ArticlePDF Available

Too Many to Fail? Evidence of Regulatory Forbearance When the Banking Sector Is Weak

Authors:

Abstract and Figures

This article studies bank failures in twenty-one emerging market countries in the 1990s. By using a competing risk hazard model for bank survival, we show that a government is less likely to take over or close a failing bank if the banking system is weak. This Too-Many-to-Fail effect is robust to controlling for macroeconomic factors, financial crises, the Too-Big-to-Fail effect, domestic financial development, and concerns due to systemic risk and information spillovers. The article also shows that the Too-Many-to-Fail effect is stronger for larger banks and when there is a large government budget deficit. The Author 2009. Published by Oxford University Press on behalf of The Society for Financial Studies. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org., Oxford University Press.
Content may be subject to copyright.
Too Many to Fail? Evidence of Regulatory
Forbearance When the Banking Sector
Is Weak
Craig O. Brown
Bert W. Wasserman Department of Economics and Finance,
Zicklin School of Business
I. Serdar Dinc¸
Department of Finance, MIT-Sloan
This article studies bank failures in twenty-one emerging market countries in the 1990s.
By using a competing risk hazard model for bank survival, we show that a government is
less likely to take over or close a failing bank if the banking system is weak. This Too-
Many-to-Fail effect is robust to controlling for macroeconomic factors, financial crises,
the Too-Big-to-Fail effect, domestic financial development, and concerns due to systemic
risk and information spillovers. The article also shows that the Too-Many-to-Fail effect is
stronger for larger banks and when there is a large government budget deficit. (JEL E58,
F30, G21, G28)
Banking is a very important part of a free-market economy. Yet exit from the
sector is not governed by market forces alone. An insolvent bank can continue
to operate by issuing new deposits to pay old liabilities until government
regulators decide to intervene. Hence, the timing and the quality of regulatory
intervention are important factors in maintaining a healthy financial system and
economy.
In principle, the government can always close a failing bank as soon as the
bank becomes insolvent. In practice, the number of options available to regu-
lators for handling the bank insolvency problem decreases with the severity of
the problem (Hoggarth, Reidhill, and Sinclair 2004; Barth, Caprio, and Levine
2006). When faced with an individual bank with a minor problem, regulatory
authorities typically seek to find a private sector solution. They grant time for
a bank turn-around and may request that the bank adopt particular measures.
When problems are severe for an individual bank, prudent regulation requires
We are grateful for comments by Paolo Fulghieri (the editor), an anonymous referee, and Maria Soledad Martinez
Peria. In addition, we would like to thank participants at the Bank of England seminar, the Office of the Comptroller
of Currency seminar, and the World Bank Banking Regulation and Corporate Governance conference. Craig
Brown wishes to thank the Research Foundation of the City University of New York for financial support. Send
correspondence to Craig O. Brown, Bert W. Wasserman Department of Economics and Finance, Zicklin School
of Business, Baruch College, City University of New York, One Bernard Baruch Way, PO Box B10-225, New
York, NY 10010; telephone: (646) 312-3519; fax: (646) 312-3451; E-mail: Craig.Brown@baruch.cuny.edu.
C
The Author 2009. Published by Oxford University Press on behalf of The Society for Financial Studies.
All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org.
doi:10.1093/rfs/hhp039
Advance Access publication May 27, 2009
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
a change in bank status through nationalization, liquidation, acquisition, or the
sale to a private entity. In times of crisis, the government may be forced to
intervene through nationalization to reduce disruption in the payments system
(Hoggarth, Reidhill, and Sinclair 2004), or to prevent fire sale prices to foreign
banks (Acharya and Yorulmazer 2008), or both.
Regulators appear to practice excessive regulatory forbearance (Hoffman
and Santomero 1998).
1
They practice regulatory forbearance when prudential
regulation dictates a change in bank status. What criteria does the government
use when deciding whether to take over or close a weak bank? Does the process
depend on the severity of problems in the banking sector? In particular, does the
government delay the closing or taking over of a bank if the banking system is
weak? These questions have become increasingly important as financial crises
become more severe in terms of depth and global scope.
The 2008 financial crisis provides a timely case study for the Too-Many-
to-Fail effect. Regulators arranged for a relatively quick resolution when Bear
Stearns experienced difficulty early in the crisis. As the crisis came to a head in
October, the sector-wide banking difficulties became evident. This realization
ultimately resulted in the U.S. government approving a massive $700 billion in
funding for the Troubled Asset Relief Program (TARP). Regulators argue that
this sort of “firepower” is necessary because of the scope of the crisis and the
number of banks in financial distress. It may be too early to draw conclusions
in this instance, but our analysis of the regulatory response to past bank failures
in emerging markets is informative, especially since crisis events in the United
States appear to be similar to those in emerging markets (Reinhart and Rogoff
2008).
Recent theoretical research argues that the Too-Many-to-Fail phenomenon
exists in bank regulation (Mitchell 2001; Acharya and Yorulmazer 2007). Reg-
ulators may choose not to take over or close a failing bank if there are many
weak banks. Alternatively, there may be reasons for aggressive regulatory in-
tervention in failing banks when the banking system is weak, precisely because
of concerns about systemic risk
2
(Allen and Gale 2000). Hence, the question
of whether there is a Too-Many-to-Fail effect cannot be settled by theoretical
arguments only; it requires empirical analysis.
Any empirical study of bank failures—or corporate failures in general—is
complicated, both conceptually and econometrically, by the fact that weak
banks may be prone to exit the sector through acquisition. Furthermore, the
likelihood of a potential bidder to materialize and obtain regulatory approval for
an acquisition is unlikely to be independent from the decision to close a failing
1
Kasa and Spiegel (2008) argue that excessive regulatory forbearance may come about as a result of a precommit-
ment to a relative closure rule in bank failure resolution. They argue that this policy, as opposed to an absolute
closure rule, permits a low number of closures when there are severe problems in the banking sector.
2
For models of the Too-Many-to-Fail effect in nonbanking contexts, see Roland and Verdier (1994) for privatization
and Perotti (1998) for monetary stabilization. Although our article is not an empirical test of any particular model,
we use the insights from theoretical models as motivation for our tests.
1379
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
bank. An empirical study of bank failures must therefore incorporate bank exit
through acquisition. This is the first article to incorporate more than one exit
alternative, not just in studying bank failures but also in studying corporate
failures and bankruptcy in general.
We follow the largest banks in twenty-one emerging market countries through
most of the 1990s. Our main finding is that a government is less likely to take
over or less likely to close a failing bank if other banks in that country are
weak. This result is robust to controlling for macroeconomic factors, financial
crises, the Too-Big-to-Fail effect, domestic financial development, and conta-
gion concerns due to systemic risk and information spillovers. This article is
the first to document the Too-Many-to-Fail channel of regulatory forbearance
in a multicountry bank setting.
The magnitude of the Too-Many-to-Fail effect is economically significant.
The rate of bank failure conditional upon past survival—also known as the
hazard rate—increases by about 15–40% as the health of other banks in that
country increases from the 25th to 75th percentile. We also find that this effect
is greater for large banks and the effect increases with the government’s budget
deficit.
This article contributes to the literature on regulatory forbearance. Several
single-country studies already suggest that the Too-Many-to-Fail approach
exists in banking regulation.
3
Our article adopts a bank-level, multicountry
approach, which allows empirical tests that are difficult to conduct in a single-
country setting. This approach allows us to separate the Too-Many-to-Fail
effect from other country-specific factors that tend to be associated with bank
failures.
Our article is also related to the literature on bank failures in emerging
markets. In contrast to our article, most of this literature consists of country-
level analyses of banking crises.
4
Two of the exceptions are Bongini, Claessens,
and Ferri (2001) and Bongini, Laeven, and Majnoni (2002), who provide a bank-
level analysis of the banking crises in four East Asian countries.
5
In another
exception, Brown and Dinc¸ (2005), with whom this article shares data, show
that regulators are more likely to take over or close failing banks shortly after
elections rather than shortly before elections.
The rest of the article is organized as follows. The next section presents
the data. The second section discusses our methodology. The third section
presents the main results. The fourth section provides robustness checks that
3
See Kane (1989), Barth (1991), White (1991), and Kroszner and Strahan (1996), who argue for the Savings and
Loan Crisis in the United States; Hoshi and Kashyap (2001) and Amyx (2004) for the Japanese Banking Crisis;
and, in a nonbanking setting, Berglof and Bolton (2002) for the implementation of corporate bankruptcy laws in
Hungary and the Czech Republic.
4
See, for example, Caprio and Klingebiel (2002); Demirguc-Kunt and Detragiache (2002); Claessens, Klingebiel,
and Laeven (2005); Barth, Caprio, and Levine (2006); and Beck, Demirguc-Kunt, and Levine (2006).
5
In a firm-level study of the East Asian financial crisis, Aguiar and Gopinath (2005) show that foreign firms
provided liquidity through acquisitions. This finding suggests that remedies to liquidity problems in a country
may not be limited to government intervention in the banking sector.
1380
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
the Too-Many-to-Fail effect we detect is not a mere reflection of the Too-Big-
to-Fail effect. I n the fifth section, we analyze the most likely drivers of the
Too-Many-to-Fail effect. The sixth section provides further robustness checks
and is followed by concluding remarks.
1. Data
The data are obtained from Brown and Dinc¸ (2005), who identify the ten
largest commercial banks in each of twenty-one emerging market countries.
These banks are followed from 1 January 1994 until one of the following three
exit events takes place: (i) failure as manifested through takeover or license
suspension/revocation by the regulators; (ii) merger with or acquisition by
another bank; or (iii) reaching 31 December 2000, the end of sample period.
Each bank merger is evaluated on a case-by-case basis to decide whether it
is, in fact, a government takeover of a failing bank. If one of the merger
partners is a private bank but the resulting entity is majority-owned by the
government, then that merger is considered as a government takeover, and
hence, the failure of that private bank. If the bank is acquired by another
bank, where there is a change of majority ownership, then it is considered a
bank acquisition exit event. If the government provides financial support f or a
bank acquisition, then it is considered a government-assisted acquisition. We
recognize that the government can intervene in a failing bank in many ways: by
providing liquidity support, limiting operations, or purchasing nonperforming
assets. We choose to focus on government takeovers and closures of failing
banks instead of other limited forms of intervention for the following reasons.
First, government takeovers and closures of failing banks are the most costly
forms of intervention. Hence, the issue of forbearance is likely to be more acute
with our chosen forms compared to other limited forms of intervention. Second,
the data on limited forms of government intervention are simply not available.
Finally, and related to the first two reasons, the data quality for limited forms of
intervention is likely to be poor. This is because in order to prevent bank runs
and other destabilizing market effects, governments actually have an incentive
not to be forthcoming about limited forms of intervention.
6
Bankscope provides the balance sheet data. Government takeovers and the
ultimate ownership of the banks are determined through manual data collection.
Press sources provided in Factiva are used to identify the failing banks and de-
termine the exact date of government intervention. The banks that are acquired
by other banks are identified using the SDC International M&A database. The
ultimate owner of each bank is determined using Bankscope, Factiva, SDC, and
various Internet sources. Based on the ultimate owner, the sample is split into
6
For example, the U.S. Federal Reserve established new channels for liquidity support to banks instead of using
the usual discount window after the “subprime” crisis so that the banks obtaining such support would remain
undisclosed. In addition, government intervention may also take place through loans from government-owned
banks, which would make that type of intervention very difficult to distinguish from normal interbank lending.
1381
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
two groups. The banks in the first group are always 50% or more owned by the
central government throughout the sample period. The second group consists
of the banks in which government ownership, if any, was less than 50% in at
least one year during the sample period. In particular, this group includes banks
that were owned by the government at more than the 50% level in 1993 and
were subsequently privatized during the sample period. We refer the reader to
Brown and Dinc¸ (2005) for the details of the dataset. We also control for sev-
eral country-level characteristics in this study. Data on deposit insurance and
stock market turnover are obtained from the World Bank Database on Financial
Structure. An index representing the quality of creditor rights is provided by
Djankov, McLiesh, and Shleifer (2007). Data on banking crises are sourced
from the dataset provided by Demirguc-Kunt and Detragiache (2005), and data
on currency crises are sourced from the dataset provided by Kaminsky (2003).
Table 1 reports the number of bank failures in 1994–2000 among the ten
largest banks (as of 1993) in each country. Three findings are worth empha-
sizing. First, bank failure is very common in the sample countries. Out of 164
private banks, 40 banks, or about 24%, were taken over by the government dur-
ing the sample period; 32 banks were acquired by other financial institutions of
which three were government assisted. These failures are not just a reflection
of the Asian Financial Crisis or another crisis. In total, twelve countries had
at least one bank failure among the largest banks during the sample period.
Second, the regulatory intervention in failing banks by suspending the banking
license of the failing bank, paying the depositors from the deposit insurance,
and liquidating the bank is an exception. In thirty-four of the forty government
action episodes, the government actually took over the bank and continued to
operate it. Third and perhaps unsurprisingly given the intervention choice of
the government, no government-owned bank in the sample ever lost its banking
license.
Given that no government-owned bank failed during the sample period, the
analysis in the rest of the article focuses on the bank-years when the banks
were private. To summarize, the following entry and exit events are adopted
for analysis: bank i enters the study in year t
i
, which is the later occurrence
of one of the following two “entry” dates: (i) 1 January 1994, the start of the
sample period; and (ii) the date the bank is privatized, so that ownership by the
central government drops below 50%. Bank i exits the study in year T
i
, which
is the earliest occurrence of one of the following three “exit” events: (i) the
bank is taken over or has its license suspended/revoked by the government; (ii)
the bank is acquired by another bank; balance sheet data are no longer available
for that bank as a separate entity; or (iii) the bank survives until 31 December
2000, the end of the sample period.
Table 2 presents sample statistics for selected balance sheet items of these
banks between their entry and exit dates and grouped by the type of bank ex-
its. Banks that are taken over or closed by the government and banks that are
acquired by other banks are smaller than the banks that survive to the end of
1382
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Table 1
Bank failures by country
Always government owned Private banks
Total number of License revoked Taken over by License revoked
Country banks (1993) Total number or liquidated Total number the government or liquidated Acquisition
Southeast Asia
Indonesia 10 5 5 5
Malaysia 10 2 8 2
Singapore 10 10
South Korea 10 2 8 5
Taiwan 10 3 7
Thailand 10 2 8 4 1
Total 60 14 0 46 14 0 3
Latin America
Argentina 10 2 8 2
Brazil 10 1 9 3 1
Chile 10 1 9 3
Colombia 10 2 8 1 2
Mexico 10 2 8 3 1
Peru 10 1 9 1 5
Venezuela 10 1 9 4 1
Total 70 10 0 60 12 0 15
Rest of the world
Czech Republic 10 10 4 2 2
Hungary 10 1 9 1 3
India 10 9 1
Israel 10 2 8 2
Poland 10 3 7 6
Russia 10 2 8 2 4
South Africa 10 1 9 1
Turkey 10 4 6 1
Total 80 22 0 58 8 6 14
Overall total 210 46 0 164 34 6 32
The table provides the number of bank failures among the ten largest banks (as of the end of 1993) in each of the twenty-one sample countries during the sample period 1994–2000. Each
bank is followed from 1 January 1994 until the first occurrence of one of the three exit events: (i) takeover or license revocation/liquidation by the government; (ii) acquisition by another
bank; or (iii) surviving to 1 January 2001. The table splits the sample based on ownership. Banks that are always government owned are the banks that were always owned by the central
government at least at the 50% level throughout 1994–2000. Private banks are the remaining banks. The banks that were owned by the government in 1993 but were later privatized are
included among the private banks unless one of the three exit events occurred first.
1383
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Tab le 2
Sample statistics
Government
Variable takeover/closure Acquisition Banks that survived All banks
Assets/GDP Mean 5.586
∗∗
3.255
∗∗∗
7.932 6.979
SE 0.580 0.298 0.444 0.337
SD 6.868 2.695 10.171 9.207
N 140 82 525 747
Total Loans/Assets Mean 0.569 0.540
∗∗
0.580 0.574
SE 0.014 0.018 0.007 0.006
SD 0.171 0.165 0.152 0.158
N 140 82 525 747
Total Deposits/Assets Mean 0.766 0.748 0.746 0.750
SE 0.013 0.014 0.007 0.006
SD 0.149 0.120 0.163 0.157
N 138 79 520 737
Capital Ratio Mean 0.044
∗∗∗
0.090 0.093 0.083
SE 0.014 0.004 0.002 0.003
SD 0.163 0.032 0.055 0.087
N 140 82 525 747
Operating Income/Assets Mean 0.019
∗∗∗
0.015 0.017 0.010
SE 0.017 0.002 0.001 0.003
SD 0.196 0.020 0.025 0.088
N 137 79 521 737
The table provides sample statistics for the banks in the sample. Government takeover/closure represents the
banks that were taken over by the government or had their licenses revoked by the government during the sample
period. Acquisition represents banks that were sold or acquired during the sample period. N denotes the number
of bank-years. Capital Ratio is the book value of shareholder equity divided by total assets. All variables are
book values.
,
∗∗
,
∗∗∗
denote statistical significance at the 10%, 5%, and 1% levels, respectively, in a two-sided
test of the mean of the type of exit with the mean of banks that survived.
our sample period. Acquired banks also have a lower loan-to-asset ratio while
there is no statistically significant difference for deposits to total assets. Unsur-
prisingly, banks taken over or closed by the government are undercapitalized
and less profitable compared to other banks. The capital ratio, defined as the
total equity divided by total assets, is only 4.4% for takeover/closure banks,
while it is 9.3% for banks that survived. Similarly, annual income per asset
is lower in failed banks with 1.9%, while the same ratio is 1.7% for banks
that survived. Both differences are statistically significant at the 1% level. The
negative average income per asset for takeovers has an interesting implication.
Unless these banks made very big losses in the year immediately before gov-
ernment intervention, these banks must have incurred losses for several years
before the government finally took them over or closed them.
2. Methodology
We adopt a hazard model to study bank failures.
7
In a traditional hazard model,
only one type of exit is considered, namely, the bank failure. The hazard rate for
7
Shumway (2001) shows the superiority of hazard models to single-period models in forecasting bankruptcy.
Studies that use hazard models in analyzing bank failures include Lane, Looney, and Wansley (1986); Whalen
(1991); Molina (2002); and Brown and Dinc¸ (2005).
1384
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
this exit, which is defined as the instantaneous rate of bank failure given survival
until that time, becomes the basis of estimation. However, there is one more
type of exit for a bank from observation: acquisition by another bank. Once a
bank is acquired by another bank, the acquired bank drops from observation,
and it can no longer be known whether or when that bank would fail. Hence,
it is desirable to incorporate into the study of bank failures the fact that some
banks may exit from observation through acquisition.
8
When multiple states of
exits are possible, the resulting hazard analysis is called competing risk hazard
analysis. When the exit events are independent from each other, the resulting
model is easy to estimate econometrically. Unfortunately, such independence
cannot be assumed in our context. Not only may a bank be more likely to
be acquired if it is weak, but the regulators’ decision to approve or reject an
acquisition may also be related to bank health. Furthermore, the regulatory
decision may also depend on the financial health of other banks, which is
the focus of this study. We describe below the competing risk hazard model
employed in our analysis.
9
Recall that, in the traditional hazard analysis with a single type of exit,
the hazard function, which represents the instantaneous rate of exit at time t
conditional on having survived until then, is given by
λ(t|X) = lim
h0
P[t < T < t + h|T > t, X ]
h
, (1)
where X is the observable control variables, which may depend on time t as
well. The survivor function is then
S(t|X) = P[t > T |X] = exp
t
0
λ(u|X)du
. (2)
Finally, the likelihood function used in estimation is based on the probability
density function for the time to exit; this density function is given by
f (t|X) = lim
h0
P[t < T < t + h|X ]
h
= λ(t|X)S(t |X). (3)
To model competing risks, we consider type-specific hazard function,also
called transition intensity, given by
λ
j
(t|X) = lim
h0
P[t < T < t + h, d = j|T > t, X ]
h
, (4)
where λ
j
(t|X) is the instantaneous rate of exit at time t due to type j having
survived until t. We make the standard assumption that at most only one type
8
We thank a referee for suggesting this addition.
9
See Kalbfleisch and Prentice (2002, Ch. 8) and Lancaster (1990, Section 5.5) for a textbook treatment of
competing risk hazard models; our exposition largely follows the former. See also the references below for
economic applications.
1385
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
of exit can occur in any given instant, so we have
λ(t|X) =
j
λ
j
(t|X). (5)
The density function for the time to a type j exit is given by
f
j
(t|X) = lim
h0
P[t < T < t + h, d = j|X]
h
= λ
j
(t|X)S(t|X ). (6)
Let us denote our sample by (t
i
, δ
i
, d
i
, X
i
), where i = 1, ..., n indexes of the
number of banks, δ
i
is the indicator that becomes 1 if the bank exits the sample
and 0 if it reaches the end of our sample period without exiting (i.e., it is “right
censored”), and d
i
denotes the type of the bank’s exit, which is unobserved and
does not enter into the likelihood function given below if δ
i
= 0. The likelihood
function then becomes
L =
i
((λ
d
i
(t
i
|X
i
))
δ
i
S(t
i
|X
i
)). (7)
Equation (7) has been obtained without any functional assumptions for the
hazard function but, without such assumptions, (7) will not be very useful
for estimation. We adopt the common exponential form for our type-specific
hazard function as given below:
λ
j
(t|X) = b
j
(t)exp [β
j
X + μ
j
], (8)
where b
j
(t) is the baseline hazard function, β
j
are the coefficients to be es-
timated, and μ
j
is the unobserved heterogeneity term, which is discussed in
more detail below. There are at least three aspects of (8) that should be em-
phasized. First, the coefficients β
j
to be estimated are indexed by the exit type
j, which implies that different sets of coefficients are (jointly) estimated for
different types of exit in each regression. Second, the baseline hazard b
j
(t)is
also allowed to be different for different types of exit.
Finally, the hazard function (8) includes the unobserved heterogeneity term
μ
j
, which is akin to “random effects,” as in Han and Hausman (1990) and
Sueyoshi (1992), among others. This term serves two purposes. First, Heckman
and Singer (1984) show that including unobserved heterogeneity and estimating
it nonparametrically increases the accuracy of coefficient estimates for the
structural equations even if the distribution of the unobserved heterogeneity
is not accurately estimated. Second, this term permits us to allow dependence
between different types of exit. In other words, by not requiring μ
j
and μ
l
to be
independent for j = l, we allow the banks that are more likely to be acquired
by other banks for reasons unobserved by the econometrician to be more (or
less) likely to be taken over or closed by the government.
10
10
For the importance of allowing dependence, see Honore and Tamer (2006).
1386
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Heckman and Singer (1984) argue for modeling the unobserved hetero-
geneity as a discrete distribution and for estimating the jump points and
their associated probabilities together with structural coefficients. Following
McCall (1996); Dolton and van der Klaauw (1999); Deng, Quigley, and Van
Order (2000); and Fallick and Ryu (2007), among others, we also adopt this
framework. More precisely, we assume that
μ
j
∈{μ
j1
,...,μ
jk
} with P[μ
j
= μ
jk
] = π
jk
and
jk
π
jk
= 1. (9)
Hence, our type-specific hazard function in (8) becomes
λ
j
(t|X) =
k
(π
jk
b
j
(t)exp[β
j
X + μ
jk
]), (10)
while the survivor function can be obtained as
S(t|X) = exp
t
0
jk
(π
jk
b
j
(u)exp[β
j
X + μ
jk
])du
. (11)
The likelihood function can then be obtained by plugging (10) and (11)
into (7).
Note that (9) has too many free parameters so it requires some normal-
ization. In estimation, we normalize μ
j1
for all j, adopt discrete unobserved
heterogeneity with K = 2, and fit a fourth-order polynomial for the natural
log functions of the baseline hazard b
j
(t), separate for each j. I n our model,
we consider only two different types of exit for banks, namely, (i) a takeover
or closure by the government; or (ii) acquisition by another bank, so j = 1,2.
Finally, our explanatory variables X depend on time and include measures of
financial health for other banks, as discussed in the next section.
3. Regression Results
The focus of our analysis is the government takeover or closure of failing
banks. However, as described in the Methodology section, a bank may also
exit from the sample when it is acquired by another bank. An acquisition
exit may (or may not) be related to bank health and need not be independent
from a government takeover or closure decision. Our hazard analysis explicitly
incorporates the exit through being acquired as a competing risk to the main
focus of the analysis. For each regression, two sets of coefficients are (jointly)
estimated: one for each exit type, as the coefficients for each type of exit are
allowed to be different. Throughout our analysis, we report marginal effects
(in percentage points) on the hazard evaluated at the sample mean. A positive
effect indicates an increase in the probability of exit (by that type) given that the
bank has survived to the current point in time. As the government action need
1387
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Tab le 3
Too-Many-to-Fail: regulatory reluctance when the banking system is weak
Government Government Government
takeover/ takeover/ takeover/
closure Acquisition closure Acquisition closure Acquisition
Total Assets/GDP 1.856 1.349
∗∗∗
1.826 1.362
∗∗∗
2.115
∗∗∗
1.038
(1.450) (0.387) (1.406) (0.383) (0.473) (0.561)
Capital Ratio 0.499
∗∗∗
0.017 0.722
∗∗∗
0.017 0.401
∗∗∗
0.014
(0.178) (0.108) (0.202) (0.118) (0.082) (0.161)
Income 0.050 0.106 0.147 0.147 0.139 0.137
(0.195) (0.119) (0.184) (0.184) (0.097) (0.095)
Before Election 4.388
∗∗
0.107 4.993
∗∗
0.109 3.333
∗∗
0.800
(1.966) (1.415) (2.300) (1.454) (1.406) (0.795)
Capital Ratio_Other Banks 1.235
∗∗∗
0.006 1.873
∗∗∗
0.427
(0.414) (0.703) (0.389) (0.440)
GDP per Capita 1.191
∗∗∗
0.030
(0.401) (0.777)
GDP Growth 0.387
∗∗∗
0.050
(0.054) (0.179)
Currency Depreciation 8.473
∗∗∗
1.364
(2.768) (2.832)
Inflation Rate 1.102 2.528
(2.194) (1.862)
Real Interest Rate 0.010 1.601
(3.950) (4.074)
IMF Loans/GDP 2.128
∗∗∗
6.119
∗∗∗
(0.502) (0.843)
Banking Crisis 0.485 1.842
(1.029) (1.218)
P-value 0.000 0.003 0.000 0.004 0.000 0.000
Observations 763 763 523
P-value (all) 0.000 0.000 0.000
The table presents the results of a competing risk proportional hazard model for bank failure, where there are
two types of bank exit: takeovers or closures of banks by the government and the acquisition of the bank by
another bank. The model allows for correlated bank exit types. Each column represents a single regression, and
the coefficients for both types of exit in a column are jointly estimated. For each variable, we report the marginal
effect evaluated at the sample mean. A positive effect (in percentage points) denotes an increasing hazard of
bank exit through that type of exit event. Total Assets/GDP is the bank’s total assets normalized by the country’s
GDP. Capital Ratio is the total e quity divided by total assets. Income is operating income divided by total assets.
Capital Ratio_Other Banks is the weighted average (by total assets) of the capital ratio of other banks in that
country. All are book values and as of year t–1. Before Election is a dummy variable that takes 1 if the bank
fails in the latter half of the electoral cycle or, in the case of no exit, the end of the bank’s accounting year falls
within the latter half of the electoral cycle. GDP per Capita is GDP for a given year divided by the population
in that year. GDP Growth is the rate of growth in the country’s GDP. Currency Depreciation is the decrease in
the local currency’s exchange rate against U.S. dollars; it is negative if the local currency appreciates. Inflation
Rate is the logarithm of 1 plus the consumer price inflation. Real Interest Rate is the nominal lending rate minus
the rate of consumer price inflation. IMF Loans/GDP is total IMF loans outstanding to the country, normalized
by the country’s GDP. All variables are as of t–1. P-values of a Wald test that all coefficients are jointly zero
are reported for each type of bank exit and then for both types of bank exit. Heteroscedasticity-robust standards
errors, corrected for clustering at the country level, are in parentheses.
,
∗∗
,
∗∗∗
denote statistical significance at
the 10%, 5%, and 1% levels, respectively.
not be independent across the banks within a country, the errors are clustered
at the country level—and robust to heteroscedasticity.
The main regression results are reported in Table 3. Each column reports
the results of a single regression, which jointly estimates the coefficients of
variables for two types of bank exit: government takeover and acquisition by
another entity. To control for size, all the regressions include Total Assets/GDP,
which is the bank’s total assets normalized by the GDP of the country where
1388
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
it is located; two variables in the regression control for the bank’s financial
health: Capital Ratio, defined as the book value of the shareholder equity
divided by total assets, and Income, defined as the operating income divided
by total assets. Brown and Dinc¸ (2005) find that governments are less likely
to take over or close a failing bank before elections, so we include a Before
Election dummy variable. This variable takes the value of 1 if the bank exits in
the latter half of the electoral cycle, or in the case of no exit, the end of bank’s
accounting year if it falls within the latter half of the electoral cycle. All these
variables, with the exception of the Before Election dummy variable, are as of
year t1.
The first regression in Table 3 serves as a benchmark and does not include
measures of financial health for other banks. For government takeover/bank
closure, there is no evidence that size is a factor for this sample when the
macroeconomic factors are not controlled for. On the other hand, the bank
financial health plays a large role, as expected. The marginal effect of Capital
Ratio on the hazard is negative and statistically significant at the 1% level,
which indicates that banks with larger capital bases are less likely to fail given
that they have survived to the current point in time. The regression also confirms
that political concerns play a major role in the government decision to take over
a failing bank, as the marginal effect for the Before Election dummy is negative
and statistically significant.
The primary finding for acquisitions is that larger banks are less likely to be
acquired. This effect is statistically significant at the 5% level. However, we
do not find any evidence that weak banks are any more likely to be acquired.
Similarly, we do not find any role for the electoral cycle or the country’s
income level for bank acquisitions. While the coefficients of these factors may
not be individually significant in explaining bank acquisitions, they are jointly
significant at the 1% level.
The second regression in Table 3 is one of our main regressions. It includes
a measure of financial health for other banks, Capital Ratio_Other Banks,
which is the average of the capital ratio measure of other banks in that country,
weighted by bank total assets. While the regression sample contains only private
banks, these measures are constructed using all the banks in the initial sample
(government and private) to capture the financial health of the banking sector
in that country. As discussed before, this variable should not have a statistically
significant effect if the government decision to take over or close a failing bank
is based only on that bank’s health. On the other hand, if there is regulatory
delay in taking over or closing a bank when the other banks in the system
are weak, Capital Ratio_Other Banks will have a positive and statistically
significant effect.
Capital Ratio_Other Banks has a positive and statistically significant effect
for the government takeover or closure. This indicates that, controlling for
individual bank-level factors, the government is more likely to take over or
liquidate a failing bank if the remaining banks have high capital ratios—a
1389
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Too-Many-to-Fail effect. We do not find a similar effect for bank exit through
acquisitions by other banks.
It is important to study the robustness of the aforementioned results to
common macroeconomic factors. The main concern is whether the health of
other banks just proxies for the general health of the economy. Though if that
were the case, there would be a negative coefficient indicating a slower rate of
failure, not a positive coefficient as we found. Nevertheless, in regression 3,
we control for five different macroeconomic variables: GDP per capita, GDP
growth rate, currency depreciation, inflation rate, and real interest rate.
11
We
also include the total IMF loans to that country, normalized by that country’s
GDP, to control for any influence by IMF. It is also important to check whether
the Too-Many-to-Fail effect we detect is a mere reflection of financial crises. To
control for this effect, we include country-year Banking Crisis dummy variable
constructed with the data from Demirguc-Kunt and Detragiache (2005).
12
All
macroeconomic variables are as of year t1.
Regression 3 in Table 3 reports the results of the regressions that include these
macroeconomic variables. Our main variable of interest in the analysis, Capi-
tal Ratio_Other Banks, continues to have positive and statistically significant
effects in all the regressions. These results indicate that the Too-Many-to-Fail
effect does not occur because the financial health measures employed in the
analysis proxy for some common macroeconomic factor.
Other bank-level risk indicators may also have predictive power in a govern-
ment takeover or closure of failing banks, so it i s important to check whether
the financial health measures for other banks in the country are robust to con-
trolling for those bank-level factors. Unfortunately, there is a lack of data about
one factor likely to determine bank failures—nonperforming loans. Data on
nonperforming loans are available for fewer than half of the bank-years in the
sample. Without those data, we turn our attention to other factors that may play
a role in determining bank failure.
Equity reserves may provide a cushion for adverse times, so banks with
greater reserves are less likely to fail. Loans are illiquid, while the deposits
are liquid, so a bank with a high proportion of loans may be more likely to
fail. Similarly, the risks taken by a bank may be reflected in the difference
between the interest paid by the bank to depositors and the interest charged to
its borrowers. Each regression reported in Table 4 controls for these factors but
none of them seem to have a statistically significant impact once we control for
the bank’s capital ratio and income. On the other hand, Capital Ratio_Other
Banks continues to have a positive and statistically significant effect. These
11
It should also be noted that, to the extent that countrywide macroeconomic factors are correlated with the financial
health of other banks, potential multicollinearity problems will make it difficult to obtain a statistically significant
coefficient for our measure of other banks’ health.
12
In an earlier version, we also checked the robustness of the Too-Many-to-Fail effect to the existence of a currency
crisis in that country, as opposed to a banking crisis, using the data from Kaminsky (2003) and obtained similar
results.
1390
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Tab le 4
Too–Many-to-Fail: additional bank-level factors
Government Government Government
takeover/ takeover/ takeover/
closure Acquisition closure Acquisition closure Acquisition
Total Assets/GDP 2.717
∗∗∗
1.054
2.283
∗∗∗
0.953 2.200
∗∗∗
1.107
(0.492) (0.575) (0.766) (3.759) (0.497) (0.598)
Capital Ratio 0.797
∗∗∗
0.034 0.443 0.018 0.550
∗∗∗
0.025
(0.155) (0.157) (0.753) (0.326) (0.155) (0.164)
Income 0.227 0.146 0.094 0.136 0.024 0.105
(0.274) (0.146) (1.314) (0.820) (0.155) (0.112)
Before Election 3.267
∗∗
0.788 3.335
0.859 2.174
∗∗
0.849
(1.477) (0.854) (1.962) (1.799) (1.008) (0.830)
Capital Ratio_Other Banks 2.332
∗∗∗
0.445 1.930
∗∗∗
0.588 1.831
∗∗∗
0.368
(0.460) (0.514) (0.531) (3.466) (0.239) (0.474)
Equity Reserves 0.327 0.021
(0.246) (0.135)
Loans 1.509 0.350
(1.941) (2.872)
Lending Margin 0.959 1.942
(0.607) (1.462)
Macro & Crisis Control Yes Yes Yes Yes Yes Yes
P-value 0.000 0.000 0.000 0.000 0.000 0.000
Observations 520 523 513
P-value (all) 0.000 0.000 0.000
The table presents the results of a competing risk proportional hazard model for bank failure, where there are
two types of bank exit: takeovers or closures of banks by the government and the acquisition of the bank by
another bank. The model allows for correlated bank exit types. Each column represents a single regression, and
the coefficients for both types of exit in a column are jointly estimated. For each variable, we report the marginal
effect evaluated at the sample mean. A positive effect (in percentage points) denotes an increasing hazard of
bank exit through that type of exit event. Total Assets/GDP is the bank’s total assets normalized by the country’s
GDP. Capital Ratio is the total e quity divided by total assets. Income is operating income divided by total assets.
Capital Ratio_Other Banks is the weighted average (by total assets) of the capital ratio of other banks in that
country. All are book values and as of year t–1. Before Election is a dummy variable that takes 1 if the bank
fails in the latter half of the electoral cycle or, in the case of no exit, the end of the bank’s accounting year
falls within the latter half of the electoral cycle. Macro and Crisis Control variables include GDP per Capita,
GDP Growth, Banking Crisis, Currency Depreciation, Inflation Rate, Real Interest Rate,andIMF Loans/GDP.
Equity Reserves represents the equity reserves of the bank, normalized by total assets. Loans represents the total
net loans divided by total assets. Lending Margin is the spread between the average interest rate charged on
loans and the average interest rate paid on deposits. All variables are as of t1. P-values of a Wald test that
all coefficients are jointly zero are reported for each type of bank exit and then for both types of bank exit.
Heteroscedasticity-robust standards errors, corrected for clustering at the country level, are in parentheses.
,
∗∗
,
∗∗∗
denote statistical significance at the 10%, 5%, and 1% levels, respectively.
results imply that the Too-Many-to-Fail effect shown above is not a proxy for
some common bank-level risk factor.
4. Too-Many-to-Fail and Too-Big-to-Fail
It is important to verify that the Too-Many-to-Fail effect we demonstrate is not
just a r eflection of the Too-Big-to-Fail effect discussed in the literature. Our
sample includes only the largest ten banks in a country, and our regressions
always include a size variable, so our analysis already suggests some robustness
in this direction. However, given the importance of Too-Big-to-Fail in banking,
it is still desirable to study this issue in detail. Instead of our usual size variable,
we create four different dummy variables for the top three banks based on assets,
1391
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
loans, deposits, or employee expenses in that country in that year.
13
While these
tests do not, of course, constitute a proof or repudiation of any potential Too-
Big-to-Fail effect in these countries, they will capture nonparametrically any
Too-Big-to-Fail effect within our sample of already large banks. These four
variables are naturally correlated with one another, but they are also different
in that they are likely to capture different effects. For example, the dummy
variables constructed using loans will capture the government concern for the
borrowers upon the failure of the largest lenders, while the dummy variable
based on the deposits will reflect the government’s concern for the burden of a
large bank failure on the deposit insurance fund. Similarly, the dummy variable
constructed using employee expenses will incorporate the government aversion
to large layoffs upon the failure of large banks.
Table 5 reports these regressions. In all the regressions, regardless of which
of the four different bank-specific characteristics we use, the dummy variable
for the largest t hree banks has a negative and statistically significant coefficient.
Our main variable of interest, Capital Ratio_Other Banks, continues to have a
positive and statistically significant coefficient. These results cannot conclude
that the Too-Big-to-Fail effect did or did not exist in these countries, but they do
imply that the Too-Many-to-Fail effect demonstrated in this article is separate
and not just a reflection of any possible Too-Big-to-Fail effect.
5. Understanding the Too-Many-to-Fail Effect
In this section, we study the potential economic drivers of the Too-Many-
to-Fail effect. In doing so, we employ a number of interaction effects. For a
nonlinear model such as the one used in this article, the study of interaction
effects is not as straightforward as it would be in a linear regression. In a linear
regression, the interaction effect is completely captured by the coefficient of the
interaction term; hence, the interaction effect remains constant for all the values
of explanatory variables. But, as Ai and Norton (2003) show, the interaction
effect cannot be completely captured by the coefficient of the interaction term
in a nonlinear regression; instead, it also depends on other coefficients and the
values of explanatory variables at which it is evaluated. In the Appendix, we
derive the interaction effect for the hazard model used in this article.
5.1 The role of government fiscal health
A potential explanation may be that the government itself may have incentives
to delay the ultimate reckoning in bank failures, as found in the Savings and
Loan Crisis by Kane (1989) and Kroszner and Strahan (1996), among others.
In particular, the takeover or closure of a bank causes the government to
incur costs of a financial cleanup in the short run. We hypothesize that the
13
We also constructed our dummy variables for the top five banks instead of three and obtained similar results.
1392
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Tab le 5
Too-Many-to-Fail versus Too-Big-to-Fail
Government Government Government Government
takeover/ takeover/ Takeover/ takeover/
closure Acquisition closure Acquisition Closure Acquisition closure Acquisition
Top 3 Assets 2.122
∗∗
0.357
(0.878) (0.633)
Top 3 Loans 2.269
∗∗∗
0.455
(0.839) (0.640)
Top 3 Deposits 2.137
∗∗
0.386
(0.885) (0.608)
Top 3 Employee 1.666
0.145
Expenses (1.004) (0.750)
Capital Ratio 0.394
∗∗∗
0.046 0.389
∗∗∗
0.048 0.399
∗∗∗
0.044 0.471
∗∗∗
0.034
(0.069) (0.159) (0.066) (0.160) (0.068) (0.160) (0.137) (0.180)
Income 0.099 0.221
∗∗
0.094 0.219
∗∗
0.095 0.218
∗∗
0.085 0 .298
∗∗
(0.104) (0.095) (0.102) (0.094) (0.102) (0.096) (0.183) (0.120)
Before Election 1.963
0.896 1.990
0.902 1.964
0.884 2.129
1.330
(1.180) (0.825) (1.146) (0.827) (1.141) (0.826) (1.293) (0.966)
Capital Ratio_ 1 .697
∗∗∗
0.662 1.668
∗∗∗
0.646 1.695
∗∗∗
0.660 1.666
∗∗∗
0.836
Other Banks (0.220) (0.592) (0.218) (0.582) (0.215) (0.579) (0.245) (1.097)
Macro & Crisis Yes Yes Yes Yes Yes Yes Yes Yes
Controls
P-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Observations 521 521 520 461
P-value (all) 0.000 0.000 0.000 0.000
The table presents the results of a competing risk proportional hazard model for bank failure, where there are
two types of bank exit: takeovers or closures of banks by the government and the acquisition of the bank by
another bank. The model allows for correlated bank exit types. Each column represents a single regression, and
the coefficients for both types of exit in a column are jointly estimated. For each variable, we report the marginal
effect evaluated at the sample mean. A positive effect (in percentage points) denotes an increasing hazard of bank
exit through that type of exit event. Top 3 Assets is a dummy variable that takes 1 if the bank is ranked in the
top three in the country based on total assets. Top 3 Loans is a dummy variable that takes 1 if the bank is ranked
in the top three in the country based on total loans. Top 3 Deposits is a dummy variable that takes 1 if the bank
is ranked in the top three in the country based on total deposits. Top 3 Employee Expenses is a dummy variable
that takes 1 if the bank is ranked in the top three in the country based on employee expenses. Capital Ratio is
the total equity divided by total assets. Income is operating income divided by total assets. Capital Ratio_Other
Banks is the weighted average (by total assets) of the capital ratio of other banks in that country. All are book
values and as of year t–1. Before Election is a dummy variable that takes 1 if the bank fails in the latter half of
the electoral cycle or, in the case of no exit, the end of the bank’s accounting year falls within the latter half of
the electoral cycle. Macro and Crisis Control variables include GDP per Capita, GDP Growth, Banking Crisis,
Currency Depreciation, Inflation Rate, Real Interest Rate,andIMF Loans/GDP. All variables are as of t1.
P-values of a Wald test that all coefficients are jointly zero are reported for each type of bank exit and then for
both types of bank exit. Heteroscedasticity-robust standards errors, corrected for clustering at the country level,
are in parentheses.
,
∗∗
,
∗∗∗
denote statistical significance at the 10%, 5%, and 1% levels, respectively.
Too-Many-to-Fail effect may be weaker for governments that run a budget
surplus or a small deficit.
To test this hypothesis, we study the interaction effect between Capital
Ratio_Other Banks and a High Budget Balance dummy variable. The High
Budget Balance dummy variable takes the value 1 if that country’s budget
balance in that year, as a ratio to its GDP, is greater than the sample median.
In our sample, the median is a budget deficit equal to 1.49% of GDP, so the
High Budget Balance dummy is 1 for countries with a budget surplus or a small
deficit.
Table 6 provides the regression results with the High Budget Balance dummy
and the interaction of that variable with Capital Ratio_Other Banks.The
1393
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Tab le 6
Understanding the reasons behind Too-Many–to-Fail
Government Government Government
takeover/ takeover/ takeover/
closure Acquisition closure Acquisition closure Acquisition
Total Assets/GDP 2.815
∗∗∗
1.330
∗∗
5.215
∗∗∗
1.581 1.885
∗∗∗
0.876
(0.550) (0.527) (1.554) (1.477) (0.455) (0.602)
Capital Ratio 0.453
∗∗∗
0.010 0.950
∗∗∗
0.091 0.369
∗∗∗
0.004
(0.086) (0.151) (0.354) (0.332) (0.084) (0.157)
Income 0.221
∗∗
0.187
0.185 0.383
0.188 0.140
(0.103) (0.104) (0.431) (0.205) (0.118) (0.084)
Before Election 3.771
∗∗
1.337 6.654
∗∗
1.376
∗∗
3.336
∗∗
0.955
(1.620) (0.912) (2.938) (0.545) (1.319) (0.746)
High Budget Balance 4.421 1.380
(4.324) (2.613)
Capital Ratio_Other Banks 2.269
∗∗∗
0.837 4.392
∗∗∗
1.775
∗∗
1.835
∗∗∗
0.417
(0.452) (1.006) (0.739) (0.812) (0.318) (0.435)
Capital Ratio_Other Banks
9.073
∗∗∗
4.006
High Budget Balance (3.133) (2.113)
Interbank Deposits/GDP 0.746 2.110
(1.671) (1.770)
Rated Bank 0.914 0.861
(0.886) (1.006)
Macro & Crisis Controls Yes Yes Yes Yes Yes Yes
P-value 0.000 0.000 0.000 0.000 0.000 0.000
Observations 490 441 523
P-value (all) 0.000 0.000 0.000
The table presents the results of a competing risk proportional hazard model for bank failure, where there are
two types of bank failure: government takeovers and sales. The model allows for correlated bank exit types. Each
column represents a single regression, and the coefficients for both types of exit in a column are jointly estimated.
For each variable, we report the marginal effect evaluated at the sample mean. A positive effect (in percentage
points) denotes an increasing hazard of bank exit through that type of exit event. Total Assets/GDP is the bank’s
total assets normalized by the country’s GDP. Capital Ratio is the total equity divided by total assets. Income is
operating income divided by total assets. Before Election is a dummy variable that takes 1 if the bank fails in the
latter half of the electoral cycle or, in the case of no failure, the end of the bank’s accounting year falls within
the latter half of the electoral cycle. High Budget Balance is a dummy variable that takes 1 if the budget balance
(the government’s fiscal budget balance normalized by the country’s GDP) is greater than the median budget
balance for the sample. Capital Ratio_Other Banks is the weighted average (by total assets) of the capital ratio of
other banks in that country. Macro and C risis Control variables include GDP per Capita, GDP Growth, Banking
Crisis, Currency Depreciation, Inflation Rate, Real Interest Rate,andIMF Loans/GDP. Interbank Deposits/GDP
are the deposits of other banks in the bank, normalized by the bank’s total assets and the country’s GDP,
respectively. Rated Bank is a dummy variable that takes 1 if the bank has any debt rated by Moody’s Investor
Service. All are book values and as of year t–1. P-values of a Wald test that all coefficients are jointly zero are
reported for each type of bank exit and then for both types of bank exit. Heteroscedasticity-robust standards
errors, corrected for clustering at the country level, are in parentheses.
,
∗∗
,
∗∗∗
denote statistical significance at
the 10%, 5%, and 1% levels, respectively.
interaction effect is evaluated at the sample mean of all other explanatory
variables. The first regression includes High Budget Balance but no interaction
term to serve as a benchmark. High Budget Balance does not have a statistically
significant effect when no interaction term is included. The second regression
includes the interaction term. It shows that the interaction effect is negative and
statistically significant at the sample mean.
5.2 The exposure of other banks to the failing bank
Prudential regulation suggests that there are legitimate reasons why government
regulators may choose forbearance and delay intervention when the banking
1394
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
system is weak. One reason is that the failure of an initial bank may trigger
failures of other banks if the other banks have loaned large sums to the initial
bank through the interbank market (Allen and Gale 2000). In turn, regulators
might delay the takeover of the failing bank to avoid triggering subsequent
industry upheavals and bank failures. To test whether these concerns are behind
the Too-Many-to-Fail delay, we control for the total interbank borrowing by
a given bank normalized by the country’s GDP. The results of the regression
analysis are reported in regression 2, Table 6.
The use of interbank deposits as a control variable is not without its own
disadvantages. Such borrowing tends to have short maturities, often overnight,
while our data come from balance sheets, so they have an annual frequency.
Low observation frequency relative to the maturities of deposits may not allow
us to detect other banks’ reaction to one bank’s deteriorating financial health.
Another disadvantage is that we have no data on the identity of the lending
banks. Some of the lending banks may in fact be government-owned banks
directed to support the failing bank through interbank deposits. Such disguised
government support may reflect the politicians’ or regulators’ hope that the
failing bank may later regain its financial health on its own or their desire to
wait until a more opportune time to intervene. Nevertheless, we still believe
that it is informative to study the role of interbank exposure in regulatory
forbearance.
We find no evidence that the exposure of other banks to a failing bank in
the system causes regulators to show forbearance and delay major intervention.
Interbank Deposits/GDP has a negative but statistically insignificant coeffi-
cient, while the effect of our measure of other banks’ financial health remains
statistically significant.
5.3 Information spillovers
Another regulatory concern might be that a failing bank potentially reveals
information about the whole banking system and that this information might
cause runs on other banks (Lang and Stulz 1992; Slovin, Sushka, and Polonchek
1999). Such fears of contagion may delay regulatory intervention. Although
there is only rare evidence of such contagion (Calomiris and Mason 1997,
2003), it is still important to study whether such regulatory concerns are behind
the Too-Many-to-Fail effect.
As a measure of publicly available information about the bank, we use a
variable for the presence of a debt rating. If the bank is rated, a regulatory
intervention is less likely to carry new information about the financial health
of banking in that country. It is also less likely to create concerns about runs
on other banks due to information spillovers. In fact, Berger, Davies, and
Flannery (2000) show that supervisory reports, when stale, tend to generate
little information about a bank over what the market already knows and that
any such information is short lived.
1395
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Regression 3 in Table 6 reports the results of regressions that include the
indicator variable Rated, which takes the value of 1 if the bank was issued
a debt rating by Moody’s in the previous year. The effect of this variable is
never statistically significant, but Capital Ratio_Other Banks continues to have
a positive and statistically significant effect. To be specific, concerns about
information spillovers do not appear to lead to the Too-Many-to-Fail effect
demonstrated in this article.
14
5.4 Other interaction effects
In regressions reported in Table 7, we explore whether the magnitude of the
Too-Many-to-Fail effect changes with some bank-level characteristics. We start
by studying whether the Too-Many-to-Fail effect is stronger for larger banks.
In the first regression, we use the base specification with a dummy variable for
large banks instead of the continuous size variable. The Large Banks dummy
variable takes the value 1 if Total Assets/GDP is greater than the sample median.
We also include an interaction between the dummy variable for Large Banks and
Capital Ratio_Other Banks. The interaction effect is positive and statistically
significant at the 1% level, which implies that the Too-Many-to-Fail effect is
indeed stronger for larger banks.
We also explore whether the Too-Many-to-Fail effect is stronger for weaker
banks. The second and third regressions show negative interaction effects when
we include variables for bank strength: a dummy variable for High Capital
Ratio and a dummy variable for High Income.TheHigh Capital Ratio dummy
variable t akes the value 1 if that bank’s capital ratio is greater than the sample
median. The High Income dummy variable takes the value 1 if t hat bank’s
operating income (as a percentage of total assets) is greater than the sample
median. Both interaction effects are negative and statistically significant at the
5% level or better, which indicates that the Too-Many-to-Fail effect is stronger
for weaker banks.
6. Robustness
6.1 Alternative measures of financial health for other banks
We start by studying the robustness of the Too-Many-to-Fail effect reported
above to different measures of financial health for other banks. The first re-
gression in Table 8 uses Liquid Reserves_Other Banks instead of Capital
Ratio_Other Banks. Carletti, Hartmann, and Spagnolo (2004) motivate Liquid
Reserves_Other Banks as a measure of banking system liquidity. It is con-
structed using the average of the liquid equity reserves of other banks in that
14
It is important to be clear about what these results do and do not imply. In particular, they do not imply that
regulators are not concerned about systemic risks in banking. In fact, their preferred method of intervention in
a failing bank, namely the government takeover as opposed to the closing of the bank, may be motivated by
concerns about systemic risks. Our results only imply that such concerns are not behind the Too-Many-to-Fail
delay demonstrated in this article.
1396
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Tab le 7
Too-Many-to-Fail: bank-level interaction effects
Government Government Government
takeover/ takeover/ takeover/
closure Acquisition closure Acquisition closure Acquisition
Total Assets/GDP 1.964
∗∗∗
1.087
2.003
∗∗∗
1.175
(0.520) (0.606) (0.464) (1.451)
Large Banks 2.246
∗∗∗
1.122
(0.846) (1.710)
Capital Ratio 0.356
∗∗∗
0.033 0.520
∗∗∗
0.001
(0.103) (0.154) (0.124) (0.163)
High Capital Ratio 0.429 0.914
(0.679) (1.812)
Income 0.126 0.206
∗∗
0.402
∗∗∗
0.179
∗∗
(0.143) (0.105) (0.096) (0.087)
High Income 3.032
0.327
(1.643) (3.759)
Before Election 3.340
∗∗
0.823 3.348
∗∗
1.276 3.287
∗∗
0.745
(1.623) (0.810) (1.372) (0.911) (1.337) (2.323)
Capital Ratio_Other Banks 1.854
∗∗∗
0.502 1.543
∗∗∗
0.518 1.773
0.309
(0.368) (0.446) (0.395) (0.331) (1.051) (0.655)
Capital Ratio_Other Banks
1.749
∗∗∗
1.701
Large Banks (0.370) (2.323)
Capital Ratio_Other Banks
2.655
∗∗
0.582
High Capital Ratio (1.247) (1.512)
Capital Ratio_Other Banks
10.363
∗∗∗
1.284
High Income (3.983) (4.321)
Macro & Crisis Controls Yes Yes Yes Yes Yes Yes
P-value 0.000 0.000 0.000 0.000 0.000 0.000
Observations 523 523 523
P-value (all) 0.000 0.000 0.000
The table presents the results of a competing risk proportional hazard model for bank failure, where there are
two types of bank exit: takeovers or closures of banks by the government and the acquisition of the bank by
another bank. The model allows for correlated bank exit types. Each column represents a single regression, and
the coefficients for both types of exit in a column are jointly estimated. For each variable, we report the marginal
effect evaluated at the sample mean. A positive effect (in percentage points) denotes an increasing hazard of
bank exit through that type of exit event. Total Assets/GDP is the bank’s total assets normalized by the country’s
GDP. Large Banks is a dummy variable that takes 1 if the size (total assets normalized by GDP) for the bank is
greater than the median size for the sample. Income is operating income divided by total assets. High Income is a
dummy variable that takes 1 if the operating income for the bank is greater than the median operating income for
the sample. Capital Ratio is the total equity divided by total assets. High Capital Ratio is a dummy variable that
takes 1 if the capital ratio for the bank is greater than the median capital ratio for the sample. Capital Ratio_Other
Banks is the weighted average (by total assets) of the capital ratio of other banks in that country. Macro and Crisis
Control variables include GDP per Capita, GDP Growth, Banking Crisis, Currency Depreciation, Inflation Rate,
Real Interest Rate,andIMF Loans/GDP. All variables are as of t1. Before Election is a dummy variable that
takes 1 if the bank fails in the latter half of the electoral cycle or, in the case of no exit, the end of the bank’s
accounting year falls within the latter half of the electoral cycle. P-values of a Wald test that all coefficients are
jointly zero are reported for each type of bank exit and then for both types of bank exit. Heteroscedasticity-robust
standards errors, corrected for clustering at the country level, are in parentheses.
,
∗∗
,
∗∗∗
denote statistical
significance at the 10%, 5%, and 1% levels, respectively.
country, weighted by bank total assets. This variable also has a positive ef-
fect that is statistically significant at the 1% level. The second regression uses
Income_Other Banks, which is the average of the income of other banks in that
country, weighted by bank total assets. This variable also has a positive effect
that is statistically significant at the 1% level. Finally, Capital Ratio_Other
Banks, our main measure of other banks’ financial health, may be endoge-
nous to the system if it includes the capital ratio of the banks that fail later
1397
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Tab le 8
Too-Many-to-Fail: alternative measures of banking system weakness
Government Government Government
takeover/ takeover/ takeover/
closure Acquisition closure Acquisition closure Acquisition
Total Assets/GDP 2.213
∗∗∗
1.088
∗∗
2.320
∗∗∗
1.099
∗∗
1.989
∗∗∗
1.032
(0.472) (0.551) (0.484) (0.556) (0.490) (0.558)
Capital Ratio 0.391
∗∗∗
0.005 0.378
∗∗∗
0.009 0.421
∗∗∗
0.034
(0.087) (0.149) (0.094) (0.154) (0.077) (0.172)
Income 0.170
0.131 0.190
0.152 0.093 0.123
(0.100) (0.102) (0.114) (0.094) (0.083) (0.102)
Before Election 3.237
∗∗
0.793 3.140
∗∗
0.774 3.245
∗∗
0.799
(1.442) (0.787) (1.381) (0.776) (1.316) (0.811)
Liquid Reserves_Other Banks 1.943
∗∗∗
0.236
(0.384) (0.426)
Income_Other Banks 2.018
∗∗∗
0.399
(0.355) (0.298)
Capital Ratio_No Fail Banks 3.541
∗∗∗
0.805
(0.924) (1.086)
Macro & Crisis Controls Yes Yes Yes Yes Yes Yes
P-value 0.000 0.000 0.000 0.000 0.000 0.000
Observations 523 523 523
P-value (all) 0.000 0.000 0.000
The table presents the results of a competing risk proportional hazard model for bank failure, where there are
two types of bank exit: takeovers or closures of banks by the government and the acquisition of the bank by
another bank. The model allows for correlated bank exit types. Each column represents a single regression, and
the coefficients for both types of exit in a column are jointly estimated. For each variable, we report the marginal
effect evaluated at the sample mean. A positive effect (in percentage points) denotes an increasing hazard of
bank exit through that type of exit event. Total Assets/GDP is the bank’s total assets normalized by the country’s
GDP. Capital Ratio is the total e quity divided by total assets. Income is operating income divided by total assets.
Before Election is a dummy variable that takes 1 if the bank fails in the latter half of the electoral cycle or, in
the case of no exit, the end of the bank’s accounting year falls within the latter half of the electoral cycle. Macro
and Crisis Control variables include GDP per Capita, GDP Growth, Banking Crisis, Currency Depreciation,
Inflation Rate, Real Interest Rate,andIMF Loans/GDP. Liquid Reserves_Other Banks is the weighted average
(by total assets) of the equity reserves of other banks in that country. Income_Other Banks is the weighted average
(by total assets) of income of other banks in that country. Capital Ratio_No Fail Banks is the weighted average
(by total assets) of the capital ratio of banks in that country that did not fail by government takeover or closure.
All values are as of year t–1. P-values of a Wald test that all coefficients are jointly zero are reported for each
type of bank exit and then for both types of bank exit. Heteroscedasticity-robust standards errors, corrected for
clustering at the country level, are in parentheses.
,
∗∗
,
∗∗∗
denote statistical significance at the 10%, 5%, and
1% levels, respectively.
in the sample period. To check the robustness of our measure to this con-
cern, we construct Capital Ratio_No Fail Banks by excluding banks that failed
at any point in time. Capital Ratio_No Fail Banks also has a positive effect
that is statistically significant at the 5% level. These results indicate that the
Too-Many-to-Fail effect identified in previous sections is robust to using dif-
ferent measures for the financial health of other banks.
6.2 Alternative specifications
We check the robustness of our results to different definitions of bank failure
and bank acquisition. The results are reported in Table 9. We first replace the
bank failure with the first sign of problems, which is defined as the first year
of negative income. We then consider only the acquisitions by foreign banks
as they exit through being acquired. Capital Ratio_Other Banks has a positive
1398
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Tab le 9
Too-Many-to-Fail: alternative specifications
Government Acquisition base
takeover/ Acquisition Government specification
closure initial initial sign takeover/ with acquisitions
sign of banking of banking closure base by foreign
problems problems specification entities only
Total Assets/GDP 2.122
∗∗∗
1.044
2.113
∗∗∗
1.523
(0.456) (0.560) (0.498) (0.910)
Capital Ratio 0.321
∗∗∗
0.012 0.401
∗∗∗
0.036
(0.106) (0.162) (0.082) (0.126)
Income 0.163 0.142 0.136 0.045
(0.112) (0.096) (0.095) (0.084)
Before Election 2.455
∗∗
0.791 3.310
∗∗
0.067
(1.239) (0.802) (1.429) (1.045)
Capital Ratio_Other 1.843
∗∗∗
0.353 1.865
∗∗∗
0.372
Banks (0.389) (0.430) (0.384) (0.406)
Macro & Crisis Controls Yes Yes Yes Yes
P-value 0.000 0.000 0.000 0.000
Observations 522 523
P-value (all) 0.000 0.000
The table presents the results of a competing risk proportional hazard model for bank failure, where there are
two types of bank exit: takeovers or closures of banks by the government and the acquisition of the bank by
another bank. The model allows for correlated bank exit types. Each column represents a single regression, and
the coefficients for both types of exit in a column are jointly estimated. For each variable, we report the marginal
effect evaluated at the sample mean. A positive effect (in percentage points) denotes an increasing hazard of bank
exit through that type of exit event. The first column uses the first occurrence of negative operating income as the
time of bank exit instead of the date of government takeover/closure. The second column uses only acquisitions
by a foreign entity as the occurrence of bank exit instead of acquisitions by all types of entities. The third column
uses the base specification in addition to feedback terms. Total Assets/GDP is the bank’s total assets normalized
by the country’s GDP. Capital Ratio is the total equity divided by total assets. Income is operating income
divided by total assets. Capital Ratio_Other Banks is the weighted average (by total assets) of the capital ratio of
other banks in that country. Macro and C risis Control variables include GDP per Capita, GDP Growth, Banking
Crisis, Currency Depreciation, Inflation Rate, Real Interest Rate,andIMF Loans/GDP. All values are as of year
t–1. Before Election is a dummy variable that takes 1 if the bank fails in the latter half of the electoral cycle or,
in the case of no exit, the end of the bank’s accounting year falls within the latter half of the electoral cycle.
P-values of a Wald test that all coefficients are jointly zero are reported for each type of bank exit and then for
both types of bank exit. Heteroscedasticity-robust standards errors, corrected for clustering at the country level,
are in parentheses.
,
∗∗
,
∗∗∗
denote statistical significance at the 10%, 5%, and 1% levels, respectively.
and statistically significant coefficient in both regressions, which indicates that
our results are robust to different definitions of bank exit.
6.3 Domestic financial development
It may be a concern that our measures of financial health for other banks may
be capturing the level of domestic financial development and other institutional
factors. In order to check the robustness of our findings, we use the following
country-year-level control variables: a creditor rights index, the presence of
a formal deposit insurance scheme, and stock market turnover. The results
are presented in Table 10. None of these factors seem to play a statistically
significant role in the government’s decision to take over or close a failing
bank. However, Capital Ratio_Other Banks continues to have a positive and
statistically significant effect, which indicates that the Too-Many-to-Fail effect
remains robust to controlling for financial development and other institutional
factors.
1399
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Tab le 10
Too-Many-to-Fail: domestic financial development
Government Government Government
takeover/ takeover/ takeover/
closure Acquisition closure Acquisition closure Acquisition
Total Assets/GDP 1.703
∗∗∗
1.057
2.124
∗∗∗
1.026
2.301
∗∗∗
1.060
∗∗
(0.480) (0.580) (0.453) (0.601) (0.414) (0.465)
Capital Ratio 0.408
∗∗∗
0.009 0.393
∗∗∗
0.032 0.409
∗∗∗
0.060
(0.109) (0.151) (0.096) (0.176) (0.076) (0.150)
Income 0.160 0.139 0.170 0.116 1.431
∗∗∗
0.133
(0.125) (0.089) (0.124) (0.114) (0.100) (0.089)
Before Election 3.533
∗∗
0.825 3.351
∗∗
0.726 3.201
∗∗
1.168
(1.423) (0.831) (1.433) (0.792) (1.375) (0.916)
Capital Ratio_Other Banks 2.038
∗∗∗
0.427 1.908
∗∗∗
0.385 1.992
∗∗∗
0.009
(0.499) (0.432) (0.327) (0.427) (0.410) (0.490)
Creditor Rights 1.018 0.061
(0.980) (0.435)
Deposit Insurance 0.750 0.666
(1.775) (1.507)
Stock Market Turnover 0.793 4.927
∗∗
(0.989) (1.944)
Macro & Crisis Controls Yes Yes Yes Yes Yes Yes
P-value 0.000 0.000 0.000 0.000 0.000 0.000
Observations 523 523 523
P-value (all) 0.000 0.000 0.000
The table presents the results of a competing risk proportional hazard model for bank failure, where there are
two types of bank exit: takeovers or closures of banks by the government and the acquisition of the bank by
another bank. The model allows for correlated bank exit types. Each column represents a single regression, and
the coefficients for both types of exit in a column are jointly estimated. For each variable, we report the marginal
effect evaluated at the sample mean. A positive effect (in percentage points) denotes an increasing hazard of bank
exit through that type of exit event. Total Assets/GDP is the bank’s total assets normalized by the country’s GDP.
Capital Ratio is the total equity divided by total assets. Income is operating income divided by total assets. Capital
Ratio_Other Banks is the weighted average (by total assets) of the capital ratio of other banks in that country. All
are book values and as of year t–1. Before Election is a dummy variable that takes 1 if the bank fails in the latter
half of the electoral cycle or, in the case of no exit, the end of the bank’s accounting year falls within the latter
half of the electoral cycle. Creditor Rights represents an index of the quality of creditor rights in that country.
Depositor Insurance is a dummy variable equal to 1 if there is the presence of depositor insurance in that country.
Stock Market Turnover is the ratio of the value of total shares traded to average real market capitalization. Macro
and Crisis Control variables include GDP per Capita, GDP Growth, Banking Crisis, Currency Depreciation,
Inflation Rate, Real Interest Rate,andIMF Loans/GDP. All variables are as of t1. P-values of a Wald test
that all coefficients are jointly zero are reported for each type of bank exit and then for both types of bank exit.
Heteroscedasticity-robust standards errors, corrected for clustering at the country level, are in parentheses.
,
∗∗
,
∗∗∗
denote statistical significance at the 10%, 5%, and 1% levels, respectively.
7. Conclusion
We study banking in major emerging economies throughout the latter part of
the 1990s to demonstrate regulatory forbearance toward failing banks when the
banking sector is weak. This Too-Many-to-Fail effect is unlikely to be limited
to emerging markets. It was present in the U.S. Savings and Loan Crisis of
the 1980s and the Japanese Banking Crisis of the 1990s (see footnote 3). To
the extent that the current U.S. crisis is similar to emerging market crises,
15
as argued by Reinhart and Rogoff (2008), we expect to see a U.S. regulatory
response characterized by Too-Many-to-Fail concerns. In fact, the $700 billion
15
Reinhart and Rogoff (2008) show remarkable similarities between emerging markets and developed countries in
the events leading up to and the impact on the government’s budget subsequent to a financial crisis.
1400
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
TARP—that aims to support weak banks—appears to be motivated by the
large number of weak banks and the widespread nature of the problems in the
banking sector.
Regulatory decisions do not depend only on the characteristics of the bank
in question. This finding has implications for recent policy debates on bank
regulation. The finding suggests that Basel II’s focus on bank characteristics
without proper emphasis on regulatory incentives may be misplaced. The di-
rect bank-level, multicountry evidence presented in this article strengthens the
arguments that designing bank regulation without due concern for regulatory
incentives is not likely to be very productive (Barth, Caprio, and Levine 2006).
Instead, market monitoring of banks may be a welcome augmentation for miti-
gating the negative impact of regulatory incentive issues (Flannery and Sorescu
1996; Berger, Davies, and Flannery 2000; Peria and Schmukler 2001).
Whether it is through acquisition, nationalization, or liquidation, prudential
regulation suggests a change in bank status when the banking system weakens.
The results presented in this article can be interpreted as evidence of neglect
by the government. We provide a note of caution on this interpretation. We
focus on two drastic and costly forms of government intervention: government
takeovers and bank closures. There are many other forms of intervention that
the government can use: liquidity support, purchase of nonperforming assets,
and other short-term aid. We do not consider limited forms of government
intervention because, to the best of our knowledge, there is no reliable dataset
on these types of actions.
16
Our findings may not extend to these limited forms
of government intervention.
The econometric methodology that we use to study bank failure may be of
independent interest for other bank failure studies and for bankruptcy studies
in general. Many weak banks exit the sector not just through the government
actions of takeovers or liquidation but also through acquisition. To the extent
that acquisitions are not independent of government action—and there are many
reasons why they may not be—bank failure studies must allow for exit through
acquisitions. We hope that the competing risk method will become a standard
approach in this regard.
The results presented in this article lead to several questions for further
research. Does the Too-Many-to-Fail effect lead to bank herding ex ante? A
banker may be more likely to take risks or lend to the same sectors (e.g., real
estate) if he/she knows that his/her bank is less likely to be closed or taken over
when subsequent problems appear to be system wide.
17
How costly is the Too-
Many-to-Fail effect? How costly is regulatory forbearance in general? Given the
16
Any such dataset is likely to be incomplete because regulators have an incentive not to disclose information about
these limited forms of intervention in order to prevent bank runs. Even during the onset of the subprime crisis
in the United States, the Federal Reserve Bank introduced a policy to provide liquidity support to weak banks
without disclosing their identity. This policy shift was made in order to supplement its usual discount window
through which it could provide liquidity support to a bank only by disclosing the bank’s identity.
17
Acharya and Yorulmazer (2007) provide evidence that suggests that this type of herding exists.
1401
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
distortions caused by government ownership in banking,
18
how are regulatory
forbearance costs affected when failing banks are taken over or “nationalized”
by the government? Finally, Gorton and Rosen (1995) and Dinc¸ (2006) show
that the corporate governance of banks plays an important role in risk taking by
banks. Does corporate governance of banks play an important role in regulatory
forbearance? We leave these interesting questions for future research.
Appendix: Interaction Effects and Competing Risk
In an effort to explain the timing of bank failures, we estimate several interaction effects. For linear
models, the coefficient for an interaction term can be easily interpreted as the interaction effect. For
example, let us allow a continuous variable y to depend on two continuous variables x
1
, x
2
, their
interaction, and a vector of additional independent variables X including a constant term. Hence
the data-generating process for y is as follows:
y
i
= β
0
+ β
1
x
i1
+ β
2
x
i2
+ β
12
x
i1
x
i2
+ X
i ,(1,2,12)
β
,(1,2,12)
+ u
i
. (A1)
The interaction effect of the independent variables is the cross-derivative of the expected value
of y
i
:
2
E
y|x
1
, x
2
, X
x
1
x
2
= β
12
. (A2)
In this article, we present a competing risk proportional hazard model for bank failure. This
model is nonlinear in the estimated coefficients. In nonlinear models, the interaction effect is not
equal to the coefficient for the interaction term. Following Ai and Norton (2003), we present the
correct way to recover interaction effect estimates and standard errors for nonlinear models using
continuous variables. We then present the correct way to recover interaction effect estimates and
standard errors in our model using one continuous variable and one binary variable.
Consider the following type-specific hazard:
λ
j
(t|X) = b
j
(t)exp(β
j1
x
1
+ β
j2
x
2
+ β
j12
x
1
x
2
+ X
j,(1,2,12)
β
j,(1,2,12)
+ ε
j
). (A3)
Let x
1
and x
2
be continuous variables. The interaction effect is given by the cross-derivative of the
type-specific hazard:
μ
12
= λ
j
(t|X)
[(β
j1
+ β
j12
x
2
)(β
j1
+ β
j12
x
1
) + β
j12
]. (A4)
Note that the interaction effect is conditional on the explanatory variables. The interaction effect
is estimated by
ˆ
μ
12
= λ
j
(t|X)
[(
ˆ
β
j1
+
ˆ
β
j12
x
2
)(
ˆ
β
j1
+
ˆ
β
j12
x
1
) +
ˆ
β
j12
]. (A5)
The continuity of the type-specific hazard and the consistency of the estimated coefficients ensure
the consistency of the interaction effect estimator. The standard error of the estimated interaction
effect is found by applying the Delta method. Hence, the asymptotic variance of the estimated
interaction effect is itself estimated consistently by
ˆ
σ
12
=
β
[
ˆ
μ
12
]
ˆ
β
β
[
ˆ
μ
12
], (A6)
18
See Sapienza (2004), Dinc¸ (2005), and Khwaja and Mian (2005).
1402
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
where
β
is the covariance for β.
In the article, we present the following type-specific hazard:
λ
j
(t|M, X) = b
j
(t)exp(β
j1
x
1
+ β
jM
M + β
j1M
x
1
M + X
j,(1,M,1M )
β
j,(1,M,1M )
+ ε
j
), (A7)
where M is a dummy variable and can take one of two values: 0 or 1. The interaction ef-
fect is given by the finite difference of the derivative of the type-specific hazard with respect
to x
1
:
μ
12
= (β
j1
+ β
j1M
)
λ
j
(t|M = 1, X ) β
j1
λ
j
(t|M = 0, X). (A8)
In the article, we present the interaction effect and the standard errors given the hazard at the sample
mean. We also provide the t-statistic for each estimate to test the hypothesis that the interaction
effect equals zero.
References
Acharya, V. V., and T. Yorulmazer. 2007. Too Many to Fail—An Analysis of Time-inconsistency in Bank Closure
Policies. Journal of Financial Intermediation 16:1–31.
Acharya, V. V., and T. Yorulmazer. 2008. Cash-in-the-Market Pricing and Optimal Resolution of Bank Failures.
Review of Financial Studies 21:2705–42.
Aguiar, M., and G. Gopinath. 2005. Fire-sale FDI and Liquidity Crises. Review of Economics and Statistics
87(3):439–52.
Ai, C., and E. C. Norton. 2003. Interaction Terms in Logit and Probit Models. Economics Letters 80(1):123–29.
Allen, F., and D. Gale. 2000. Financial Contagion. Journal of Political Economy 108:1–33.
Amyx, J. A. 2004. Japan’s Financial Crisis. Princeton, NJ: Princeton University Press.
Bankscope. CD-ROM and Internet access, various years.
Barth, J. 1991. The Great Savings and Loan Debacle. Washington, DC: AEI Press.
Barth, J. R., G. Caprio, and R. Levine. 2006. Rethinking Bank Regulation. Cambridge: Cambridge University
Press.
Beck, T., A. Demirguc-Kunt, and R. Levine. 2006. Bank Concentration and Crises: First Results. Journal of
Banking and Finance 30(5):1581–603.
Berger, A., S. M. Davies, and M. J. Flannery. 2000. Comparing Market and Regulatory Assessments of Bank
Performance: Who Knows What When? Journal of Money, Credit and Banking (Pt. 2):641–67.
Berglof, E., and P. Bolton. 2002. The Great Divide and Beyond: Financial Architecture in Transition. Journal of
Economic Perspectives 16(1):77–100.
Bongini, P., S. Claessens, and G. Ferri. 2001. The Political Economy of Distress in East Asian Financial
Institutions. Journal of Financial Services Research 19:5–25.
Bongini, P., L. Laeven, and G. Majnoni. 2002. How Good Is the Market at Assessing Bank Fragility? A Horse
Race between Different Indicators. Journal of Banking and Finance 26:1011–28.
Brown, C. O., and I. S. Dinc¸. 2005. The Politics of Bank Failures: Evidence from Emerging Markets. Quarterly
Journal of Economics 120(4):1413–44.
Calomiris, C. W., and J. R. Mason. 1997. Contagion and Bank Failures during the Great Depression: The June
1932 Chicago Banking Panic. American Economic Review 87:863–83.
1403
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Calomiris, C. W., and J. R. Mason. 2003. Fundamentals, Panics and Bank Distress during the Depression.
American Economic Review 93:1615–47.
Caprio, G., and D. Klingebiel. 2002. Episodes of Systematic and Borderline Financial Crises. Working Paper,
World Bank.
Carletti, E., P. Hartmann, and G. Spagnolo. 2004. Bank Mergers Competition and Liquidity. Working Paper,
ECB.
Claessens, S., D. Klingebiel, and L. Laeven. 2005. Crisis Resolution, Policies, and Institutions: Empirical
Evidence. In P. Honohan and L. Laeven (eds.), Systemic Financial Crises: Containment and Resolution,
pp. 169–94. Cambridge and New York: Cambridge University Press.
Demirguc-Kunt, A., and E. Detragiache. 2002. Does Deposit Insurance Increase Banking Stability? An Empirical
Investigation. Journal of Monetary Economics 49:1373–406.
Demirguc-Kunt, A., and E. Detragiache. 2005. Cross Country Empirical Studies of Banking System Distress: A
Survey. In P. Davis (ed.), Financial Instability, Asset Prices, and Credit, No. 192 (April), pp. 68–83. London:
National Institute Economic Review.
Deng, Y., J. M. Quigley, and R. Van Order. 2000. Mortgage Terminations, Heterogeneity and the Exercise of
Mortgage Options. Econometrica 68(2):275–307.
Diamond, D., and R. Rajan. 2005. Liquidity Shortages and Banking Crises. Journal of Finance 60(2):615–
48.
Dinc¸, I. S. 2005. Politicians and Banks: Political Influences on Government-owned Banks in Emerging Markets.
Journal of Financial Economics 77:453–79.
Dinc¸, I. S. 2006. Monitoring the Monitors: The Corporate Governance in Japanese Banks and Their Real Estate
Lending in the 1980s. Journal of Business 79(6):3057–81.
Djankov, S., C. McLiesh, and A. Shleifer. 2007. Private Credit in 129 Countries. Journal of Financial Economics
84(2):299–329.
Dolton, P., and W. Van Der Klaauw. 1999. The Turnover of Teachers: A Competing Risks Explanation. The
Review of Economics and Statistics 81(3):543–50.
Factiva. online access, www.factiva.com, Dow Jones and Reuters.
Fallick, B., and K. Ryu. 2007. The Recall and New Job Search of Laid-off Workers: A Bivariate Propor-
tional Hazard Model with Unobserved Heterogeneity. The Review of Economics and Statistics 89(2):313–
23.
Flannery, M. J., and S. M. Sorescu. 1996. Evidence of Bank Market Discipline in Subordinated Debenture Yields:
1983–1991. Journal of Finance 51(4):1347–77.
Gorton, G., and R. Rosen. 1995. Corporate Control, Portfolio Choice, and the Decline in Banking. Journal of
Finance 50:1377–420.
Han, A., and J. Hausman. 1990. Flexible Parametric Estimation of Duration and Competing Risk Models. Journal
of Applied Econometrics 5:1–28.
Heckman, J., and B. Singer. 1984. A Method for Minimizing the Impact of Distributional Assumptions in
Econometric Models for Duration Data. Econometrica 52(2):271–320.
Hoffman, P., and A. Santomero. 1998. Problem Bank Resolution: Evaluating the Options. Working Paper,
Wharton Financial Institutions Center.
Hoggarth, G., J. Reidhill, and P. Sinclair. 2004. On the Resolution of Banking Crises: Theory and Evidence.
Working Paper No. 229, Bank of England.
Honore, B. E., and E. Tamer. 2006. Bounds on Parameters in Panel Dynamic Discrete Choice Models. Econo-
metrica 74(3):611–29.
1404
The Review of Financial Studies / v 24 n 4 2011
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
Hoshi, T., and A. Kashyap. 2001. Corporate Financing and Governance in Japan. Cambridge, MA: MIT Press.
Kalbfleisch, J., and R. Prentice. 2002. The Statistical Analysis of Failure Time Data, 2nd ed. New York: Wiley.
Kaminsky, G. 2003. Varieties of Currency Crises. Working Paper No. W10193, NBER. Available at SSRN:
http://ssrn.com/abstract=483124.
Kane, E. J. 1989. The S&L Insurance Mess: How Did It Happen? Washington, DC: Urban Institute.
Kasa, K., and M. Spiegel. 2008. The Role of Relative Performance in Bank Closure Decisions. Economic Review,
Federal Reserve Bank of San Francisco, pp. 19–29.
Khwaja, A. I., and A. Mian. 2005. Do Lenders Favor Politically Connected Firms? Rent Provision in an Emerging
Financial Market. Quarterly Journal of Economics 120(4):1371–411.
Kroszner, R. S., and P. E. Strahan. 1996. Regulatory Incentives and the Thrift Crisis: Dividends, Mutual-to-Stock
Conversions, and Financial Distress. Journal of Finance 51:1285–319.
Lancaster, T. 1990. The Econometric Analysis of Transition Data. Cambridge: Cambridge University Press.
Lane, W. R., S. W. Looney, and J. W. Wansley. 1986. An Application of the Cox Proportional Hazards Model to
Bank Failure. Journal of Banking and Finance 10:511–31.
Lang, L., and R. Stulz. 1992. Contagion and Competitive Intra-industry Effects of Bankruptcy Announcements.
Journal of Financial Economics 32:45–60.
McCall, B. 1996. Unemployment Insurance Rules, Joblessness, and Part-time Work. Econometrica 64:647–82.
Mitchell, J. 2001. Too Many to Fail and Regulatory Response to Banking Crises. Working Paper, Facultes
Universitaires Saint-Louis, Brussels.
Molina, C. A. 2002. Predicting Bank Failures Using a Hazard Model: The Venezuelan Banking Crisis. Emerging
Markets Review 3:31–50.
Peria, M. S. M., and S. L. Schmukler. 2001. Do Depositors Punish Banks for Bad Behavior? Market Discipline,
Deposit Insurance, and Banking Crises. Journal of Finance 56(3):1029–51.
Perotti, E. 1998. Inertial Credit and Opportunistic Arrears in Transition. European Economic Review 42:1703–25.
Reinhart, C. M., and K. S. Rogoff. 2008. Banking Crises: An Equal Opportunity Menace. Working Paper No.
W14587, NBER. Available at SSRN: http://ssrn.com/abstract = 1320833.
Roland, G., and T. Verdier. 1994. Privatization in Eastern Europe: Irreversibility and Critical Mass Effect. Journal
of Public Economics 54(2):161–83.
Sapienza, P. 2004. What Do State-owned Firms Maximize? Evidence from the Italian Banks. Journal of Financial
Economics 72:357–84.
Shumway, T. 2001. Forecasting Bankruptcy More Accurately: A Simple Hazard Model. Journal of Business
74:101–24.
Slovin, M. B., M. E. Sushka, and J. A. Polonchek. 1999. An Analysis of Contagion and Competitive Effects at
Commercial Banks. Journal of Financial Economics 54:197–225.
Sueyoshi, G. 1992. Semi-parametric Proportional Hazards Estimation of Competing Risks Models with Time-
varying Covariates. Journal of Econometrics 51:25–58.
Whalen, G. 1991. A Proportional Hazards Model of Bank Failure: An Examination of Its Usefulness as an Early
Warning Tool. Economic Review 27:21–31.
White, L. J. 1991. The S&L Debacle: Public Policy Lessons for Bank and Thrift Regulation. New York: Oxford
University Press.
1405
Too Many to Fail?
at Rutgers University on September 21, 2011rfs.oxfordjournals.orgDownloaded from
... As discussed in Section 2.1, regulators may engage in forbearance when facing resource constraints that inhibit their ability to intervene into all troubled banks. This idea is supported by prior research suggesting that forbearance is more likely when the banking sector is weaker and resources are thus more likely to be constrained (Brown and Dinç 2011). Furthermore, when banking sectors are weaker, intervening into one bank may spark concerns about the viability of nearby banks. ...
... While I employ several approaches to mitigate concerns about specific sources of endogeneity, I cannot account for all potential sources. Second, I cannot directly observe forbearance; instead, I infer it using actual interventions, following prior research (Brown and Dinç 2011). While the evidence from the mechanism analyses is consistent with financial reporting opacity facilitating forbearance, I cannot eliminate the possibility that other explanations are responsible. ...
... Second, the optimality of bank opacity depends on the motivation(s) behind forbearance. Prior research suggests that forbearance could be beneficial if it helps regulators manage constrained resources, minimize resolution costs, or mitigate contagion (Santomero and Hoffman 1998;Brown and Dinç 2011;Morrison and White 2013). My analyses provide evidence consistent with financial reporting opacity facilitating forbearance motivated by these reasons. ...
Article
Full-text available
I study the association between bank financial reporting opacity, measured by delayed expected loan loss recognition, and the intervention decisions made by bank regulators. Examining U.S. commercial banks during the 2007–09 financial crisis, I find that delayed expected loan loss recognition is negatively associated with the likelihood of regulatory intervention (measured by either severe enforcement action or closure). This result is robust to using various specifications and research designs. In additional analyses, I find evidence suggesting that this association is driven by regulators exploiting financial reporting opacity to practice forbearance. My findings contribute to the extant literature on bank opacity, regulatory forbearance, and the consequences of loan loss provisioning by suggesting that delayed expected loan loss recognition affects regulatory intervention decisions.
... However, the formation and survival of zombie firms is not only derived from the blood transfusion from banking sector, but also from the "assistant hand" of government. For the sake of its own interests, government may prevent firms from withdrawing from the market by providing direct financial assistance or deregulation (Brown & Dinc, 2011;Hoshi & Kashyap, 2010;Jaskowski, 2015;Willam, 2014). As a transitional economy, government intervention in the operation of China's economy is prominent. ...
Article
Full-text available
One of the most important tasks in China’s supply-side structural reform is to effectively clean up zombie firms. Using industrial firm-level data in China from 1998 to 2007, we identified zombie firms in Chinese industrial sectors. Based on the identified zombie firms, we measured the survival duration of zombie firms and empirically studied the effect of fiscal subsidies on the survival duration of zombie firms by using survival analysis. Results revealed that fiscal subsidies significantly lengthened the survival duration of zombie firms. Specifically, fiscal subsidies not only reduced the possibility of zombie firms exiting the market, but more importantly, they reduced the possibility of zombie firms’ resurrection. Thus, fiscal subsidies do not contribute to the “death” or “resurrection” of zombie firms. These results are robust despite changing the identification method and the survival analysis method, and considering the endogeneity of fiscal subsidies. Our findings show that we should adhere to market-oriented mechanism in disposing zombie firms and cut off the channels that government transfuses to zombie firms by fiscal subsidies.
... Sử dụng độ lệch chuẩn của ROE như là một thước đo chính cho rủi ro của ngân hàng, DeYoung và Rice (2004) cho thấy rằng các ngân hàng có quy mô vừa và nhỏ có mức độ cao hơn về rủi ro so với các ngân hàng có quy mô lớn. Điều này có thể được giải thích bởi các ngân hàng lớn có khả năng đa dạng hóa danh mục đầu tư, từ đó giảm thiểu rủi ro tốt hơn so với các ngân hàng nhỏ (Brown & Dinç, 2011). ...
Article
Full-text available
This research examines the influence of geopolitical risk on the financial stability of the banking system in Asian countries. Using a sample of 776 commercial banks in 8 Asian countries from 2006 to 2019, we find that geopolitical risk has a negative impact on bank stability. The research results are robust to the GMM estimation method to eliminate endogeneity caveats. We also add interactive variables, which will help to more accurately reflect the role of moderating factors such as bank size and the capital ratio on the association between geopolitical risk and bank financial stability. The test results show that the negative effect of geopolitical risk on bank financial stability is more severe in banks with small sizes and low capital rates.
... [Insert Table 8 about here] 1, 2, 4, 5)), which is supported by the literature since large banks are more likely to benefit from regulatory forbearance and this might create incentives to invest in risky activities (Brown and Dinc, 2009), weakening the effectiveness of macroprudential policies. Although large banks actively build-up capital that exceeds the regulatory minimum requirements (Berger et al., 2008), they are also bettercapitalized associated with a lower contribution to systemic risk, but they might engage in risky operations (Perotti et al., 2011), which leads to lower financial stability. ...
Preprint
Full-text available
This paper investigates the complementarity between the different macroprudential policies to contain bank systemic risk. We use a newly updated version of the IMF survey on Global Macroprudential Policy Instruments (GMPI). By disentangling the aggregate macroprudential policy index, we assess the complementarity between borrower-targeted and lender-targeted instruments in mitigating systemic risk arising from intra-financial system vulnerabilities. We investigate the effect of boom-bust cycle on such a relationship by analyzing the financial upturns and downturns and show the effectiveness of the macroprudential policies during calm period. We also show that their efficacy in mitigating instability is quite heterogeneous and may vary depending on the set of tools implemented, as well as bank' size, TBTF, leverage, liquidity and concentration. Our results bear critical policy implications for implementing optimal macroprudential tools and provide insights into the trade-off between financial vis-à-vis price stability. JEL Codes: E32, E58, F44, G18, G21, G32
... According to Brown and Dinç (2011), the Banking Sector is a highly regulated sector of activity, which means, for each institution, the imperative need to respond adequately, accurately, correctly and in a timely manner to the multiple requests, both regular and untimely, requested by the regulatory authorities. ...
Article
Full-text available
For some years now, master data has become extremely relevant to business success and continuity in an increasingly competitive and global business environment. The banking sector is one example of how the implementation of well-structured and designed master data management policies and initiatives is crucial for reaching positive results. One of the areas in which banks need to ensure extremely fruitful master data management approaches and data governance procedures is when dealing with risk-related data, as it not only ensures accurate and well-supported management and decision-making, but also because banks are required to do so by imposed regulations, such as the BCBS 239. Drawing on a DSR methodology supported research project, where banking and IS-related expertise was continuously merged with existing theoretical knowledge on MDM and BCBS 239 related topics, and a permanent focus on the technical and functional complexity associated with implementing master data management and well-established data governance procedures that ensure regulatory compliance, we propose a novel, six-phase action plan that will allow banks to ensure compliance with BCBS 239 and, consequently, ensure efficient and effective risk data management and reporting.