Institutions, Moral Hazard and Expected Government Support of Banks
Angelos A. Antzoulatos*
[email protected]
(+30) 210-414-2185
Chris Tsoumas
[email protected]
(+30) 210-414-2155
Department of Banking & Finance, University of Piraeus
&
School of Social Sciences, Hellenic Open University
Abstract.
We model the expected support of banks with credit ratings from Moody’s, taking explicitly
into account the capacity and willingness of governments to provide support in case of need,
as well as their concerns about moral hazard (i.e., that the expected support may induce banks
to assume bigger risks). Our results, which are bolstered by extensive robustness checks,
suggest that moral hazard concerns are relatively weak. In addition, a substantial part of the
expected support can be attributed to the quality of a country’s institutions. These findings
have important implications for the dynamics of banking crises, the value of the ‘fair’
insurance premium banks might be called upon to pay for the expected support, as well as for
ways to reduce the pertinent negative externalities.
J.E.L. classification numbers: G21, G24, G28
Keywords: Banks, Credit Ratings, Government Support, Institutions, Moody’s,
Moral Hazard
*
Corresponding author. We thank Alex Cukierman, Fotis Pasiouras, Emmanouil Tsiritakis and seminar
participants at the conference on ‘Banking, Finance, Money and Institutions: The Post-crisis Era’
(University of Surrey), November 2013. We also thank Ekaterini Tarasidou for able and conscientious
research assistance, and the University of Piraeus Research Center for generous financial support. The
usual disclaimer applies.
1. Introduction
The global financial crisis, which started in 2007, has brought forcefully to the center
of academic and policy debates the expected support of banks in need by
governments. The growing literature has explored several issues, such as, the size and
determinants of the expected support, the funding advantage banks derive from it, the
pertinent distortions in competition and the moral hazard it creates for banks. The
latter refers to the concern that, in the expectation that support will be extended in
case of need, banks will ex-ante assume more risks, thus making support more likely,
bigger and, perhaps, less affordable.
Indicatively, Schich and Lindh (2012) measure the expected support as the
difference between two credit ratings: an all-in rating —which encompasses expected
support, and a stand-alone rating, both from Moody’s (details are provided below).
This support lowers banks’ cost of funding (Morgan and Stiroh [2005]), as it is
recognized by market participants. The estimates of the pertinent funding advantage
vary widely, but are substantial. For example, Ueda and Weder di Mauro (2012),
using credit ratings from Fitch, estimate it to between 60 – 80bp (For a neat
presentation of ways to measure it, see Noss and Sowerbutts [2012].). The expected
support also affects banks in subtler ways as it reduces the capital they must set aside
for their holdings of other banks’ debt; it also allows the use of such debt as collateral
in central bank funding.
Since its level varies across countries and, within countries, across banks (Schich
and Lindh [2012]), the expected support may distort competition. Briefly, it may
tempt banks to assume more risks, as market discipline weakens, or even push their
competitors to do so – the rationale being that a bank’s funding advantage leads to
fiercer competition that reduces the franchise value of its competitors (Gropp et al.
[2011]). Moreover, if, as is widely believed, it is positively related to the size of a
bank, it may tempt banks to expand – another form of moral hazard which
exacerbates the negative externalities of bank fragility. Last, but not least, the
expected support creates a potentially destabilizing feedback loop between
government creditworthiness and bank fragility (see, for example, Estrella and Schich
[2012]), that is, as bank fragility increases, the contingent fiscal cost of the expected
2
support may perversely affect government’s creditworthiness. Closing the loop, the
decrease in government’s creditworthiness may lead to increased bank fragility
through the reduced capacity to support.
Market participants question the strength of moral hazard, but, as the literature on
financial safety nets indicates, it is a possibility and, hence, an open question. The
theoretical papers support both views, i.e., for and against the safety nets (see, for
example, Diamond and Dybvic [1983] and Diamond and Rajan [2000]). The
empirical evidence may seemingly support the negative view (see, for example,
Demirgüç-Kunt and Detragiache [2002], Barth et al. [2004] and, more recently, Dam
and Koetter [2012]), but one cannot dismiss the possibility of ‘near-misses’ which are
not recorded in the literature – that is, of crises that have been averted because of the
existence of safety nets. In any event, the existing literature on the expected support of
banks is primarily focused on the ‘negatives’ and largely overlooks the potential
benefits of a government safety net, part of which is the expected support. This
implies that its policy recommendations must be viewed with caution.
Thus motivated, we take a fresh and broader look at the expected support of
banks, using –as Schich and Lindh (2012)– bank credit ratings from Moody’s. In
doing so, we depart from the existing literature in several important ways. To begin
with, we try to explicitly consider a government’s capacity to provide support to a
bank in need, its willingness to do so or not being in default when support is needed.
Everything else equal, government’s capacity to support a bank in need will affect
expected support positively; and vice-versa. Likewise for government’s willingness.
Willingness is associated with the systemic significance of a bank or of the banking
sector in an economy and, hence, with the potential economic repercussions of bank
fragility: The bigger the systemic significance, the higher the willingness; and viceversa. Potentially restraining the expected support are the aforementioned moral
hazard concerns.
This approach allows us to evaluate the relative significance of moral hazard
concerns embedded in Moody’s ratings and, by extension, in market expectations of
government support. In addition, it guides us to use potential explanatory variables
related to the size and structure of a country’s banking sector, as well as to the quality
of its institutional environment. To the best of our knowledge, such country-specific
3
variables have not been used in the literature so far. Proxies of the former include
private credit to GDP, bank assets to GDP and concentration. The rationale is that, in
evaluating the expected support of a bank, one has to take into account the potential
repercussions of generalized financial fragility caused by this bank’s fragility.
As for the variables pertaining to the institutional environment, the reaction, or
fear of such reaction, by the markets and the public in general, may be a strong
proponent for or deterrent of support. For example, in a country with strong
institutions, which –among other things— may restrain excessive risk-taking by banks
through more effective supervision and stronger market discipline, non-support for a
bank in need might be regarded negatively by the markets as well as by a public
apprehensive of the potential repercussions of financial fragility. Conversely, in a
country with weak institutions support might be construed as rewarding imprudent
banks and heightening moral hazard and, hence, regarded negatively. Hence, public
hostility to bank support may be smaller in the first country and, as a result, capacity
and willingness to support may be higher while moral hazard concerns lower. Our
proxies for the quality of the institutional environment are two indices from the Fraser
Institute that measure legal and property rights, and credit regulations.
Lastly, we conduct the analysis year-by-year for the period 2007 – 2011. Data
availability changes over time, and 2007 is the first year with enough observations.
Importantly, 2007 is likely the last year for which Moody’s ratings were not affected
by the unfolding and intensifying global financial crisis. The crisis forced
governments to switch to a crisis-containment mode and, as a result, probably affected
Moody’s assessment about expected support, suggesting a structural break.
The results, from a sample of about 600 banks from all over the world, suggest
that moral hazard concerns were weaker than governments’ capacity and willingness,
and more so after the eruption of the crisis. This is in tune with market-participants’
claim that this fear is exaggerated. As such, it has important implications for the
dynamics of banking crises —provided, of course, that Moody’s ratings capture/affect
market expectations. Briefly, if a government withholds support to a bank in need for
a fear which market-participants regard as inconsequential, destabilizing expectations
might develop, turning a manageable situation into a full-blown banking crisis. This is
similar to the ‘too little, too late’ criticism often directed at central banks’ monetary-
4
policy decisions. Strengthening this implication, the results also suggest that Moody’s
ratings did not embed the possibility of collective risk-taking by banks, in the spirit of
Bonfim and Kim (2012); that is, in the expectation of support in case of need, banks
may engage in collective risk-taking strategies which increase systemic fragility.
Perhaps more importantly, a substantial part of the expected support can be
attributed to a country’s institutional environment and to the size and structure of its
banking sector. A good institutional environment is associated with higher expected
support, while a large and concentrated banking sector with lower. Among the
pertinent policy implications, when quantifying the funding advantage banks derive
from the expected support and pricing a ‘fair’ insurance premium they must pay for,
one should take into account that similar benefits from a good institutional
environment accrue to other private enterprises as well. In addition, such benefits
accrue to both big and small banks, hence the focus of the pertinent policy discussions
on the former may be misplaced. Lastly, policies to reduce the expected support may
have undesirable side effects. For example, forcing banks to downsize has two
opposing effects on expected support: negative from the smaller size, and positive
from the lower concentration.
The remainder of the paper is as follows. Section 2 discusses the expected
support of banks and the potential explanatory variables. Section 3 presents the data
sources and the econometric specification. Section 4 presents the empirical results,
while Section 5 elaborates on their policy implications.
2. Logical Foundations
Related studies
In Schich and Lindh (2012), whose work is closer to this one, the all-in rating is a
bank’s long term deposit rating, which includes the expected support, while the standalone rating its financial stability rating (symbol BFSR). Using a sample of 123 large
European banks, they find that their difference, the proxy of the expected support, is
positively related to the sovereign rating and negatively related to the stand alone
rating. In a similar spirit, Ueda and Weder di Mauro (2012) use ratings from Fitch to
5
explore the effect of expected support on a bank’s long-term rating and to quantify the
pertinent funding cost advantage mentioned above. They find that the long-term rating
is explained by a bank’s financial strength rating –Fitch’s equivalent of BFSR—, the
expected support plus the sovereign rating; all with a positive sign.
Our work is also related, but distinct from, papers that explain BFSR using
publicly known data, starting with the work of Poon et al. (1999). Interestingly, Poon
et al. find that country risk indicators do not appear to be significant predictors of
BFSR – in line with Moody’s assertion that BFSR is “intended to provide a globally
consistent measure of a bank’s financial condition before considering external support
factors that might reduce default risk, or country risks that might increase default
risk.” (Moody’s, 2007b, p.6). Provided this holds for the period of our analysis, 2007
– 2011, it increases the significance of our findings which suggest that such indicators
do affect the expected support. In a more recent paper, which additionally provides an
illuminating review of related studies, Shen et al. (2012) introduce the quality of
information and of the institutional environment as a potential determinant of the
long-term credit ratings of commercial banks in 86 countries, for the period 2002 –
2008, in addition to bank indicators pertaining to capital, efficiency, liquidity and
profitability, and to the size of a bank and the sovereign rating. They find that better
information quality is associated with higher stand-alone ratings. Lastly, Peresetsky
and Karminsky (2008) find that, in addition to bank-specific ratios, a corruptionperceptions index affects both BFSR and deposit ratings negatively.
Dependent variable
Our variable of interest, DIFF, is the difference between the bank rating that
encompasses all kinds of expected support in case of need, the so-called all-in rating,
and the stand-alone rating, both from Moody's. To arrive at the long-term deposit
rating, Moody’s starts from the latter and sequentially takes into account expected
support from operating parent, cooperative group, regional government and national
government, (Moody’s 2007b). We focus on the last type of support. This means that
DIFF is a noisy measure of expected government support. Yet, as elaborated below,
this noise is likely smaller than it appears at a first glance.
6
The stand-alone rating, BFSR, is intended to provide a measure of a bank’s
financial condition that is comparable across countries. As such, it does not
incorporate any expected support due, for example, to ‘too big to fail’ considerations,
nor does it take into account the risk of a deposit freeze (Moody’s [2007b]). In
essence, BFSR is the local currency deposit rating that would be assigned by Moody’s
without any expected external support. In contrast, as Packer and Tarashev (2011)
neatly summarize, the all-in-ratings reflect “the guarantor’s capacity to provide
support, its willingness to provide support and the probability that it is in default when
the bank needs support.”
The guarantor’s capacity to provide support is typically measured by its own
credit rating. Pertaining to a national government, its capacity to provide support may
exceed that implied by its own domestic-currency debt obligations, for support may
take different forms, such as, liquidity provision. As a result, the local-currency
deposit rating – symbol LTDR— may exceed the government’s local-currency bond
rating. In addition, a government’s capacity may not change one-to-one with its
rating.
Moody’s assesses the willingness of a government to provide support in two
steps. The first is associated with the banking sector as a whole, the second with each
individual bank. Thus, the first step relies on the importance of the banking sector for
the national economy and its overall strength, while the second relies on the
importance of a bank in the national economy.
Staying as close as possible to the spirit of the rating process, we use LTDR as the
all-in-rating. Other researchers use different, but highly correlated with this, ratings.
For example, Packer and Tarashev (2011) use the issuer rating, while Schich and
Lindh (2012) use the long-term issuer rating and the senior unsecured rating and –
when not available— the long-term bank deposit rating.
Long-term   Financial 
DIFF  

 LTDR  BFSR
Deposit
Rating   Strength Rating 

DIFF incorporates all four types of expected support. Given that our focus is on the
expected support by national governments, this introduces measurement error to the
7
dependent variable. Yet, this error is smaller than it seems at a first glance, for the
capacity –and, probably, the willingness— of the operating parent, the cooperative
group and the regional government to provide support is likely boosted by the
national government. As Moody’s states, “when multiple forms of support are
anticipated, systemic support (the final stage) will be further shaded in order to avoid
double-counting external support.” (Moody’s, 2007b, p. 5).
LTDR and BFSR are in different scales. The first, like all other measures of the
all-in ratings, are in the standard Moody’s scale, Aaa to C. The second is in the scale
A to E. Following the typical practice in the empirical literature on credit ratings, we
convert the first scale to a numerical one, with Aaa assigned the value 20, Aa1 the
value 19, all the way to C which is assigned the value zero. As for the BFSR, a typical
assignment is as follows: A  12, A-  11, ..., E  0. Due, however, to the different
scales, a notch does not have the same value for the two ratings. To overcome the
difficulties arising from the different scales, we employ Moody’s mapping of BFSR to
the standard scale (Moody’s 2007a and 2009), as shown in appendix table A1.
Potential independent variables
Building upon the existing studies, we use an extensive set of potential explanatory
variables. Going on step further, however, we also try to explicitly consider which
variables, both bank-specific and country-specific, might affect the government’s
capacity, its willingness or not being in default when support is needed, using as guide
Moody’s rating process outlined above. By doing so, we identify several variables
that, to the best of our knowledge, have not been considered in previous research,
notably, variables that pertain to the size and structure of a country’s banking sector,
and to its institutional environment.
Another guide for the analysis is the strategic interaction of governments and
banks which, presumably, is behind the aforementioned moral hazard concerns. That
is, the fear that a bank, in the expectation that support will be extended in case of
need, will ex-ante assume more risks, thus making support more likely and, possibly,
bigger and less affordable.
8
The setting is as follows. A government that contemplates whether to support a
bank in need is concerned about the creation of perverse incentives in case of support.
Such incentives may lead to developments that in the future will be beyond its
capacity to handle/remedy, say, an overblown banking sector or a very large bank. In
addition, the prospect of such developments may tie its hands now because of the fear
of hostile reactions by the markets and the public. The government’s hands may also
be tied in case a bank’s need is –or is perceived to be— the result of pre-existing
perverse incentives; that is, the result of excessive risk-taking in the past, in
anticipation of the expected support.
In mathematical terms,

 




Moral
 Expected 
 Capa- Willin-  Pr ob.of Not  


    
, 
, 
, 

  city   gness   being in Default  Hazard Concerns 
 Support 




(1)
where the ‘+’ or ‘-’ sign above a variable indicates that it is positively or negatively
related with expected support.
More succinctly,
 

 
DIFF     C,W , PND, MH 

(2)

where the dependent variable DIFF has replaced ‘Expected Support’, while the
symbols C, W, PND and MH correspond to the variables in equation (1).
Variables that affect positively government’s capacity, will affect DIFF
positively; and vice-versa. For such a variable Z,

DIFF DIFF C
 DIFF 
 C 

 sign
  sign

Z
C Z
 Z 
 Z 
9
Likewise for variables that affect positively government’s willingness or not being in
default. In contrast, variables that affect positively moral hazard concerns will affect
DIFF negatively; and vice-versa.
More generally, the marginal effect of a variable Z on the expected support is




DIFF DIFF C DIFF W DIFF PND DIFF MH




Z
C Z
W Z PND Z
MH Z
(3)
where the ‘+’ or ‘-’ signs above the partial derivatives denote their expected signs.
Capacity depends positively upon the government’s own financial condition and
the room for support, and negatively upon the size of the potential support. Capacity
may also depend upon other characteristics of the banks, as well as on the country’s
economic and institutional environment.
Willingness is associated with the systemic significance of a bank and the
banking sector in an economy and, hence, with the potential repercussions of bank
fragility. The bigger the systemic significance, the higher the willingness. A bank’s
systemic significance has three dimensions: ‘too big to fail’, ‘too interconnected to
fail’ and ‘too important for other reasons’ (Schich and Lindh [2012]). Their proxies
are expected to be positively related with the willingness to support. The systemic
significance of a country’s banking sector is associated with its size. Willingness may
also depend upon other characteristics of the banks, as well as upon the country’s
economic and institutional environment.
Likewise, moral hazard concerns may be related to both bank-specific and
country-specific variables. The former are widely recognized, with the ‘too big to fail’
and ‘too interconnected to fail’ being the most characteristic ones. For the latter, as
Bonfim and Kim (2012) analyze, banks, in the expectation of support in case of need,
may engage in collective risk-taking strategies which increase systemic fragility. A
big banking sector, which raises the potential cost of bank fragility, or high
concentration, which increases the probability of contagion and –through it— of
generalized financial fragility, may be the outcome of such strategies, i.e., of
collective ‘too big to fail’ and ‘too interconnected to fail’ strategies.
10
The potential explanatory variables are summarized in table 1. The first column
shows the name of the variable and its symbol. The remaining columns show the sign
of their expected effect on DIFF through capacity, willingness and moral hazard
concerns. To economize on space, table 1 does not show the expected sign through
PND. It applies to only one of the potential explanatory variables, anyway; to the
sovereign rating.
Insert Table 1 here
The sovereign credit rating, SOV_RATING, is a proxy of the government’s own
financial condition. The higher the rating, the stronger the condition and the higher
the capacity. In addition, a higher rating is associated with higher probability of the
government not being in default in case of a bank’s need, which again implies a
positive sign.
Private credit to GDP, CREDIT_GDP, a measure of the size of a country’s banking
sector, is related to the size of the potential support. The bigger the size, the lower the
capacity to support, but the higher the systemic significance of the banking sector and,
hence, the higher the willingness. Owing to the potential repercussions of financial
fragility, moral hazard concerns may also be positively correlated with private credit
to GDP – as noted above; hence the negative expected effect on DIFF of
CREDIT_GDP through moral hazard concerns (MH). We also experimented with other
proxies of the size of the banking sector, such as, bank assets to GDP and deposits to
GDP. The results were virtually the same.
BFSR is related to the room for support and to moral hazard concerns. Everything
else equal, a higher BFSR leaves smaller room for support, i.e., how much the rating
can increase. Moral hazard concerns are likely to be stronger for banks with low
BFSR and weaker for banks with high BFSR. Hence, the positive expected effect on
DIFF through MH.
The ratio of a bank’s assets to the country’s GDP, ASSETS_GDP, is a bankspecific proxy of the potential size of support. As such, it is negatively related to
capacity, and positively related to willingness and moral hazard. The signs of the
11
expected effect on DIFF are as in CREDIT_GDP. Ueda and Weder di Mauro (2012)
use a similar measure.
‘Too interconnected to fail’ reflects concerns of contagion, i.e., the spreading of
problems from one bank to other banks of a country. It is proxied with the interbank
ratio, INTERBANK, that is, the ratio of loans to other banks and loans from other banks –
the latter includes central bank funding. The higher this ratio, the stronger the
interconnections and, hence, the lower the capacity to support and the higher the
willingness. Also, a higher ratio is potentially associated with stronger moral hazard
concerns, the justification provided by Bonfim and Kim (2012) – as above.
The proxy of the ‘Too important for other reasons’, the ratio of loans to total
assets, LOANS_ASSETS, is related to the role of banks in mobilizing financial resources
for the financing of the domestic economy. The higher this ratio, the more involved in
financial intermediation a bank is –as opposed, for example, to be more trading
oriented— and, thus, the higher the willingness to support. The results were virtually
the same with customer deposits to total assets.
The capacity to support may also be affected by the accounting standards used by
a bank. More informative accounting standards, which make bank financial ratios
more transparent, reduce the problem of asymmetric information between a bank, on
the one side, and governments, supervisors and markets, on the other. As a result, they
may lead to higher BFSR (Shen et al. [2012]). They may additionally lead to smaller
room for regulatory forbearance and, hence, to smaller capacity for support. More
transparent ratios may also be associated with higher capacity and willingness, and
weaker moral hazard concerns – see the discussion below about the legal and property
rights. The corresponding dummy variable ACC_STD is set to one when a bank follows
IFRS or is a US bank, and zero otherwise. Presumably, IFRS and US GAAP are more
informative than other accounting standards.
A higher ratio of equity over total assets, EQUITY_ASSETS, indicates a bank with a
higher capacity to absorb losses, reducing the need to and, hence, the willingness for
support by the government. Additionally, a higher ratio, which denotes a more
‘conservative’ bank, is likely to be negatively associated with moral hazard concerns,
hence its positive expected effect through MH. Strengthening the above, this ratio has
the benefit of transparency (EBA [2013], Le Leslé Sofiya Avramova [2012]). In
12
addition, its reciprocal, i.e., a simple leverage ratio, is as good a predictor of bank
failure as more complex risk-weighted ratios (Estrella et al. [2000], Drehmann and
Tarashev [2011]).
The last variable provides a powerful example of the dynamic interplay between
banks and governments. A positive sign of DIFF EQUITY_ASSETS
DIFF
EQUITY_ASSETS




W
DIFF
MH
DIFF


W EQUITY_ASSETS MH EQUITY_ASSETS
would be a strong indication that moral hazard concerns are indeed embedded in
market expectations of government support. Perversely, a negative sign, which would
indicate that willingness exerts more influence than moral hazard concerns (the
negative effect before the ‘+’ dominates the positive effect after), would exacerbate
moral hazard as it would induce banks –which, sooner or later, would recognize it—
to take bigger risks.
A better institutional environment, by reducing the possibility of destabilizing
expectations and negative reactions by the markets and the public, may increase both
capacity and willingness. Briefly, in such an environment, the markets and the public,
trusting the institutions of the country, may be more positively inclined towards
support and less negatively inclined against it. In addition, the resolution of bank
distress may be faster and less costly. Moreover, bank distress may be regarded as the
unfortunate, but unavoidable, result of economic forces, such as, unforeseen adverse
macro-economic developments and contagion, instead of the outcome of bad
management of a bank in an economy dominated by special interests.
Moreover, a better institutional environment, provided that it is associated with
more effective supervision and stronger market discipline, may be negatively
associated with moral hazard concerns. Similar arguments justify a positive
relationship of ACC_STD with capacity and willingness, and a negative relationship
with moral hazard concerns.
We proxy the quality of the institutional environment with the index “Legal
Structure and Security of Property Rights” – symbol LEGAL, from the Fraser Institute.
Its key ingredients are rule of law, security of property rights, an independent
13
judiciary, and an impartial court system (Gwartney et al. [2010]). Higher values of
this index denote a better institutional environment, hence the positive signs in table
1.1
Higher bank concentration, CONC, which is related to the structure of a country’s
banking sector, implies higher required support in the case of contagion and, thus,
lower capacity. It also implies higher probability of a systemic crisis in case a bank’s
fragility, which suggests higher willingness to support. Moreover, based on the logic
of
Bonfim and Kim (2012), higher concentration may also be associated with
stronger moral hazard concerns.
We also use GDP per capita, PPP-adjusted at 2005 dollars, as a proxy for the
level of economic development.
3. Data and Econometric Specification
Bank credit ratings and sovereign ratings were collected from Bloomberg. Ratings
availability, summarized in Appendix table A2, determines the sample. Several points
are of interest. To begin with, there are relatively few observations of the local
currency long-term deposit rating – LTDR (column 3) until 2006. BFSR (column 2),
in contrast, is available for a large number of banks since 2000. Taking additionally
into account Moody’s changes in the estimation of the first rating (Moody’s 2007a),
the sample period is chosen 2007 – 2011.
From the two sovereign ratings Moody’s reports, for long-term debt in foreign
and local currency (columns 4 and 5), the latter is more appropriate. Moreover, it has
more observations. In greater detail, the number of banks with the BFSR, LTDR and
long-term debt rating in local currency varies from 486 in 2011 to 571 in 2008
(column 6). Just to have an idea about the differences in the two ratings, the number
of cases with BFSR and LTDR when both sovereign ratings were available and equal
varied from 391 in 2011 to 445 in 2008 (column 7). When both ratings were
1
The conclusions are virtually the same when one more Fraser index is included: “Credit Market
Regulations”, with key ingredients ownership of banks – whether from the government or private
agents, foreign bank competition, private sector credit – the extent to which credit is supplied to the
private sector, and interest rate controls / negative real interest rates. Higher values of this index also
denote a better institutional environment.
14
available, the cases where they were not the same were few and diminishing towards
the end of the sample (column 8).
With this in mind, we use the long-term debt rating in local currency and, when
not available, the long-term debt rating in foreign currency. In this way, the number of
observations rises to 557 in 2011 and 625 in 2007 (column 9). Note that columns (7)
and (8) in table A2 do not add to column (9) because there are cases where only one
of the two sovereign ratings is available. Nevertheless, the results were essentially the
same when only the first of the two –column (6)— was used. In the econometric
analysis, the number of usable observations is further restricted by the availability of
the potential explanatory variables – especially those pertaining to individual banks.
The cross-tabulation of the sovereign rating (local currency long-term debt) and
the BFSR, in panel A of table 2, indicates that the two ratings are clustered along the
diagonal, in two quadrants: (high, high) and (low, low). That is, countries with high
sovereign rating tend to have banks with high stand-alone rating; and vice-versa.
Panel A refers to year 2007, but, as panel B indicates, the picture is broadly similar
for the whole sample period. A notable difference in panel B is that there are
relatively more off-diagonal entries when the sovereign rating is high and BFSR is
low. These are mainly from the years 2008 – 2010, at the height of the global
financial crisis, when governments had probably switched to a crisis-containment
operational mode.
Insert Table 2 here
Note that BFSR is mostly below the sovereign rating, with the few exceptions being
concentrated in the Ba1 – B3 sovereign rating categories. As said above, Moody’s
occasionally assigns a bigger capacity to a government to support its domestic banks
than to service its own debt. Illustrating this possibility, the handful of cases from the
industrial countries pertain to Greece, Ireland, Italy and Spain in 2010 and 2011,
when, at the height of the European sovereign debt crisis, the sovereign ratings were
very low while there was support for the banks from EU institutions. For Greece and
Italy, the expected support was 0 notches; for Ireland and Spain, 1 to 2 notches. For
other countries, such as, Argentina, Brazil, Bulgaria, Chile, Colombia, Croatia, Egypt,
15
Hungary, India, Indonesia, Malaysia, Mauritius, Morocco, Pakistan, Peru, Philippines,
Turkey, Ukraine, Uruguay, Venezuela and Vietnam, the expected support was mostly
0, 1 or 2 notches above the sovereign rating.
The bank-specific variables are retrieved from Bankscope and matched with the
bank credit ratings data. The nominal GDP, in millions of local-currency units, which
is used for the calculation of the ratios of bank assets and bank deposits to GDP,
comes from World Bank’s World Development Indicators. So does the PPP-adjusted
GDP per capita, in constant 2005 US dollars.
From the country-specific variables, private credit to GDP (CREDIT_GDP), bank
deposits to GDP and concentration (CONC) come from World Bank’s Financial
Development & Structure database. Concentration refers to the percent of total bank
assets in a country held by the three biggest banks. The variable pertaining to the legal
system and property rights, LEGAL, comes from the Fraser Institute.
Table 3 reports the pair-wise correlation matrix of the variables used in the
analysis for the year 2007. As it indicates, some of the explanatory variables, namely,
SOV_RATING, BFSR, CREDIT_GDP, LEGAL and log GDP per capita, exhibit relatively
high correlation. However, as mentioned in the empirical analysis section below, to
ease concerns on possible multicollinearity issues and to ensure that this does not
affects our findings, as a robustness check we repeated the analysis in a panel setting
and employed the mean-centered relevant variables. The results were virtually the
same.
Insert Table 3 here
The equation to be estimated is
DIFFi,j,t = α0 + α1 SOV_RATINGj,t + a2 CREDIT_GDPj,t-1
+ β1 BFSRi,j,t + β2 ASSETSi,j,t_GDPj,t + β3 INTERBANKi,j,t +
+ β4 LOANS_ASSETSi,j,t +β5 ACC_STDi,j,t + β6 EQUITY_ASSETSi,j,t
+ γ1 LEGALj,t-1 + γ2 CONCj,t-1 + ui,j,t
(3)
16
where i denotes banks, j countries and t time; ui,j,t is the disturbance term. From the
coefficients to be estimated, the β’s refer to bank-specific variables, while the α’s and
γ’s to country-specific ones. The country-specific variables, with the exception of the
sovereign rating, are lagged once to account for delays in data collection. The results
though –available upon request—were essentially the same with contemporaneous
country-specific variables. The estimation is conducted by year, while the estimation
technique is OLS with robust standard errors.
The ambiguous sign of several variables in table 1 suggests that their effect on
DIFF is likely to be conditional on other observable characteristics of banks and their
countries. This insight will be used in order to deduce the relative significance of
moral hazard concerns.
Consider the coefficients of CREDIT_GDP and ASSETS_GDP, the country-specific
and bank-specific proxies of the ‘too big to fail’. Negative coefficients, α2 < 0 and β2
< 0, would be consistent with –but not unambiguous evidence of— such concerns.
Not unambiguous because the negative sign could reflect lower capacity. In contrast,
positive coefficients would be strong evidence of willingness dominating the
combined effect of capacity and moral hazard concerns. Yet, α2 < 0 and β2 > 0 would
tilt the balance towards capacity (α2 < 0) and willingness (β2 > 0), for moral hazard
concerns are more likely to be reflected on the bank-specific proxy of ‘too big to fail’
than on the country-specific one.
For the coefficient of BFSR, a positive sign would be unambiguous evidence of
the existence of moral hazard concerns. Yet, a negative one would not be
unambiguous evidence that such concerns do not exist; rather that capacity
considerations are relatively stronger.
4. Empirical Analysis
The discussion focuses on the relative significance of moral hazard concerns. It is
assessed in three different ways. The first entails a discussion along the lines in the
previous subsection for the year 2007. The second entails a comparison of the
estimated coefficients for the years 2008 – 2011 with those for 2007. The working
hypothesis is that a structural break might have occurred sometime after 2007. Both
17
ways are summarized in table 4. The third way to assess the relative significance of
moral hazard concerns entails the estimation of equation (3) using the observations for
which moral hazard concerns are likely to be weaker, and the comparison of the
estimated coefficients with those for the whole sample. These are the observations
with high BFSR. The results are summarized in table 5.
Overall, the empirical results indicate that moral hazard concerns were mainly
associated with ASSETS_GDP (positively –which is consistent with the prevailing ‘too
big to fail’ view) and BFSR (negatively). Yet, these concerns were relatively less
significant than capacity and willingness. In addition, the evidence for the presumed
structural break is stronger for the years 2009 – 2011. Moreover, the quality of
institutions was also associated with moral hazard concerns and significant
determinant of expected support.
Main Results
The main results are summarized in table 4. The first column shows the independent
variables, while the remaining ones report the estimated coefficients (and standard
errors in parentheses). There is one column per year.
Insert Table 4 here
Year 2007.
The results for 2007 are summarized in the second column of table 4.
The sovereign rating has a positive and statistically significant –at the 1% level—
coefficient, α1 = 0.280 with standard error 0.040. This is consistent with capacity to
support.
The negative coefficient of lagged private credit to GDP, α2 = -0.017 with
standard error 0.002, is –as discussed above— consistent with moral hazard concerns.
Strictly speaking, it is consistent with the combined effect of capacity and moral
hazard concerns dominating the effect of willingness. However, taking into account
the positive coefficient of ASSETS_GDP (see below), α2 < 0 likely reflects the negative
18
relationship between the size of a country’s banking system and the government’s
capacity to support.
BFSR’s negative coefficient, β1 = -0.194 with standard error 0.046, is consistent
with capacity being more significant than moral hazard concerns.
The positive coefficient of a bank’s assets to GDP, β2 = 0.693, indicates that
willingness exerts more influence than capacity and moral hazard concerns together.
This result is also consistent with Hau et al.’s (2012) finding that the rating agencies
assign higher ratings to bigger banks.2
The accounting standards dummy is also significant. Its coefficient, β6 = -0.666,
significant at the 1% level, is consistent with capacity being negatively correlated
with ACC_STD and exerting more influence than willingness and moral hazard
concerns together.
The positive coefficient of legal and property rights, γ1 = 0.129, significant at the
1% level, suggests that a better institutional environment is associated with higher
expected support. With the available information, one cannot deduce to what extent
this is due to the amelioration of moral hazard concerns in countries with a good
institutional environment. Nevertheless, this result is powerful indication of the
potential beneficial effect of such an environment. To be noted, this effect is enjoyed
by all banks in a country, small and large ones alike, and –most likely— by all private
enterprises as well.
The above findings are quite robust (all the pertinent results are available upon
request). Specifically, we repeated the analysis with Weighted Least Squares instead
of OLS to account for the possible impact of outliers. The results were the same. Also,
to ease any multicollinearity concerns, we mean-centered the variables with high
correlation, i.e., SOV_RATING, BFSR, CREDIT_GDP, LEGAL and log GDP per capita,
and repeated the analysis in a panel setting. The results were essentially unchanged. In
addition, using only the observations for which the local-currency sovereign rating
was available reduced the number of usable observations but resulted in the same
significant variables and virtually the same magnitude of the estimated coefficients.
2
Yet, it is not ex-ante obvious why and how this would affect DIFF. If Moody’s assigned a higher
BFSR and LTDR to bigger banks, DIFF might not be affected or even to be affected negatively.
19
Furthermore, the results were virtually the same when the other proxies mentioned in
the subsection ‘Potential independent variables’ were used, and when the insignificant
variables were eliminated using a general-to-specific modeling approach.
Looking forward, the results for the years 2008 – 2011 (in table 4) and for the
high-BFSR banks (in table 5) are also robust to the same robustness checks.
Years 2008 – 2011.
The working hypothesis, under which the estimated coefficients for the years 2008 –
2011 will be compared to those for 2007, is that a structural break may have happened
after the eruption the global financial crisis. This crisis, precipitated by the collapse of
Lehman Brothers in September 2008, forced governments to switch to a crisiscontainment mode and, hence, caused a structural break in equation (4).
The presumed structural break might entail reduced capacity, increased
willingness and reduced importance of moral hazard concerns,
W
DIFF
DIFF
W
,
, and
 or
.
C
W
MH
MH
Briefly, during a period of generalized financial distress, the capacity to support a
specific bank may be reduced as the governments’ limited resources will have to be
spread to more banks. Pertaining to willingness, while ex-ante, i.e., before the
eruption of the crisis, it might be optimal for the governments to preclude support, expost, i.e., after the eruption, it might be optimal to provide such support to banks in
need; hence the increased willingness. Lastly, facing the prospect of an immediate
financial collapse, governments might ex-post be less concerned about the moral
hazard implications of their support or that the crisis might be the result of moral
hazard; hence, the reduced importance of such concerns.
Reduced capacity implies that variables positively related to capacity will exert
smaller positive influence on expected support. Hence, their coefficients are expected
ceteris paribus to decline. Increased willingness implies that variables positively
related to willingness will exert bigger positive influence on expected support. Hence,
their coefficients are expected ceteris paribus to increase. Reduced moral hazard
concerns implies that variables positively related to such concerns will exert smaller
20
negative influence on expected support. Hence, their coefficients are expected ceteris
paribus to increase. The opposite holds for variables negatively related to C, W and
MH.
To begin with, looked at independently, the results for the years 2008 – 2011
indicate that capacity and willingness were also more significant than moral hazard
concerns. Briefly, the results are virtually the same with those for 2007.
Compared with those for 2007, the results for 2008 – 2011 confirm a structural
break in equation (4) which is consistent with reduced capacity, increased willingness,
and reduced importance of moral hazard concerns. The evidence for the break is
stronger for the years 2009 – 2011. Perhaps more importantly, the effect of increased
willingness is stronger for the variables that are associated with stronger institutions
and market discipline, i.e., LEGAL and ACC_STD. To economize on space, the
discussion focuses on the cases where the estimated coefficients differ by more than
two standard deviations from their values in 2007.
In greater detail, the coefficient of the sovereign rating was within two standard
deviations in 2008 and 2009 from its value in 2007, but lower by more than two
standard deviations in 2010 and 2011 – which is consistent with reduced capacity.
The coefficient of CREDIT_GDP remained negative, with the exception of 2010
when it was insignificant. Its value was essentially the same in 2008 as in 2007, but
lower in absolute value in the remaining years with the difference from 2007
exceeding two standard deviations. This is consistent with reduced capacity, increased
willingness and reduced importance of moral hazard concerns.
The coefficient of BFSR remained negative and within two standard deviations
from its value in 2007, with the exception of 2009 where β1 = -0.299 – which
indicates that in 2009 the effect of reduced moral hazard concerns dominated that of
decreased capacity.The results for ASSETS_GDP point to the same direction. The
progressive decline of the coefficient of ACC_STD, until it became insignificant in
2010, is consistent with the effect of reduced capacity and increased willingness
dominating that of the reduced importance of moral hazard concerns.
The higher coefficient of LEGAL is consistent with increased willingness dominating
the effects of the reduced capacity and the reduced importance of moral hazard
concerns.
21
Sensitivity analysis: High-BFSR banks
To further explore the relative significance of moral hazard concerns, we repeated the
analysis in Table 4 using only the observations for which BFSR exceeds some
threshold, i.e., C- for 2007 and 2008, and D+ for 2009 – 2011, in the original BFSR
scale. The thresholds were chosen so as to include more than half of the observations
in table 4 (see panel C in table 2). The results are summarized in table 5, which has
the same structure as table 4.
The idea is as follows. The banks in the shorter sample have higher stand-alone
ratings. Everything else equal, for these banks moral hazard concerns should be
W MH  or W MH  .
weaker. Technically speaking,
Using this as an
identifying hypothesis, we analyze the evidence in table 5 by comparing year-by-year
the estimated coefficients with those in table 4. The expectation is that variables
positively correlated with moral hazard concerns – and hence negatively correlated
with expected support, will have higher coefficient.
Insert Table 5 here
The results in table 5 are broadly consistent with this hypothesis. The coefficient of
BFSR was negative in all years and lower than in the whole sample, though the
difference exceeded two standard deviations in 2009 – 2011 only. In addition, the
coefficient of LEGAL was lower for all years, becoming insignificant in 2011. The
coefficient of CONC was higher than in the whole sample, yet the difference was
significant only in 2007 and 2008.
The virtually similar coefficients of CREDIT_GDP and ASSETS_GDP indicate that
the size of the banking sector and of individual banks may not have been correlated
with moral hazard concerns; technically speaking, MH Z  0, where Z stands for
the two variables.
The higher coefficients of ACC_STD and EQUITY_ASSETS are not consistent with
the hypothesis, indicating that moral hazard concerns may have not been associated
with these two variables.
22
Moreover, the results by themselves, i.e., without reference to those in table 4, are
qualitatively the same with those in the previous section. That is, capacity and
willingness were relatively more significant than moral hazard concerns, for all
sample years. In addition, the results are largely consistent with the presumed
structural break after 2007.
5. Policy Implications
In this paper, we tried to model the expected government support of banks that is
embedded in Moody’s ratings and –by extension in market expectations of such
support—, staying as close as possible to Moody’s rating process – a process that
rests upon the capacity and willingness of governments to provide support to banks in
need and, as such, takes into account a multitude of factors, both quantitative and
qualitative, bank-specific as well as country-specific. Our approach led us to use
proxies for the size, structure and institutional environment of a country’s banking
sector, which have not been used in the existing literature. Implicit in this approach is
the strategic interaction of banks and governments. The governments, is presumed,
take into account the strategic behavior of banks by considering the moral-hazard
implications of their actions. Banks, on the other hand, decide taking into account all
the parameters of the economic, institutional and regulatory environment – part of
which is the ex-ante expected support.
Though no econometric model can adequately capture the subtleties and
qualitative aspects of the rating process, our results, which are bolstered by extensive
robustness checks, have important policy implications. First, since in Moody’s ratings
moral hazard concerns are relatively less significant than government capacity and
willingness, there is the danger of destabilizing dynamics in cases where the markets
expect support to be forthcoming while the governments hesitate because of such
concerns. Such dynamics may turn a manageable situation into a full-blown crisis.
Second, when quantifying the funding advantage banks derive from the expected
support and pricing a ‘fair’ insurance premium they must pay for, one should take into
account that a substantial part of the expected support can be attributed to the size,
structure and institutional environment of a country’s banking sector, as well as to the
23
creditworthiness of the government. Yet, similar benefits from a good institutional
environment and a creditworthy government are derived by all banks in a country,
large and small alike, as well as by all private enterprises. Pertinent casual evidence is
rife in the financial press (see, for example, Stothard and Atkins [2013]). Raising
more concerns, any insurance premium, especially if it is deemed as excessive, may
tempt banks to assume more risks. Further highlighting the issue of the funding
advantage, Davies and Tracey (2012) find that it may drive the scale economies that
are found in empirical studies for large banks.
Perhaps, more important are the policy implications regarding ways to reduce the
size of the expected support. Forcing banks to downsize does not seem very
promising. Leaving aside the potential side effects of the repressive regulatory and
supervisory actions that would be required to this end, the result would not be certain.
For one thing, the (relative) size of a bank adds comparatively little to the expected
support. In contrast, a result of the downsizing, i.e., lower concentration, by
increasing the capacity of governments to provide support, might increase the
expected support by more.
Further supporting this assessment, banks of all sizes enjoy this support. ‘Too big
to fail’ is not the only determinant, nor does it seem to be the most important. True,
the ‘too big to fail’ issue may weaken market discipline, which has other undesirable
effects. On the other hand, a good institutional environment, which, according to our
results, contributes comparatively more to the expected support, may strengthen
market discipline.
All in all, the most promising way to reduce the size of the expected support and
the pertinent distortions –moral hazard is one of them— seems to be to nudge banks
to improve their risk profiles. In the spirit of this paper, to improve their stand-alone
ratings, the BFSR. To attain this, policy-makers must take a holistic approach as BFSR
evolves endogenously, together with institutions, expected support, economic and
financial development. Indispensable parts of such an approach are better regulation
and more efficient supervision, plus measures to strengthen market discipline, in order
to tackle the problem of excessive risk-taking at its roots. To this end, bailing-in
unsecured bank creditors seems to be in the right direction.
24
References
Barth, James R., Gerald Caprio Jr. and Ross Levine (2004). “Bank regulation and
supervision: What works best?”, Journal of Financial Intermediation, Vol. 13, pp.
205 – 248.
Beck, Thorsten, Asli Demirgüç-Kunt and Ross Levine (2006). “Bank concentration,
competition and crises: First results”, Journal of Banking and Finance, Vol. 30, pp.
1581 – 1603.
Bonfim, Diana and Moshe Kim (2012). “Liquidity risk in banking: Is there herding?”,
mimeo, July.
Dam, Lammertjan and Michael Koetter (2012). “Bank Bailouts and Moral Hazard:
Evidence from Germany”, Review of Financial Studies, Vol. 25, No. 8, pp. 2343 –
2380.
Davies, Richard and Belinda Tracey (2012). “Too big to be efficient? The impact of
implicit funding subsidies on scale economies in banking”, mimeo, June.
Demirgüç-Kunt, Asli and Enrica Detragiache (2002). “Does deposit insurance
increase banking system stability? An empirical investigation”, Journal of
Monetary Economics, Vol. 49, No. 7, pp. 1373 – 1406.
Diamond, Douglas W. and Philip H. Dybvick (1983). “Bank Runs, Deposit Insurance,
and Liquidity”, Journal of Political Economy, Vol. 91, No. 3, pp. 401 – 419.
Diamond, Douglas W. and Raghuram Rajan (2000). “A Theory of Bank Capital”,
Journal of Finance, Vol. 55, No. 6, pp. 2431 – 2465 (February).
Drehmann, Mathias and Nikola Tarashev (2011). “Systemic importance: some simple
indicators”, BIS Quarterly Review, March, pp. 25 – 37.
Estrella, Arturo, Sangkyun Park and Stavros Peristiani (2000). “Capital Ratios as
Predictors of Bank Failure”, Economic Policy Review, Federal Reserve Bank of
New York, No. 6, pp. 33 – 52 (July).
Estrella, Arturo and Sebastian Schich (2011). “Sovereign and Banking Sector Debt:
Interconnections through Guarantees”, Financial Market Trends, Vol. 2011, Issue
2, pp. 21 – 45.
European Banking Authority (2013). “Interim results of the EBA review of the
consistency of risk-weighted assets. Top-down assessment of the banking book”,
http://www.eba.europa.eu/cebs/media/aboutus/News%20and%20Communications/
Interim-results-EBA-review-consistency-RWAs_1.pdf, assessed February 27,
2013.
Gropp, Reint, Hendrik Hakenes and Isabel Schnabel (2011). “Competition, RiskShifting, and Public Bail-out Policies”, Review of Financial Studies, Vol. 24, No.
6, pp. 2084 – 2120 (June).
Gwartney, James, Joshua Hall and Robert Lawson (2010). “Economic Freedom of the
World – 2010 Annual Report”, Fraser Institute.
25
Hau, Haralf, Sam Langfield and David Marques-Ibanez (2012). “Bank Ratings –
What Determines their Quality?”, Working Paper Series – ECB, No. 1484 (2012).
Le Leslé Vanessa and Sofiya Avramova (2012). “Revisiting Risk-Weighted Assets –
Why Do RWAs Differ Across Countries and What Can Be Done About It?”, IMF
Working Paper WP/12/90 (March).
Levine, Ross (2005). “Finance and Growth: Theory & Evidence”, chapter 12 in
Handbook of Economic Growth, edited by Aghion, Philippe and Steven N.
Durlauf, North-Holland, pp. 865 – 934.
Moody’s (2007a). “Bank Financial Strength Ratings: Global Methodology”
(February).
_______ (2007b). “Incorporation of Joint-Default Analysis into Moody’s Bank
Ratings: A Refined Methodology” (March).
_______ (2009). “Financial Crisis More Closely Aligns Bank Credit Risk and
Government Ratings in Non-Aaa Countries” (May).
Morgan, Donald P. and Kevin J. Stiroh (2005). “Too big to fail after all these years”,
Staff Report No. 220, Federal Reserve Bank of New York.
Noss, Joseph and Rhiannon Sowerbutts (2012). “The implicit subsidy of banks”,
Financial Stability Paper No. 15, Bank of England (May).
Packer, Frank and Nikola Tarashev (2011). “Rating methodologies for banks”, BIS
Quarterly Review, June, pp. 39 – 52.
Peresetsky, Anatoly and Alexander Karminsky (2008). “Models for Moody’s bank
ratings”, BOFIT Discussion Papers No 17 – Bank of Finland.
Poon, Winnie P.H., Michael Firth and Hung-Gay Fung (1999). “A multivariate
analysis of the determinants of Moody’s bank financial strength ratings”, Journal
of International Financial Markets, Institutions and Money, Vol. 9, No. 3, pp. 267
– 283 (August).
Schich, Sebastian and Sofia Lindh (2012). “Implicit Guarantees for Bank Debt:
Where Do We Stand?”, Financial Market Trends, Vol. 2012, Issue 1, pp. 1 – 22.
Shen, Chung-Hua, Yu-Li Huang and Iftekhar Hasan (2012). “Asymmetric
benchmarking in bank credit rating”, Bank of Finland Research – Discussion
Paper 13.
Stothard, Michael and Ralph Atkins (2013). “Eurozone core cashes in on cheaper
borrowing”, Financial Times (www.ft.com), February 17, assessed on February 22,
2013.
Ueda, Kenichi and Beatrice Weder di Mauro (2012). “Quantifying Structural Subsidy
Values for Systematically Important Financial Institutions”, IMF Working Paper
WP/12/128 (May).
26
Table 1. Potential Explanatory Variables – Expected Effect on Expected Support
through Capacity, Willingness and Moral Hazard Concerns
Variable
Capacity
Country-specific variables
Sovereign credit rating (SOV_RATING)
+
Size of the banking sector
 Private credit to GDP (CREDIT_GDP)
-
Willingness
Moral Hazard
+
-
Bank-specific variables
-
BFSR
Too big to fail
 Assets of a bank to GDP (ASSETS_GDP)
Too interconnected to fail
 Interbank ratio (INTERBANK)
Too important for other reasons
 Loans to assets (LOANS_ASSETS)
Accounting standards (ACC_STD)
Equity over total assets (EQUITY_ASSETS)
+
-
+
-
-
+
-
-/+
+
+
-
+
+
Economic and institutional environment
Legal system and property rights (LEGAL)
+
+
Bank concentration (CONC)
+
+
-
Notes.
1. The sovereign credit ratings come from Bloomberg, while Private Credit to GDP from the
Financial Development and Structure database, World Bank.
2. The bank-specific variables are calculated with data from BankScope.
3. The Bank Financial Stability Rating, BFSR, is retrieved from BankScope.
4. The variable Accounting Standards take the value 1, ACC_STD = 1, if a bank uses IFRS or
is a US bank; and zero otherwise. The presumption is that IFRS are more informative than
local accounting standards; likewise for US GAAP.
5. The variable Legal System and Property Rights, LEGAL, comes from the Fraser Institute.
6. Bank concentration, CONC, is the share of assets of the three biggest banks in a country. It
comes from the Financial Development and Structure database.
27
Table 2. Sovereign Ratings vs. BFSR
Panel A. Cross-tabulation, Year 2007
Sovereign
rating
BFSR
Aaa
Aa1 - Aa3 A1 - A3 Baa1 - Baa3 Ba1 - Ba3 B1 - B3
Caa1 - C
Aaa
36
124
28
2
Aa1 - Aa3
2
28
36
2
A1 - A3
1
49
43
26
2
12
17
25
1
55
25
37
1
1
69
3
32
22
1
58
147
116
56
3
567
Caa1 - C
Row
Total
4
832
Baa1 - Baa3
Ba1 - Ba3
5
B1 - B3
6
Row
Total
196
68
121
Caa1 - C
Column
Total
39
206
Panel B. Cross-tabulation, Years 2007 – 2011
Sovereign
rating
BFSR
Aaa
Aa1 - Aa3 A1 - A3 Baa1 - Baa3 Ba1 - Ba3 B1 - B3
Aaa
103
409
225
56
35
Aa1 - Aa3
10
157
272
51
7
A1 - A3
1
107
199
87
12
3
409
Baa1 - Baa3
19
114
141
200
24
498
Ba1 - Ba3
11
152
141
15
4
323
4
69
72
8
153
7
7
50
2719
B1 - B3
Caa1 - C
Column
Total
114
703
966
545
341
497
Panel C. Approximate median ratings
Year
2007
Sovereign rating
Rating
% of Obs.
2008
2009
2010
2011
Aa3
A1
Aa3
Aa2
Aa2
A1
46.6
59.3
52.7
50.9
49.2
47.9
2007 – 2011
Aa3
48.9
BFSR
Rating
% of Obs.
C
CCCCCD+
C-
43.2
57.8
56.4
46.4
46.4
44.6
60.9
50.8
28
Notes.
1. For variable definitions and sources, see table 1.
2. BFSR is converted to the regular rating scale for easier comparison with the sovereign
rating (see table A1).
3. In Panel C, the entries in the ‘% of Obs.’ columns indicate the percentage of cases with a
rating equal to or higher than that indicated in the previous column. For example, in 2007,
46.6% of observations had a sovereign rating of Aa3 or above, and 59.3% a rating of A1
or above. Each observation corresponds to a bank-year.
29
Table 3. Pair-wise correlation matrix for the year 2007
DIFF
DIFF
SOV_RATING
CREDIT_GDP lagged
BFSR
ASSETS_GDP
INTERBANK
ACC_STD
EQUITY_ASSETS
LEGAL lagged
CONC lagged
log GDP per capita, lagged
1
0.155
0.000
-0.079
0.028
0.014
-0.086
-0.131
0.173
-0.057
0.104
SOV_
RATING
1
0.781
0.715
-0.073
-0.095
0.240
-0.260
0.747
0.102
0.855
CREDIT_
GDP
lagged
1
0.608
-0.037
-0.025
0.067
-0.268
0.682
0.138
0.591
BFSR
1
-0.010
-0.049
0.105
-0.280
0.588
0.123
0.623
ASSETS_
GDP
1
-0.032
0.017
-0.008
-0.023
0.012
-0.068
INTER
BANK
ACC_
STD
EQUITY_
ASSETS
LEGAL
lagged
CONC
lagged
1
-0.133
0.036
0.016
0.082
-0.063
1
-0.002
0.023
0.125
0.233
1
-0.291
-0.024
-0.172
1
0.058
0.609
1
0.014
Notes.
1. For variable definitions and sources, see table 1.
2. BFSR is converted to the regular rating scale for easier comparison with the sovereign rating (see table A1).
log GDP
per capita,
lagged
1
Table 4. Main Results – Years 2007 – 2011
Constant
2007
(1)
2008
(2)
2009
(3)
2010
(4)
2011
(5)
4.297***
(1.297)
4.872***
(1.363)
4.420***
(1.387)
1.662
(1.387)
1.529
(1.379)
0.146***
(0.038)
-0.003
(0.002)
0.191***
(0.035)
-0.010***
(0.002)
-0.194***
-0.162***
-0.299***
-0.245***
(0.046)
(0.039)
(0.030)
(0.034)
0.693***
0.516***
1.088***
0.022***
(0.175)
(0.157)
(0.228)
(0.004)
0.000
-0.000
-0.001*
-0.000
(0.000)
(0.000)
(0.000)
(0.000)
-0.666***
-0.401**
-0.393***
-0.226
(0.158)
(0.155)
(0.152)
(0.168)
-0.058***
-0.097***
-0.058***
-0.054***
(0.018)
(0.015)
(0.017)
(0.021)
Economic and institutional environment
0.129***
0.378***
0.281***
0.451***
(0.041)
(0.075)
(0.072)
(0.075)
-0.007*
-0.012***
-0.019***
-0.020***
(0.004)
(0.004)
(0.004)
(0.005)
-0.260
-0.372**
-0.309*
-0.016
(0.173)
(0.165)
(0.173)
(0.172)
-0.200***
(0.035)
0.727***
(0.254)
-0.000
(0.000)
-0.088
(0.185)
-0.047***
(0.015)
Country-specific variables
SOV_RATING
CREDIT_GDP
lagged
0.280***
(0.040)
-0.017***
(0.002)
0.223***
(0.042)
-0.019***
(0.002)
0.288***
(0.037)
-0.006***
(0.002)
Bank-specific variables
BFSR
ASSETS_GDP
INTERBANK
ACC_STAND
EQUITY_ASSETS
LEGAL lagged
CONC lagged
log GDP per
capita, lagged
Diagnostics
Adj. R2
# of obs.
0.21
490
0.29
503
0.45
490
0.42
439
0.428***
(0.105)
-0.015***
(0.005)
-0.118
(0.158)
0.39
314
Notes.
1. For variable definitions and sources, see table 1.
2. Standard errors are reported in parentheses. One (*), two (**) and three (***) asterisks
denote significance at, respectively, the 10%, 5% and 1% levels.
3. The dependent variable is the difference between a bank’s long-term deposit rating in local
currency, LTDR, and bank financial stability rating, BFSR, both converted to the same
scale as shown in appendix table A1.
Table 5. Focus on Moral Hazard Concerns – Only High-BFSR Banks
Constant
2007
(1)
2008
(2)
2009
(3)
2010
(4)
2011
(5)
1.444
(2.045)
2.643
(2.084)
3.759**
(1.682)
0.772
(1.724)
1.053
(1.884)
0.195***
(0.051)
-0.004*
(0.002)
0.276***
(0.071)
-0.011***
(0.002)
-0.300***
(0.050)
0.019***
(0.005)
-0.000
(0.001)
-0.178
(0.189)
-0.033
(0.027)
-0.282***
(0.053)
1.094***
(0.311)
0.000
(0.001)
-0.099
(0.209)
-0.029
(0.025)
Country-specific variables
SOV_RATING
CREDIT_GDP
lagged
0.131**
(0.051)
-0.015***
(0.002)
0.096*
(0.056)
-0.016***
(0.002)
0.285***
(0.050)
-0.008***
(0.002)
Bank-specific variables
BFSR
ASSETS_GDP
INTERBANK
ACC_STD
EQUITY_ASSETS
-0.245***
(0.050)
0.556***
(0.151)
0.000
(0.001)
-0.206
(0.154)
-0.024
(0.015)
-0.212***
(0.043)
0.450***
(0.147)
0.000
(0.000)
0.006
(0.153)
-0.088***
(0.021)
-0.371***
(0.041)
1.116***
(0.211)
-0.001**
(0.000)
-0.102
(0.165)
-0.030
(0.022)
Economic and institutional environment
LEGAL lagged
CONC lagged
log GDP per
capita, lagged
Diagnostics
Adj. R2
# of obs.
0.119***
(0.037)
-0.001
(0.004)
0.265
(0.239)
0.261***
(0.076)
-0.005
(0.004)
0.128
(0.253)
0.126*
(0.076)
-0.014***
(0.004)
-0.072
(0.208)
0.224***
(0.085)
-0.013***
(0.005)
0.181
(0.223)
0.094
(0.130)
-0.011**
(0.006)
0.095
(0.229)
0.19
311
0.31
287
0.43
302
0.37
273
0.40
176
Notes.
1. Compared to table 4, the sample includes banks whose BFSR exceeds a limit: BFSR  Cfor 2007 and 2008, and BFSR  D+ for 2009 – 2011. The limits were chosen so that the
sample contains at least half of the banks in table 4.
2. For variable definitions and sources, see table 1.
3. Standard errors are reported in parentheses. One (*), two (**) and three (***) asterisks
denote significance at, respectively, the 10%, 5% and 1% levels.
4. The dependent variable is the difference between a bank’s long-term deposit rating in local
currency, LTDR, and the financial stability rating, BFSR, both converted to the same scale
as shown in appendix table A1.
32
Appendix
Table A1. Mapping BFSR to Standard Credit Ratings
Moody’s Ratings
Standard credit
BFSR
ratings
A
AB+
B
BC+
C
CCD+
D+
D
DE+
E+
E+
E
E
E
E
E
Aaa
Aa1
Aa2
Aa3
A1
A2
A3
Baa1
Baa2
Baa3
Ba1
Ba2
Ba3
B1
B2
B3
Caa1
Caa2
Caa3
Ca
C
Assigned
numerical
values
20
19
18
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
Notes.
1. Source: Moody’s (2007a).
2. BFSR: Bank Financial Stability Rating.
33
Table A2. Rating Availability – Number of Observations
Sovereign Ratings
Long-term Debt
(2), (3)
Foreign
Local
and
Currency Currency
(5)
Availability
(2) - (5)
with
(4)=(5)
(2) - (5)
with
(4)(5)
(2), (3)
with
(4) or (5)
Year
BFSR
LTDR
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
420
446
469
507
548
590
650
695
717
690
691
621
38
43
46
54
69
97
150
661
705
685
687
618
272
297
312
350
391
429
479
507
524
503
516
465
323
353
375
408
466
505
565
595
582
558
544
489
24
29
32
37
57
84
129
567
571
554
541
486
11
13
19
24
18
65
113
429
445
444
433
391
8
11
8
8
32
11
5
18
15
4
2
0
27
30
33
41
60
88
137
601
625
605
619
557
Notes.
1. Sources: Bankscope, Bloomberg and authors’ calculations.
2. BFSR, the stand-alone rating, is Moody’s Bank Financial Stability Rating.
3. LTDR, the all-in rating, is Moody’s local currency Long-term Deposit Rating.
4. The shaded shells identify the sample years.
34
Table A3. Descriptive Statistics – Year 2007
Country
Argentina
Australia
Austria
Bahrain
Belarus
Belgium
Bolivia
Brazil
Bulgaria
Canada
Chile
China
Colombia
Croatia
Cyprus
Czech Repu
Denmark
Egypt
Estonia
Finland
France
Germany
Ghana
Greece
Hong Kong
Hungary
Iceland
India
Indonesia
Ireland
Israel
Italy
Japan
Jordan
Kazakhstan
Korea
Kuwait
Latvia
Lebanon
Lithuania
Luxemburg
Macau
Malaysia
Malta
Mauritius
Mongolia
Morocco
Netherlands
New Zealand
Norway
Oman
Pakistan
Panama
Sovereign
Rating
5
20
20
20
19
10
11
20
16
16
11
13
16
16
20
11
20
20
16
18
15
20
9
8
20
15
18
16
15
15
15
20
14
15
12
10
20
20
20
7
Bank Private
Assets Credit
to GDP to GDP
22.6
12.0
113.6
113.1
124.4
111.6
29.2
107.1
35.1
74.6
54.8
141.5
66.8
109.5
36.4
70.0
201.2
55.6
199.0
67.5
82.1
81.3
115.9
126.6
52.4
103.5
147.6
71.7
146.8
57.8
31.5
186.4
98.4
114.6
153.8
102.9
48.3
99.5
59.3
79.6
19.4
84.2
32.9
36.4
49.7
130.4
65.9
99.7
29.9
57.8
178.1
42.5
190.6
43.0
80.9
77.5
99.4
104.9
47.2
86.3
132.3
58.1
272.8
41.1
22.7
184.8
89.3
95.1
97.4
81.9
46.9
93.1
53.1
77.0
53.9
166.2
51.8
106.6
134.4
90.8
34.0
73.1
184.5
132.3
50.4
162.0
50.9
100.5
111.2
68.7
33.0
60.9
174.5
130.2
34.1
38.4
31.2
27.2
Concentration
38.8
54.8
25.5
56.1
82.6
79.7
61.1
44.3
45.7
51.1
51.9
53.8
48.6
52.4
76.8
59.4
57.0
46.7
92.7
87.7
43.8
37.1
92.1
65.1
71.6
55.3
92.4
32.4
41.6
40.6
76.3
31.2
26.1
78.6
62.0
69.9
53.7
55.5
50.9
72.7
25.8
82.5
37.4
68.5
53.1
69.6
64.2
89.1
90.1
65.9
61.1
39.2
GDP,
$ PPPadjusted
12544.6
33848.1
35834.6
25404.4
10284.5
33538.3
3995.0
9196.4
11249.1
36073.6
13047.0
5238.7
8085.2
16941.6
25851.1
22862.3
34595.3
4955.2
19626.2
33500.8
30554.4
33402.8
5658.1
26258.0
39961.0
17711.3
36875.4
2685.6
3403.4
41024.6
25130.4
28765.6
31659.9
4851.3
10258.6
25021.4
49541.5
16284.1
10170.1
17026.5
74113.9
57158.9
12554.0
22046.2
11024.2
3324.2
3801.7
37576.7
25673.4
48799.7
22495.8
2321.8
13765.8
Legal &
Property
Rights
3.7
9.1
9.0
6.7
Credit
Regulations
7.9
10.0
9.9
10.0
8.3
3.3
4.9
4.9
9.0
6.9
5.8
3.8
7.8
8.8
6.8
9.5
6.3
7.4
9.4
8.1
8.9
8.2
6.7
8.1
6.9
9.1
6.8
4.4
8.9
6.8
6.0
7.7
7.0
9.9
9.0
6.6
10.0
10.0
9.3
7.2
9.0
9.8
10.0
9.8
10.0
6.3
9.7
10.0
9.6
8.3
9.1
8.1
9.7
9.0
9.7
7.2
8.1
9.3
8.3
8.7
9.2
9.3
6.9
8.3
6.9
9.3
9.0
9.7
6.6
9.0
9.5
9.3
6.8
9.0
7.6
9.6
9.5
9.3
7.5
9.2
9.1
9.1
9.0
3.0
8.0
10.0
10.0
10.0
9.6
8.8
35
Paraguay
Peru
Philippine
Poland
Portugal
Qatar
Romania
Russia
Saudi Arabia
Singapore
Slovakia
Slovenia
South Africa
Spain
Sweden
Switzerland
Taiwan
Thailand
Tunisia
Turkey
UK
Ukraine
United Arab Emir.
Uruguay
USA
Venezuela
Vietnam
Mean
Std. Dev.
11
7
15
18
78.4
45.9
55.9
71.4
74.6
67.7
39.6
51.2
82.6
63.2
55.0
71.9
37.8
64.3
83.2
4752.4
7332.6
3303.3
15654.5
21993.3
72813.9
10760.7
14016.2
20242.9
49877.0
19326.7
26323.7
9373.8
28521.6
34782.2
37854.4
22.3
59.4
17.6
76.4
45.4
43.1
43.9
33.4
26.1
43.4
64.5
20.6
33.3
48.9
7249.2
7908.8
12488.2
34115.8
6547.1
57092.0
10783.2
43659.7
11467.8
2481.9
79.8
53.2
57.2
18.4
22256.9
16384.4
19.8
34.9
45.3
159.0
18.3
22.4
34.7
153.4
29.5
34.4
47.8
100.6
52.5
28.0
31.1
36.2
80.1
38.5
80.6
182.1
122.1
176.7
74.4
171.2
113.5
166.0
8
20
7
100.9
57.9
46.3
173.1
46.5
89.6
52.5
26.4
173.1
44.2
7
20
7
8
27.6
65.0
22.2
84.8
15.0
4.6
88.4
49.7
12
16
20
16
18
15
20
20
20
17
13
4.6
4.3
5.9
7.6
8.7
9.2
9.0
7.7
6.1
4.7
8.0
8.0
8.3
6.2
6.9
6.9
7.0
9.3
8.9
6.7
5.8
7.5
5.0
8.4
4.9
7.4
5.7
7.4
1.6
10.0
8.8
9.3
10.0
10.0
10.0
9.3
8.8
9.3
7.0
6.5
9.4
8.2
6.7
7.1
9.2
9.3
10.0
7.0
1.8
9.0
1.0
Note. For variable definitions and sources see table 1.
36