Journal of Banking & Finance 37 (2013) 761772

Contents lists available at SciVerse ScienceDirect

Journal of Banking & Finance

journal homepage: www.elsevier.com/locate/jbf

Bank capital buffer and portfolio risk: The inuence of business cycle
and revenue diversication
Jeungbo Shim
Illinois Wesleyan University, Department of Business Administration, 205 Beecher Street, Bloomington, IL 61701-2900, United States

a r t i c l e

i n f o

Article history:
Received 28 April 2012
Accepted 11 October 2012
Available online 22 October 2012
JEL classication:
G21
G28
Keywords:
Capital buffer
Portfolio risk
Business cycle
Diversication

a b s t r a c t
The relationship between macroeconomic developments and bank capital buffer and portfolio risk
adjustments is relevant to assess the efcacy of newly created countercyclical buffer requirements. Using
the U.S. bank holding company data over the period 1992:Q12011:Q3, we nd a negative relationship
between the business cycle and capital buffer. Our results offer some support for the Basel III agreements
that countercyclical capital buffer in the banking sector is necessary to help the performance of the real
economy during recessions. We nd a robust evidence of inverse relationship between business cycle and
bank default risk. Our analysis provides evidence of diversication benets. The probability of insolvency
risk decreases for diversied banks and banks with high revenue diversity achieve capital savings.
2012 Elsevier B.V. All rights reserved.

1. Introduction
During the recent economic recession, the U.S. banking industry
has experienced a signicant number of bank failures.1 The experiences from bank crises have made regulators, shareholders and bank
themselves more aware of the importance of sufcient capital buffers. Banks retain capital buffers above the regulatory minimum as
a cushion to absorb adverse nancial consequences due to unexpected asset returns.2 Excess capital acts as an insurance against
costs that may occur due to unexpected capital shocks and difculties in raising new capital. A breach of the regulatory minimum capital requirements triggers costly supervisory intervention, possibly
even leading to the rms closure. As a consequence, banks have
an incentive to hold a buffer of excess capital to avoid costs associated with supervisory action if they approach or fall below the regulatory minimum capital ratio (e.g., Marcus, 1984; Furne, 2001).
Banks may maintain excess capital as a signal of soundness to the
market and satisfy the expectations of rating agencies (Jackson

et al., 1999). These market disciplines may lead banks to hold more
capital than required by regulators.
As Basel III agreements on banking supervision are recently endorsed, the management of capital buffers over the business cycle
is increasingly important to reinforce the nancial stability of the
banking systems.3 Under a new regime, banks are required to build
up the extra capital above the regulatory minimum requirement that
can be used in stress. The new rules create a countercyclical capital
buffer within a range of 02.5% of common equity that is designed
to dilute lending bubbles by requiring banks to increase their capital
buffers in cyclical upturns.4 The proximate objectives of the countercyclical capital buffer regime are to constrain loan growth during a
credit boom and to ensure that a sufcient buffer of capital is available to absorb negative capital shocks in downturns (Drehmann
et al., 2010; Francis and Osborne, 2012). In other words, the countercyclical buffer requirement is developed to ensure that the banking
sector in aggregate has a buffer of capital to protect it against future
potential losses and to facilitate the ow of credit in the economy
when the whole nancial system experiences stress after a period
of credit boom.

Tel.: +1 309 556 3308; fax: +1 309 556 1719.

E-mail address: jshim@iwu.edu
According to the failed bank list of the Federal Deposit Insurance Corporation
(FDIC), 361 banks have failed from January 2008 to April 2011, while only 27 banks
failed between October 2000 and December 2007.
2
See Berger et al. (1995) for detailed discussions about why banks should hold
capital.
1

0378-4266/$ - see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jbankn.2012.10.002

3
The agreement on the Basel III reforms for international banking has reached in
September 12, 2010 by the Group of Governors and Heads of Supervision.
4
The capital buffer will be phased in from January 2016 and will be fully effective
in January 2019. The countercyclical capital buffer will be met at 0% before 2016,
0.625% on January 2016, 1.25% on January 2017, 1.875% on January 2018, and 2.5% of
common equity on January 2019.

762

J. Shim / Journal of Banking & Finance 37 (2013) 761772

In this paper, we investigate the question about whether banks

capital buffers behave procyclically or anticyclically over the business cycle. We examine in particular whether capital buffers rise in
business upturns and fall in business downturns (positive comovement) or whether capital buffers exhibit the opposite behavior (negative co-movement) for the U.S. bank holding companies.5
Our evidence on the relationship between macroeconomic developments and banks capital buffer adjustments is relevant to assess the
efcacy of newly created countercyclical buffer requirements.
The capital shock is likely to be driven by materialization of default (i.e., credit) risk in lending, which tends to be associated with
business cycle. In an economic downturn when counterparties are
more likely to be downgraded, the expected credit risk will increase, while the expected credit risk will decrease during an economic upturn. Empirical studies show that credit risk is highly
correlated with changes in the business cycle. Allen and Saunders
(2003) nd that credit quality deteriorates and the probability of
default gets higher during recessions. Curry et al. (2008) show that
the probability of default rises during recession and default risk is
decreasing during periods of recovery and expansion.
It is argued in the literature that, given this anticyclical behavior
(moving in the opposite direction with the business cycle) of credit
risk, banks behavior on their capital buffers is likely to vary
according to the stage of the business cycle and the banks own
nancial situation (e.g., Ayuso et al., 2004). A forward-looking bank
is likely to expand their loan portfolio during periods of economic
upturn. Banks are also expected to build up their capital buffers to
provide protection against the associated credit risk. An underlying
rationale for building up of excess capital during upturns is that
portfolio risks may increase in good times (Crockett, 2001). The
building up of capital defenses may help to moderate excessive
credit growth in periods when economic conditions are buoyant.
These capital buffers would be consumed for credit losses during
severe downturns. If the capital buffers were sufcient to conquer
a downturn, lending activities would not be strictly restricted.
Increasing capital buffer s is actually easier and more cost effective
in booms than in economic troughs. Hence, the capital buffer is expected to behave procyclically for the forward-looking bank.
In contrast, banks may expand their loan portfolios during upturns without increasing their capital buffers accordingly since
some banks tend to underestimate potential risks during periods
of economic expansion when risks are less likely to immediately
materialize. When the economic downturn sets in, raising external
capital is extremely costly and retained earnings as a main source
of building up capital may not be an option due to lower returns.
The inability to raise capital buffer may restrict banks lending
activity. Thus, banks are forced to increase their capital buffers
through a reduction in risk-weighted assets. In this case, the capital
buffer is expected to behave anticyclically with potentially adverse
effects on bank lending during business cycle downturns. The subsequent credit squeeze would contribute to a deeper downturn in
the economy that ultimately undermines the stability of banking
sector, generating a vicious circle.
These countervailing predictions highlight the importance of
empirical studies that would provide evidence on the relationship
between banks building up of capital buffer and business cycle.
Some prior studies have investigated the relationship between
the capital buffers of banks and the rate of GDP growth, although
the evidence is still limited. Ayuso et al. (2004) nd a negative relationship between the business cycle and the capital buffers of
Spanish banks. Jokipii and Milne (2008) show that the cyclical
behavior of European bank capital buffers varies according to size

5
For brevity, we use banks in referring to the bank holding companies in the
remainder of the paper.

and type of bank. Capital buffers rise in recession for large banks,
commercial and savings banks, while capital buffers co-move positively with the business cycle, falling in recession for small banks
and co-operative banks as well as those in accession countries.
These substantial differences among banks draw attention to further research to elucidate the cyclical behavior of banks capital
buffers.
This paper makes several contributions to the literature in the
following aspects. First, as the GrammLeachBliley Act of 1999 allows full afliation of banking with underwriting and agency activities in securities and insurance, U.S. banks have expanded their
business activities from traditional loan making toward a broader
range of nancial services that generate fee income, trading revenue and other types of non-interest income. The increased shifts
toward non-interest activities provide an interesting environment
where we can examine how these diversication choices impact
the changes of banks capital buffer and portfolio risk.
Diversication is an important management strategy for the
banks revenue growth. The motives for diversifying the sources
of bank revenue and its effects have received considerable attention in the literature (e.g., DeYoung and Roland, 2001; Stiroh,
2004; Stiroh and Rumble, 2006). Banks might increase the production and sale of fee-based nancial services to exploit cost scope
economies by sharing inputs in joint production. Banks may take
advantage of revenue scope economies by providing cross-selling
opportunities to customers who are willing to pay for the extra
convenience of nancial supermarkets (e.g., Gallo et al., 1996).
Banks may reduce their capital requirements if increasing the
share of non-interest income in the banks revenue portfolio mitigates overall earnings volatility. Perold (2001) argues that diversication across business segments diminishes the rms
deadweight cost of risk capital. Despite the potential benets associated with diversication, one line of research suggests that there
may be costs arising from corporate diversication. Berger and
Ofek (1995) argue that diversication may lead to inefcient
cross-subsidization across segments because of the agency
problems that allow poor segments to drain resources from
better-performing segments. Managers may seek to diversify in
their own interests, sacricing rm value (Jensen, 1986).
In this paper, we are particularly interested in investigating
whether the increased non-interest income is associated with
changes in the banks capital buffer and risk and whether the revenue diversication contributes to lowering the probability of bank
default and helps banks build up their capital buffer. To our knowledge, this is the rst study that investigates the impact of both revenue diversication and increased non-interest income on the
banks capital buffer and risk adjustments. This is an important
question for bank regulators who should take into account diversication effects in assessing the riskiness of banks portfolio and the
adequacy of capital buffer.
Second, we use a series of pooled cross sectional and time-series
data over the period 19922011, which embraces the most recent
economic downturns that has been one of the hardest-hit periods
for U.S. banks failures. This study provides new empirical evidence
on how banks capital buffers behave throughout the business cycle in response to counter-cyclicality of the recent Basel III capital
legislation for the U.S. bank holding companies.
Third, banks consider risk prole of their portfolios when deciding on the amount of capital buffers. The literature suggests that
the level of banks asset risk will be changed according to the general economic and nancial conditions. More importantly, capital
buffers and risk exposures are simultaneously determined and closely linked to business cycle. Unlike the previous studies that focus
on the cyclical patterns of capital buffer, we incorporate the relationship between banks capital buffer and risk adjustment and
real economy simultaneously in our empirical model.

J. Shim / Journal of Banking & Finance 37 (2013) 761772

763

We build on a partial adjustment framework to examine

whether banks build up their capital buffers during economic upturns to be able to accommodate the materialization of credit risk
in recession or whether banks maintain capital buffers to a certain
level by reducing their risky loans during economic downturns.
The relationship between banks risk taking behavior and the cyclical position will also be examined in this framework. The partial
adjustment model is widely used to examine the dynamic nature
of a rms capital (or debt) adjustments and to estimate the adjustment speed with which the rm moves toward its desired level
(Gilson, 1997; Shyam-Sunder and Myers, 1999; Leary and Roberts,
2005; Flannery and Rangan, 2006; Shim, 2010). The previous studies on the banks capital and risk adjustments are based on the
simultaneous equations (Shrieves and Dahl, 1992; Jacques and
Nigro, 1997; Rime, 2001). Similar to the literature (Ayuso et al.,
2004; Jokipii and Milne, 2011), the form of banks capital buffer
and risk adjustment can be written as:

the variables, DRISKi,t and DBUFi,t are included as the regressor in

buffer and risk equation, respectively to control for simultaneous
relationship between capital buffer and risk adjustments. We investigate whether adjustment costs are relevant by including the
lagged endogenous variable, BUFi,t1 and RISKi,t1, respectively. Because the analysis of the impact of revenue diversication on the
changes of banks capital buffer and portfolio risk is one of our primary interests, revenue diversication variable (DIV) is included in
both equations.
The standard ordinary least squares (OLS) approach may produce biased and inconsistent coefcient estimates in the simultaneous Eqs. (3) and (4) due to the presence of lagged endogenous
variables. To correct for this endogeneity problem, we employ a
three-stage least squares (3SLS) estimation method with instrumental variables which provides consistent estimates of the
parameters. The 3SLS procedure is widely used in the literature
to address the nancial institutions capital and risk decisions in
a simultaneous framework (e.g., Jacques and Nigro, 1997; Rime,
2001; Shim, 2010). Wickens (1982) show that 3SLS estimate is
asymptotically efcient in a full information context. Because the
3SLS takes into account the cross-equation correlations, it yields
asymptotically more efcient parameter estimates than two-stage
least squares (2SLS). The instrumental variables used in 3SLS estimation are described in Section 3.2.

DBUFi;t /BUFi;t  BUFi;t1 ei;t

3. Data and variables

DRISKi;t wRISKi;t  RISKi;t1 gi;t

3.1. The data set

The paper is organized as follows. Section 2 describes the

empirical framework and methodology. Data and variables used
in the analysis are discussed in Section 3. Section 4 presents the results of simultaneous equations and Section 5 concludes.
2. Empirical framework and methodology

where DBUFi,t(BUFi,t  BUFi,t1) and DRISKi,t(RISKi,t  RISKi,t1) represent the observed changes in the capital buffer and risk between
two periods. BUFi,t(RISKi,t) and BUFi;t RISKi;t denote the actual and
the target level of capital buffer (risk) of bank i at time t, respectively. BUFi,t1(RISKi,t1) is the actual level of capital buffer (risk)
in the previous period and ei,t and gi,t are the error terms. The coefcients, / and w measure the adjustment costs (or the speeds of
adjustments). In the absence of adjustment costs, banks would
not hold extra capital than required by the regulators or by the market. In practice, however, adjusting the capital is costly and markets
may be illiquid. Hence, banks may not be able to adjust their desired capital buffers instantaneously in the presence of adjustment
costs. The trade-off between the cost of holding capital and the cost
of failure determines the optimum (or target) capital buffer (Milne
and Whalley, 2001). The partial adjustment Eqs. (1) and (2) indicate
that the observed changes in capital buffer and risk in period t are a
function of the target capital buffer and risk levels, the lagged buffer
and risk levels and exogenous shocks. However, the target capital
buffer BUFi;t and risk RISKi;t levels are not observable (Flannery
and Rangan, 2006). As suggested in the literature, it is assumed that
they depend on the business cycle which tends to be associated
with excess credit growth and bank specic characteristics. Thus,
the empirical model that includes the variables of business cycle
(CYCLE) and bank characteristics is specied as follows:

DBUFi;t /0 /1 CYCLEt /2 DRISKi;t  /3 BUFi;t1 /4 DIVi;t

/0 X i;t qt ei;t

We examine the U.S. bank holding companies rather than the

subsidiary commercial banks because banks strategic decisions
such as capital structure and new activities like brokerage and advisory services are likely made at the level of entire institution. The
primary data source for our analysis is from the SNL DataSource
provided by SNL Financial Corporation. The SNL dataset contains
historical balance sheet and income statement, and provides detailed information about capital adequacy, asset quality and complete sources of revenue for all U.S. bank holding companies and
nancial holding companies regulated by the Federal Reserve System. Similar to Stiroh and Rumble (2006), we focus on the toptiered bank holding companies due to the possibility of multiple
counting of the same activities.6 To maximize the number of observations, our sample is on a quarterly frequency and consists of an
unbalanced panel of U.S. bank holding companies from 1992:Q1 to
2011:Q3. We begin in 1992 since the Basel I international capital
standards were not fully implemented in the U.S. until 1992. In addition, total risk-weighted capital (Tier 1 plus Tier 2) ratios needed for
our analysis were not available in earlier years. The sample is constructed in the following manner. All bank holding companies that
report missing values for dependent variables are excluded. We eliminate bank holding companies that do not have at least twelve continuous quarterly time series observations. We only include
observations for which capital buffer ratios are between the 1st and
99th percentile. Some rms automatically exit from our dataset if
they fail or are acquired by another rm. This procedure results in
a nal sample of approximately 43,000 quarterly observations, which
account for 0.85 of total observations during the sample period.

DRISKi;t w0 w1 CYCLEt w2 DBUFi;t  w3 RISKi;t1 w4 DIVi;t

w0 Y i;t qt gi;t

where CYCLEt is a measure of the business cycle at time t, Xi,t (Yi,t) is

a vector of bank specic control variables for bank i at time t, qt is a
vector of time xed-effect and ei,t (gi,t) is error term. It is well documented that banks manage their capital in accordance with the level of risk they are taking and banks capital and risk choices are
interdependent (e.g., Shrieves and Dahl, 1992; Rime, 2001). Thus,

3.2. Dependent and explanatory variables

Capital buffers: Banks hold more capital than the required regulatory minimum to reduce the likelihood of bank failures. The regulatory minimum capital level is the amount of capital needed for a
6
Stiroh and Rumble (2006) state that one holding company may own another due
to the tiered nature of the U.S. nancial structure.

764

J. Shim / Journal of Banking & Finance 37 (2013) 761772

bank to be regarded as a viable going concern by creditors and

counterparties, while the capital buffer is an additional amount that
banks retain in excess of the regulatory minimum capital requirement to withstand a stressful period. Similar to the literature
(Jokipii and Milne, 2008), we dene banks capital buffer as the
difference between the actual total risk-weighted capital (Tier
1 + Tier 2) ratio and the minimum total required capital ratio of 8%.7
Risk: There is no single denition of risk and no consensus on
how to measure bank risk in the literature. The risk weighted assets to total assets ratio is frequently used to represent the risk
prole of banks assets (Shrieves and Dahl, 1992; Aggarwal and
Jacques, 2001; Rime, 2001). The rationale for using this proxy is
that the allocation of assets across the different risk categories
determines banks portfolio risk. It also assumes that the risk
weights assigned to each category correctly reect the actual credit
risk. However, recent literature argues that the risk weight formula
used for the derivation of supervisory capital charges does not take
into account the varying levels of risk among assets within the
specic portfolio category (Jokipii and Milne, 2011). For instance,
although borrowers level of credit losses is different, the same
weight (100%) is assigned to all commercial loans. To overcome
this problem, we rst employ the ratio of non-performing loans
to total loans and credits as a measure of bank risk.8 The non-performing loans ratio can be considered a good proximate indicator for
asset quality and bank risk since there is clear evidence that the
non-performing loans ratio deteriorates rapidly before the actual
bank failure occurs (Gonzalez-Hermosillo, 1999). Reinhart and
Rogoff (2011) argue that banking problems arise from a prolonged
deterioration in asset quality and the beginning of a banking crisis
can be marked by a large increase in non-performing loans. The
non-performing loans ratio has been commonly used to proxy for
ex post credit risk in the banking literature (Ayuso et al., 2004;
Fiordelisi et al., 2011; Louzis et al., 2012). As a robustness check,
we introduce an additional proxy for risk in Section 4.2.
Cycle: We specify the macroeconomic variable to investigate the
relationship between business cycle and capital buffer and risk
movements. The real GDP growth rate (CYCLE), the most natural
indicator of the aggregate business cycle for an economy, is used
as a proxy for business cycle indicator. We predict that capital
buffers are positively associated with the business cycle if
forward-looking banks build up their capital buffers during
economic upturns. Alternatively, negative relationship is expected
if banks increase their capital buffers through a decrease of the
risk-weighted assets during economic downturns. We expect that
the level of portfolio risk is higher in business downturns, while
it is lower in business upturns given a countercyclical materialization
of credit risk. However, Crockett (2001) argues that portfolio risks
increase in upturns. Thus, the relationship between business cycle
and bank risk is undetermined a prior.
Revenue diversication: To examine whether the revenue diversication is associated with the changes in portfolio risk and capital buffer, the revenue diversication measured by Herndahl
index (HHI) approach is included as an explanatory variable in
both capital buffer and risk equations. To identify the level of revenue diversication, we turn to the income statement and focus on
the sources of total operating revenue. Bank operating revenue is
partitioned into two primary categories: total interest income
and total non-interest income. The sources of total interest income
are composed of seven primary components9: interest and fee in-

7
The international Basel Accord capital guidelines specify that total regulatory
capital (Tier 1 plus Tier 2) must be at least 8% of risk-weighted assets.
8
Non-performing loans are those loans which are delinquent and are placed on a
non-accrual basis. The loan account is considered delinquent when payments of
principal and interest are past due by 90 days or more.
9
This breakdown is based on the classication of the SNL DataSource.

come on loans, income from leases, interest income on balances

due from depository institutions, interest and dividend income on
securities, interest income from assets held in trading accounts,
interest income on federal funds sold, and other interest income.
Interest and fee income on loans includes all interest and
charges associated with real estate loans, commercial and industrial loans, consumer loans, foreign loans, agricultural loans and
other domestic loans. Income from leases includes income related
to lease nancing receivables. Interest income on balances due from
depository institutions includes income received on balances carried with domestic and foreign banks and other depository institutions. Interest and dividend income on securities includes all
interest and dividend income from U.S. Treasury securities, securities issued by states and local government, and other debt and equity securities including investments in mutual funds. Interest
income from assets held in trading accounts includes income
earned on assets that are included in trading assets. Interest income
on federal funds sold includes income on transactions involving the
disposal of immediately available funds for one business day, i.e.,
loans and other extensions of credit resulting from credits to deposit balances at Federal Reserve Banks. Other interest income includes
interest and dividend income received on other assets not specied
elsewhere, i.e., interest income on interest-only strips receivable.
Similar to Stiroh (2004) and Stiroh and Rumble (2006), the
sources of total non-interest income are broken down into four primary components: duciary activities, service charges on deposit
accounts, trading revenue, and other non-interest income. Fiduciary activities generate revenue related to the banks duciary
operations such as managing corporate and individual pension
and retirement plans and corporate security transactions. Service
charges on deposit accounts include revenue related to the maintenance of checking or savings accounts such as fees for writing
excessive checks or overdraft fees. Trading revenue includes income
from trading cash instruments and off-balance sheet derivative
contracts (including commodity contracts). It also includes revaluation adjustments to the carrying value of assets and liabilities due
to periodic marking to market. The sources of other non-interest
income include consulting and advisory services, collection of
utility bills, credit card fees, loan commitment fees and rental of
safe deposit boxes.10 Details on the sources of total interest income
and non-interest income are provided in Appendix A.
Our revenue diversity measure takes into account these eleven
breakdowns of total operating revenue. Thus, the Herndahl index
(HHI) measure of revenue diversication is calculated by the sum
of the squares of income share of a banks operating revenue across
all sources of income (y = 111) for each bank i in each quarter q.

HHIi;q

2
11 
X
Incomei;y;q
y1

Incomei;q

We dene revenue diversity as one minus Herndahl index

(HHIi,q). As a result, banks that operate with only one or a few
sources of revenue (either interest income sources or non-interest
income sources) take a lower value of the revenue diversity. In contrast, banks that offer a wider range of services from traditional
sources of revenue to growing non-interest income or vice versa
have a higher value of the revenue diversity, which indicates that
revenue becomes more diversied. If the increased revenue diversication reduces bank portfolio risk and capital requirements, we
expect this variable to have a negative sign in both capital buffer
and risk equations.
Revenue volatility: We include the volatility of revenue portfolio
to capture the impact of the variation of bank revenue on the cap10
The descriptions of all these components are partly borrowed from the Federal
Reserve Board (www.federalreserve.gov).

J. Shim / Journal of Banking & Finance 37 (2013) 761772

ital buffer and risk adjustments.11 To calculate the volatility for the
banks revenue portfolio, we assume that bank operating revenue
consists of a simple portfolio of two types of assets: one (INT) that
yields total interest income and the other (NON) that produces
non-interest income. Because it is difcult to identify the specic assets associated with each income category to estimate the rate of return, we use the ratio of each income scaled by bank total assets as a
proxy for the rate of return of each asset. Then, the volatility of the
banks revenue portfolio (rREVP) can be expressed as a weighted
sum of covariance of asset INT and asset NON returns12:

rREVP w2 r2INT 1  w2 r2NON 2w1  wCOVrINT ; rNON 1=2

6
where rINT, and rNON represent the standard deviation of returns of
asset INT and asset NON, respectively, w = INT/(INT + NON) is the
total interest income share of bank operating revenue, and
COV(rINT, rNON) is the covariance between the rate of returns of asset
INT (rINT) and asset NON (rNON). The standard deviation of each asset
return and their covariance are computed for each quarter on a
moving average basis over the past twelve quarters. The Eq. (6)
shows that the volatility of the banks revenue portfolio accounts
for the uctuation of the individual asset return and the covariability between the returns of portfolio constituents. The standard portfolio theory implies that the portfolio risk is less than the sum of
individual stand-alone risk if returns of portfolio assets are not perfectly positively correlated. It might be reasonable to reduce capital
buffers for banks with lower volatility of revenues. Thus, we expect
this variable to be positively related to the changes of capital buffer.
We also expect a positive relationship between the volatility of
bank revenue and the changes of portfolio risk.
Non-interest share: To examine how the increased non-interest
income inuences bank capital buffer and risk adjustments, we include the non-interest share of a banks operating revenue in both
equations. The expected sign on this variable is ambiguous a prior.
If non-interest income is more volatile than net interest income,
the increased non-interest income may lead to higher levels of risk
and capital requirement. On the other hand, a negative relationship
between the non-interest share and levels of capital and risk is expected if banks expanded services and cross-selling opportunities
offer the scope of economies or diversication benets.
Bank size: Bank size (SIZE) is included to capture size effects on
capital buffer movement and risk adjustment. Large banks are
likely to hold relatively lower buffer since larger banks tend to
be more diversied and have easier access to the capital markets
than smaller banks. According to the too big to fail hypothesis,
there is a high probability that larger banks will be bailed out by
government assistance in the event of nancial distress. Francis
and Osborne (2012) nd an inverse relationship between capital
ratios and bank size, arguing that larger banks tend to set a smaller
buffer over the regulatory minimum requirement. In contrast,
smaller banks may have an incentive to retain higher capital ratios
since smaller banks are more likely to be liquidated or could be the
target of unfavorable takeovers when they are in nancial distress.
We expect this variable to have a negative impact on the capital
buffer. However, the effect of bank size on risk adjustment is not
determined a prior. The moral hazard issue is likely to occur for
11
We thank a referee for the suggestion to include some proxy for the volatility of
bank operating revenues.
12
Alternatively, we compute the volatility by separating the banks revenue
portfolio into eleven components (seven sources of total interest income plus four
sources of total interest income). Following Stiroh (2004)s approach, we also
calculate the volatility of the banks revenue portfolio using the revenue growth rate
of each component as a proxy for its rate of return. The results for revenue volatility
with these alternative volatility measures are not changed in the regression analysis
that follows.

765

larger banks due to a governments safety net through implicit

too big to fail policies. Thus, larger banks may involve in riskier
lending activities. In contrast, the charter value acts as a restraint
against moral hazard (Keeley, 1990). Larger banks may deter
excessive risk-taking behavior to protect their charter or franchise
value. Bank size is measured by the natural log of total assets
(SIZE). Alternatively, we include indicator variables, LARGE and
SMALL, to detect differences in capital buffer and risk adjustments
according to the size of the institution. LARGE (SMALL) takes the
value one for banks in the highest (lowest) decile in terms of the
size distribution of assets and otherwise takes zero.
Protability: Myers and Majluf (1984) contend that rms prefer
retained earnings rather than external nance as their main source
of nancing investment. As raising capital through the capital markets is costly, banks may rely on the retained earnings to build up
their capital buffers. It is relatively easy to increase capital buffers
when earnings are high. The protability measured as the ratio of
the earnings before interest and tax to total assets is included in
the buffer equation with a positive expected sign.
Liquidity: Banks with a higher level of liquid assets that can
readily be turned into cash when needed indicate that they have
a greater ability to meet short-term nancial obligations without
having to resort to untimely sale of investments or xed assets.
Banks that are required to maintain a certain level of liquid assets
may have less incentive to engage in riskier lending activities. We
use the ratio of liquid assets to total assets as a proxy for liquidity.13 The expected sign on this variable is positive (negative) in buffer (risk) equations.
Loan loss reserve: Loan loss reserve measured by the ratio of loan
loss reserve to gross loans is included in both buffer and risk equations. The loan loss reserve represents the amount that banks think
is adequate to cover estimated losses in the loan portfolios. Thus,
the level of loan loss reserve banks hold may reect the quality
of their loan portfolios. We predict that the loan loss reserve is positively related to the capital buffers because banks with greater expected losses are assumed to boost their capital to comply with
regulatory requirement and to reduce insolvency risk. Aggarwal
and Jacques (1998) argue that the deterioration of asset quality
prevents banks from assuming higher level of risks. As a consequence, a negative relationship is expected between loan loss reserve and bank-risk taking.
Asset growth: The bank specic variables include the asset
growth rates. We expect this variable to be negative in the buffer
equation and positive in the risk equation as rapid asset growth
may increase the banks portfolio risk and impact adversely on capital buffer.
We use several instrument variables in the rst-stage regression to correct for the endogeneity problem. The commonly suggested instrumental variables consist of lagged or historically
averaged measures of rm characteristics, industry growth, and
general economic growth (e.g., Campa and Kedia, 2002). Accordingly, we include average rm size for the prior 3 years, 3-year
average of bank loan growth, rm age and average GDP growth
for the previous 3 years. Finally, year indicator variables are included to control for time effects.
Table 1 presents the breakdown of primary sources of bank
operating revenue. Bank-level data from the SNL DataSource are
aggregated on a quarterly basis. The data reported are for the rst
quarter of selected years from 1990 to 2011 and deated with the
GDP deator. As shown in Table 1, the non-interest income share of
total operating revenue (total interest income plus non-interest income) has steadily increased from 15.24% in 1990:Q1 to 32.06% in

13
Liquid assets include cash and balances due from depository institutions,
securities, federal funds and trading account assets less pledged securities.

766

J. Shim / Journal of Banking & Finance 37 (2013) 761772

Table 1
Composition of bank operating revenue.
1990:Q1

1993:Q1

1996:Q1

1999:Q1

2002:Q1

2005:Q1

2008:Q1

2011:Q1

Percent of total operating revenue

Total interest income
Income on loans
Income from leases
Income on balances due
Income on securities
Income on trading accounts
Income on federal funds
Other interest income
Non-interest Income
Fiduciary income
Service charges on deposit
Trading revenue
Other non-interest income
Total operating revenue

84.76
61.88
0.02
2.67
16.27
0.68
3.20
0.04
15.24
4.28
3.00
0.85
7.10
100

77.37
54.43
1.00
1.27
18.01
0.86
1.73
0.07
22.63
5.74
4.57
2.26
10.05
100

77.83
57.30
1.14
1.13
13.29
2.37
2.60
0.01
22.17
5.09
3.79
1.69
11.60
100

67.94
46.94
1.67
1.01
11.39
2.49
4.23
0.20
32.06
4.06
3.70
4.31
19.99
100

61.16
39.43
1.81
0.76
11.37
4.49
2.90
0.39
38.84
4.05
4.34
3.76
26.69
100

61.57
40.69
1.08
0.89
9.62
4.51
4.35
0.44
38.43
3.85
3.97
4.06
26.54
100

69.14
44.70
0.77
1.33
9.55
7.20
5.09
0.51
30.86
4.20
3.67
0.53
22.47
100

54.37
37.42
0.79
0.61
9.06
4.93
1.15
0.42
45.63
3.40
3.00
9.12
30.12
100

Billions of 2005 Dollars

Total Interest Income
Income on loans
Income from leases
Income on balances due
Income on securities
Income on trading accounts
Income on federal funds
Other interest income
Non-interest income
Fiduciary income
Service charges on deposit
Trading revenue
Other non-interest income
Total operating revenue

19.76
14.43
0.00
0.62
3.79
0.16
0.75
0.01
3.55
1.00
0.70
0.20
1.66
23.31

19.16
13.48
0.25
0.31
4.46
0.21
0.43
0.02
5.60
1.42
1.13
0.56
2.49
24.77

31.67
23.31
0.46
0.46
5.41
0.96
1.06
0.00
9.02
2.07
1.54
0.69
4.72
40.68

61.09
42.21
1.51
0.91
10.24
2.24
3.81
0.18
28.83
3.65
3.33
3.87
17.98
89.92

70.67
45.57
2.09
0.88
13.14
5.18
3.35
0.45
44.89
4.68
5.01
4.34
30.85
115.55

93.50
61.79
1.64
1.35
14.61
6.85
6.60
0.66
58.36
5.85
6.03
6.17
40.30
151.86

138.44
89.50
1.54
2.67
19.11
14.42
10.19
1.01
61.80
8.40
7.35
1.05
44.99
200.23

114.56
78.85
1.66
1.28
19.09
10.39
2.41
0.88
96.14
7.16
6.31
19.21
63.46
210.70

This table presents the breakdown of primary sources of bank operating revenue. Bank-level data from the SNL DataSource are aggregated on a quarterly basis. The data
reported are for the rst quarter of selected years from 1990 to 2011 and deated with the GDP deator. Total operating revenue is dened as total interest income plus noninterest income. Based on the classication of the SNL DataSource, the sources of total interest income are composed of seven primary components: income on loans, income
from leases, income on balances due from depository institutions, income on securities, income from assets held in trading accounts, income on federal funds sold, and other
interest income. Similar to Stiroh (2004) and Stiroh and Rumble (2006), the sources of total non-interest income are broken down into four primary components: duciary
income, service charges on deposit accounts, trading revenue, and other non-interest income.

1999:Q1 to 45.63% in 2011:Q1.14 Among the components of noninterest income, other non-interest income and trading revenue
show the greatest increase during the sample period. Interest and
fee income on loans is the dominant source for total interest income
accounting for 61.88% in 1990:Q1, but its proportion is generally
decreasing over time, down to 37.42% in 2011:Q1. The second largest
share among the sources of total interest income, income on securities, also shows a similar decreasing pattern over the sample period.
Table 1 shows that U.S. banks have shifted away from traditional
interest income towards activities that generate non-interest income
for decades. Table 2 presents the denition of the variables. Table 3
provides summary statistics on the variables of capital buffer, risk,
business cycle, revenue diversication, and rm specic characteristics for all U.S. bank holding companies. The Pearson correlation
coefcients between variables used in our analysis are presented
in Table 4.

4. Estimation results
4.1. First results
To show the reaction of capital buffer and risk to business cycle
uctuations, the changes in capital buffers and bank risk are regressed on real GDP growth rate after controlling for bank-specic
characteristics. We use a three-stage least squares (3SLS) estimation method to account for potential endogeneity between
14
The income composition trend changes in 2008:Q1 due to banks unexpected
asset returns during the nancial crisis.

variables. The results of estimating the simultaneous Eqs. (3) and

(4) are presented in Table 5. Model specication I includes the natural log of total assets to control for rm size, while Model specication II differentiates between large and small banks by dening
large (small) banks as those in the highest (lowest) decile of asset
distribution.
The estimated coefcients of GDP growth rate (CYCLE) are negative and statistically signicant in capital buffer equations, demonstrating that capital buffers uctuate anticyclically over the
business cycle. The result indicates that capital buffers fall in business upturns and rise in business downturns. This nding is in line
with prior empirical studies (Ayuso et al., 2004; Jokipii and Milne,
2008). The increase of capital buffer in downturns can be achieved
either by raising new capital or reducing riskier assets. Banks are
likely to react to capital ratios pressures in the most cost effective
manner. Because external nancing is more costly during economic downturns, banks might be forced to increase their capital
ratios by cutting back loans during cyclical downturns. Jackson
et al. (1999) state that bank capital pressures during cyclical downturns have limited bank lending in those periods, causing a credit
crunch and affecting the real economy. Hyun and Rhee (2011)
show that banks prefer shrinking loans to recover capital ratio
rather than recapitalizing when the economy is in recession.
The coefcients of CYCLE are negative and signicant in risk
equations, showing that bank risk patterns are inuenced by the
macroeconomic environment. The result provides evidence that
loan quality is related to business cycle in an anticyclical manner.
Specically, an expansionary phase of the economy is associated
with relatively lower non-performing loans ratio. During periods
of economic expansion banks tend to underestimate their risk

767

J. Shim / Journal of Banking & Finance 37 (2013) 761772

Table 2
Denition of variables.
Variable

Description

BUF
DBUF
RISK(1)
DRISK(1)
RISK(2)
DRISK(2)
REVENUE DIVERSITY
REVENUE VOLATILITY
NON INTEREST SHARE
SIZE
LARGE
SMALL
ROA
LIQUIDITY
LOAN RESERVE
ASSET GROWTH
CYCLE
AVG OF SIZE
AVG OF LOANGROWTH
AGE
AVG OF GDP GROWTH

Capital buffer: the difference between actual total risk-weighted capital (Tier 1 + Tier 2) ratio and minimum total required capital ratio of 8%
Changes in capital buffer
Ratio of non-performing loans to total loans and credits
Changes in non-performing loans to total loans and credits ratio
Z-score measured by the return on assets (ROA) plus the capital to asset ratio divided by the standard deviation of ROA
Changes in Z-score
Revenue diversication measured by one minus Herndahl index (HHI)
Weighted sum of covariances of returns of the banks revenue portfolio
Non-interest income share of the banks operating revenue
Natural log of total assets
Indicator variable: value 1 for banks in the highest decile of assets
Indicator variable: value 1 for banks in the lowest decile of assets
Return on assets
Ratio of liquid assets to total assets
Ratio of loan loss reserve to gross loans
Rate of growth of total asset
Rate of growth of real GDP
Average rm size for the prior 3 years
Three-year average of bank loan growth
Firm age
Three-year average of GDP growth

Table 3
Summary statistics of regression variables.
Variable
BUF(t1)
DBUF
RISK(1)(t1)
DRISK(1)
RISK(2)(t1)
DRISK(2)
REVENUE DIVERSITY
REVENUE VOLATILITY
NON INTEREST SHARE
SIZE
ROA
LIQUIDITY
LOAN RESERVE
ASSET GROWTH
CYCLE
AVG OF SIZE
AVG OF LOANGROWTH
AGE
AVG OF GDP GROWTH

Mean

Median

Std. dev.

Minimum

Maximum

0.0613
0.0119
0.0138
0.0702
48.7121
0.6652
0.4782
0.0020
0.1611
13.7725
0.0092
0.1762
0.0148
0.0226
0.0055
5.9700
0.0255
26.1378
0.0058

0.0521
0.0200
0.0063
0.0000
36.1037
0.1273
0.4867
0.0011
0.1396
13.4434
0.0103
0.1539
0.0132
0.0163
0.0064
5.8283
0.0211
27.0000
0.0067

0.0382
1.1205
0.0247
0.9237
46.1872
20.6639
0.1204
0.0024
0.1035
1.3225
0.0109
0.1053
0.0072
0.0556
0.0072
0.5737
0.0401
12.4931
0.0040

0.0002
0.7272
0.0001
0.7200
3.3252
215.6827
0.0014
0.0001
0.0001
11.9270
0.1797
0.0072
0.0001
0.7675
0.0230
4.9905
0.2465
1.0000
0.0020

0.9200
0.2122
0.7310
0.7224
297.3562
253.3176
0.8493
0.0616
0.9550
21.5864
0.1980
0.8639
0.2078
1.2603
0.0195
9.3676
0.7209
162.0000
0.0116

This table presents the summary statistics for the variables used in the regression. The data are an unbalanced panel of 43,217 quarterly observations for U.S. bank holding
companies for the period 1992:Q12011:Q3. BUFt1 is the lagged capital buffer. DBUF is measured by changes in capital buffer. RISK(1) is the ratio of non-performing loans to
total loans and credits. RISK(2) is the Z-score measured by the return on assets (ROA) plus the capital to asset ratio divided by the standard deviation of ROA. DRISK(1) and
DRISK(2) are the changes in RISK(1) and RISK(2), respectively. REVENUE DIVERSITY is equal to one minus Herndahl index (HHI). REVENUE VOLATILITY is dened as a
weighted sum of covariances of returns of the banks revenue portfolio. NON INTEREST SHARE is the non-interest income share of the banks operating revenue. SIZE is
measured by the natural log of total assets. ROA is the return on assets. LIQUIDITY is the ratio of liquid assets to total assets. LOAN RESERVE is the ratio of loan loss reserve to
gross loans. ASSET GROWTH is the rate of growth of total assets. CYCLE is the rate of growth of real GDP. AVG OF SIZE is the average rm size for the prior 3 years. AVG OF
LOANGROWTH is the 3-year average of bank loan growth. AGE is the rm age. AVG OF GDP GROWTH is the average GDP growth for the previous 3 years.

exposures and expand lending activity to increase their market

share. Banks often reach this goal by lending to borrowers of lower
credit quality. Thus, when recession sets in, the non-performing
loans ratio increases as borrowers ability to meet obligations
worsens.
The relationship between DBUFi,t and DRISKi,t is not statistically
signicant in both buffer and risk equations. According to the partial adjustment model in Eqs. (1) and (2), the coefcient estimates
of the lagged endogenous variables measure the speed of adjustment. The coefcients of the lagged buffer and risk are 0.147
(0.148) and 0.183 (0.173), respectively and statistically signicant
in Model I (II), indicating that the speed of capital buffer adjustments in banking is relatively slow possibly due to the presence
of adjustment costs.

The coefcients of revenue diversication are negative and signicant at 1% level in both buffer and risk equations, indicating
that diversication benets exist. The result suggests that revenue
diversication is associated with the banks capital buffer savings.
Note that the ratio of non-performing loans to total loans and credits is used to measure bank risk in Table 5. Hence, the revenue
diversication appears to provide effective hedges against risks
from loan portfolio quality. The coefcient on the proxy for revenue volatility is not signicant in buffer equation, but positive
and marginally signicant in risk equation, indicating that banks
with higher revenue volatility may have higher non-performing
loans ratio. The non-interest share has a negative and signicant
impact on changes in capital buffer, while it shows no signicant
relationship with risk movement. The result implies that the

768

J. Shim / Journal of Banking & Finance 37 (2013) 761772

Table 4
Correlation matrix of regression variables.
Variable

10

11

12

13

14

1. BUFt1
2. DBUF

1
0.2025
<.0001
3. RISK(1)t1
0.0404
<.0001
0.0282
4. DRISK(1)
<.0001
5. RISK(2)t1
0.1027
<.0001
6. DRISK(2)
0.0047
0.3248
7. REVENUE DIVERSITY
0.2434
<.0001
8. REVENUE VOLATILITY
0.0066
0.1724
9. NON INTEREST SHARE
0.0322
<.0001
10. SIZE
0.1033
<.0001
11. ROA
0.1135
<.0001
12. LIQUIDITY
0.4271
<.0001
13. LOAN RESERVE
0.1191
<.0001
14. ASSET GROWTH
0.0009
0.8492
15. CYCLE
0.0783
<.0001

0.0418
<.0001
0.0250
<.0001
0.0435
<.0001
0.0563
<.0001
0.0018
0.7164
0.0213
<.0001
0.0230
<.0001
0.0210
<.0001
0.0510
<.0001
0.0025
0.5970
0.0593
<.0001
0.2583
<.0001
0.0539
<.0001

0.0833
<.0001
0.2194
<.0001
0.0135
0.0050
0.0803
<.0001
0.0339
<.0001
0.0889
<.0001
0.0724
<.0001
0.3206
<.0001
0.0083
0.0856
0.4492
<.0001
0.1314
<.0001
0.0773
<.0001

0.0334
<.0001
0.0259
<.0001
0.0418
<.0001
0.0100
0.0377
0.0004
0.9400
0.0221
<.0001
0.1092
<.0001
0.0487
<.0001
0.0642
<.0001
0.0184
0.0001
0.0869
<.0001

0.3648
<.0001
0.0821
<.0001
0.2180
<.0001
0.0561
<.0001
0.0782
<.0001
0.1537
<.0001
0.0296
<.0001
0.1703
<.0001
0.0578
<.0001
0.0473
<.0001

0.0239
<.0001
0.0360 0.1607
<.0001
<.0001
0.0076
0.4557 0.0245
0.1155
<.0001
<.0001
0.0175
0.2820 0.1637
0.3639
0.0003
<.0001
<.0001
<.0001
0.0976
0.1180
0.0349
0.1139 0.0529
<.0001
<.0001
<.0001
<.0001
<.0001
0.0162
0.4400
0.0316
0.1024 0.0275
0.0855
0.0008
<.0001
<.0001
<.0001
<.0001
<.0001
0.0011
0.0619
0.0694
0.1512
0.1666 0.1632 0.0847
0.8165
<.0001
<.0001
<.0001
<.0001
<.0001 <.0001
0.0698 0.0199
0.0856 0.0051
0.0040
0.1242 0.0396 0.1191
<.0001
<.0001
<.0001
0.2851
0.4024
<.0001 <.0001
<.0001
0.0639
0.0651
0.0663 0.0313 0.1179
0.1898 0.1648 0.0078 0.0377
<.0001
<.0001
<.0001
<.0001
<.0001
<.0001 <.0001
0.1057 <.0001

This table presents the correlation matrix between variables used in the regression. BUFt1 is the lagged capital buffer. DBUF is measured by the changes in capital buffer.
RISK(1) is the ratio of non-performing loans to total loans and credits. RISK(2) is the Z-score measured by the return on assets (ROA) plus the capital to asset ratio divided by
the standard deviation of ROA. RISK(1)t1 and RISK(2)t1 are the lagged risk variables. DRISK(1) and DRISK(2) are the changes in RISK(1) and RISK(2), respectively. REVENUE
DIVERSITY is equal to one minus Herndahl index (HHI). REVENUE VOLATILITY is dened as a weighted sum of covariances of returns of the banks revenue portfolio. NON
INTEREST SHARE is the non-interest income share of the banks operating revenue. SIZE is measured by the natural log of total assets. ROA is the return on assets. LIQUIDITY is
the ratio of liquid assets to total assets. LOAN RESERVE is the ratio of loan loss reserve to gross loans. ASSET GROWTH is the rate of growth of total assets. CYCLE is the rate of
growth of real GDP.

increased non-interest shares in bank revenue portfolio may reduce capital buffer requirement if non-interest income and net
interested income are negatively or weakly correlated and their
interactions lower total volatility of bank revenue.
The results of other control variables are generally consistent
with our hypotheses. The estimated coefcient of bank size is negative and signicant in buffer equation of Model I, indicating that
larger banks seem to hold relatively less capital buffers due to
greater diversication, scale economies in risk management and/
or greater support from government in case of nancial difculties.
The signicant and negative sign of LARGE dummy variable in
Model II supports this result. Bank size has a signicant and positive effect on changes in risk in Model I, consistent with the argument of excessive risk-taking behavior by the larger banks under a
governments nancial safety net. Again, the positive sign of LARGE
dummy variable in the risk equation of Model II supports the view
that large banks are more likely to engage in riskier lending activities. The signicant and positive sign of SMALL dummy variable in
the capital equation of Model II indicates that smaller banks tend
to hold more capital buffers that protect them against unexpected
capital shocks since smaller banks are less likely to receive public
subsidies. The protability measured by ROA has a positive impact
on changes in capital buffer in both Model I and Model II, indicating that more protable banks have higher buffers. Because dividend payments tend to be oppressive, it is relatively easy to
increase capital ratios when earnings are high (Berger et al.,
2008). This result supports the capital structure theory that rms
prefer retained earnings as a main source of nancing since external nancing is costly (Myers and Majluf, 1984). The estimated
coefcients of liquidity are signicant and positive in buffer

equation, suggesting that a greater proportion of liquid assets are

positively associated with capital buffers. In the presence of nancial market frictions where external nancing is costly, internal
sources of liquidity might be readily used to build up buffers.
The liquidity ratio is signicant and negative in risk equation,
indicating that banks with a higher degree of liquid assets tend
to have a lower non-performing loans ratio. The loan loss reserve
has a positive (negative) inuence on buffer (risk) adjustment, as
expected if higher levels of loan loss reserve increase incentives
for building up buffers and prevent banks from assuming risky
loans. The coefcients of asset growth are negative and signicant
in buffer equation, showing that a banks asset growth inuences
the change of capital buffer. Since rapid asset growth is likely to increase the proportion of certain risky assets in a banks portfolio,
the risk-weighted capital ratio of banks with high asset growth
may decrease. The coefcients of asset growth are positive and
signicant in risk equation, implying that banks with higher asset
growth rate tend to have higher non-performing loans ratio.
4.2. The analysis of an alternative measure of bank risk
Our primary measure of bank risk is based on the ratio of nonperforming loans to total loans and credits (RISK1). We reexamine
the cyclical variations in capital buffer and risk using a Z-score
(RISK2) as an alternative measure of bank risk. The Z-score is measured by the return on assets (ROA) plus the capital to asset ratio
divided by the standard deviation of ROA. The standard deviation
of return on assets (SDROA) is calculated on a moving average basis
over the preceding twelve quarters. That is, the SDROA in the rst
quarter of 2000 is the standard deviation of a bank quarterly ROA

769

J. Shim / Journal of Banking & Finance 37 (2013) 761772

Table 5
Results for portfolio risk measured by the ratio of non-performing loans to total loans and credits.
Variable

INTERCEPT
CYCLE

DRISK(1)

Model I

Model II

DBUF

DRISK(1)

DBUF

DRISK(1)

1.8525***
(0.4364)
0.0192***
(0.0055)
0.0650
(0.5300)

0.6858***
(0.1479)
0.0115***
(0.0028)

0.8787***
(0.0919)
0.0151***
(0.0051)
0.7227
(0.4746)

0.3132***
(0.0695)
0.0104***
(0.0026)

DBUF
BUFt1

0.0533
(0.0430)
***

0.1476
(0.0493)

REVENUE VOLATILITY
NON INTEREST SHARE
SIZE

0.1482
(0.0205)
***

RISK(1)t1
REVENUE DIVERSITY

0.2681***
(0.1029)
0.0474
(0.0385)
0.1426**
(0.0711)
0.0602***
(0.0208)

0.1832
(0.0448)
0.2209***
(0.0583)
0.0400*
(0.0218)
0.0612
(0.0623)
0.0215***
(0.0060)

LARGE
SMALL
ROA
LIQUIDITY
LOAN RESERVE
ASSET GROWTH
Adjusted R2
Observations

0.0363
(0.0340)
***

0.0545***
(0.0058)
0.0348***
(0.0071)
0.1071**
(0.0545)
5.7030***
(0.2025)
0.133
43217

0.0016***
(0.0005)
0.1879***
(0.0518)
0.2943**
(0.1192)
0.022
43217

0.2877***
(0.1021)
0.0513
(0.0389)
0.1411**
(0.0696)

0.1131***
(0.0396)
0.0827**
(0.0332)
0.0514***
(0.0054)
0.0273***
(0.0031)
0.0298
(0.0370)
5.5622***
(0.1679)
0.134
43217

0.1735***
(0.0360)
0.2137***
(0.0566)
0.0361*
(0.0216)
0.0708
(0.0604)

0.0560***
(0.0197)
0.0164
(0.0181)

0.0016***
(0.0005)
0.1289***
(0.0415)
0.2605**
(0.1073)
0.024
43217

This table presents the regression results for the 3SLS estimation. The dependent variables are DBUF measured by changes in capital buffer, and DRISK(1) calculated by the
changes in non-performing loans to total loans and credits ratio. CYCLE is the rate of growth of real GDP. BUFt1 is the lagged capital buffer and RISK(1)t1 is the lagged risk
variable. REVENUE DIVERSITY is equal to one minus Herndahl index (HHI). REVENUE VOLATILITY is dened as a weighted sum of covariances of returns of the banks
revenue portfolio. NON INTEREST SHARE is the non-interest income share of the banks operating revenue. SIZE is measured by the natural log of total assets. LARGE (SMALL)
takes the value one for banks in the highest (lowest) decile of assets distribution and otherwise zero. ROA is the return on assets. LIQUIDITY is the ratio of liquid assets to total
assets. LOAN RESERVE is the ratio of loan loss reserve to gross loans. ASSET GROWTH is the rate of growth of total assets. The coefcient estimates of year dummies are not
reported here to conserve space. Standard errors are presented in parentheses.
*
Statistical signicant at 10% level.
**
Statistical signicant at 5% level.
***
Statistical signicant at 1% level.

over the 12 quarterly observations between the rst quarter of

1997 and the fourth quarter of 1999. Thus, the SDROA can be
considered a useful indicator of bank total risk because it captures
the total asset return volatility. The total volatility of prots is
costly if large variations in earnings increase the probability of default and banks are forced to raise external capital to ensure that
the likelihood of bankruptcy is low. Hence, the Z-score increases
as banks protability and capital ratio improves, and the Z-score
decreases with increasing volatility of asset returns. Following
the literature, we consider a Z-score as a proxy measure of the likelihood of rm insolvency (Hannan and Hanweck, 1988; Stiroh and
Rumble, 2006; Laeven and Levine, 2009; Barry et al., 2011). Since
the Z-score as a measure of banks distance-to-default is inversely
related to the probability of insolvency, a higher (lower) Z-score
indicates a lower (higher) probability of bank default.
The estimation results using Z-score are presented in Table 6.
The key results are robust to using an alternative bank risk measure. The coefcients of GDP growth rate (CYCLE) are negative
and highly signicant at 1% level in buffer equations. The results
conrm our preceding ndings that capital buffer movements are
negatively associated with business cycle. The coefcients of CYCLE are positive and highly signicant at 1% level in risk equations.
The positive association between GDP growth rate and changes in

Z-score supports the previous ndings in Table 5 and is consistent

with the view that the probability of bank default is increasing during recessions (Allen and Saunders, 2003; Curry et al., 2008). Interestingly, the relationship between DBUFi,t and DRISKi,t is now
positively signicant in both equations, implying that the amount
of capital buffer a bank holds is closely related to the probability of
bank default. This observation seems to support the assumption
that banks with higher proportion of buffers would have lower
probability of default. Thus, higher buffer needs to be held against
riskier portfolios to keep their default risk low.
The robustness checks that use an alternative risk measure
show potential benets of revenue diversication and reinforce
the results of Table 5. Revenue diversication is associated with
lower default risk and banks with diversied revenue portfolio
are likely to hold less capital buffer as the coefcients on revenue
diversity are positive (negative) and statistically signicant in risk
(buffer) equations. The negative and signicant signs of revenue
volatility in risk equations suggest that the probability of bank default may increase as the bank operating revenue becomes more
volatile. The signicant and negative signs of non-interest share
in buffer equations show that there is an inverse relationship between the increased exposure to non-interest activities and bank
capital buffers. The result is consistent with the previous ndings

770

J. Shim / Journal of Banking & Finance 37 (2013) 761772

Table 6
Results for using the Z-score (RISK2) as an alternative measure of bank risk.
Model I

Model II

Variable

DBUF

DRISK(2)

DBUF

DRISK(2)

INTERCEPT

0.9472**
(0.3697)
0.0180***
(0.0032)
0.0464***
(0.0072)

3.2515**
(1.4038)
0.2046***
(0.0454)

0.4219***
(0.0872)
0.0174***
(0.0031)
0.0563***
(0.0061)

3.7722***
(0.8394)
0.1595***
(0.0439)

CYCLE

DRISK(2)

4.2775***
(0.6403)

DBUF
BUFt1

0.0814*
(0.0442)

REVENUE VOLATILITY
NON INTEREST SHARE
SIZE

0.0178
(0.0212)
***

RISK(2)t1
REVENUE DIVERSITY

0.2686***
(0.0778)
0.0472
(0.0292)
0.1723**
(0.0841)
0.0238**
(0.0120)

0.2162
(0.0317)
1.3764**
(0.6979)
1.4232***
(0.3870)
1.3463
(1.0976)
0.8667***
(0.1652)

LARGE
SMALL
ROA
LIQUIDITY
LOAN RESERVE
ASSET GROWTH
Adjusted R2
Observations

3.1662***
(0.6105)

0.0892***
(0.0063)
0.0153**
(0.0068)
0.0437*
(0.0232)
6.4016***
(0.1880)
0.115
43217

0.0175*
(0.0103)
1.9470***
(0.2799)
11.9450***
(2.3700)
0.021
43217

0.2658***
(0.0748)
0.0609
(0.0285)
0.1920**
(0.0897)

0.0749**
(0.0358)
0.0487**
(0.0240)
0.0801***
(0.0059)
0.0264**
(0.0120)
0.0508**
(0.0236)
6.5229***
(0.1768)
0.118
43217

0.0820***
(0.0189)
1.5656**
(0.7487)
1.5126***
(0.3883)
1.3701
(1.0732)

0.5220*
(0.2974)
0.3577
(0.3267)

0.0014
(0.0096)
0.8421***
(0.1887)
15.0079***
(2.2842)
0.022
43217

This table presents the regression results for the 3SLS estimation. The dependent variables are DBUF measured by changes in capital buffer, and DRISK(2) measured by
changes in Z-score. The Z-score is calculated by the return on assets (ROA) plus the capital to asset ratio divided by the standard deviation of ROA. CYCLE is the rate of growth
of real GDP. BUFt1 is the lagged capital buffer and RISK(1)t1 is the lagged risk variable. REVENUE DIVERSITY is equal to one minus Herndahl index (HHI). REVENUE
VOLATILITY is dened as a weighted sum of covariances of returns of the banks revenue portfolio. NON INTEREST SHARE is the non-interest income share of the banks
operating revenue. SIZE is measured by the natural log of total assets. LARGE (SMALL) takes the value one for banks in the highest (lowest) decile of assets distribution and
otherwise zero. ROA is the return on assets. LIQUIDITY is the ratio of liquid assets to total assets. LOAN RESERVE is the ratio of loan loss reserve to gross loans. ASSET GROWTH
is the rate of growth of total assets. The coefcient estimates of year dummies are not reported here to conserve space. Standard errors are presented in parentheses.
*
Statistical signicant at 10% level.
**
Statistical signicant at 5% level.
***
Statistical signicant at 1% level.

of Table 5. However, this variable is not statistically signicant in

risk equations. The results of most other explanatory variables
are generally similar to those presented in Table 5. In sum, the results in Table 6 emphasize a robust connection between business
cycle and capital buffer and risk movements and show a positive
impact of revenue diversication on the banks capital and portfolio risk decisions.

5. Conclusion
The countercyclical capital buffer recently proposed in Basel III
is designed to ensure that the capital requirements in the banking
sector should take account of the macroeconomic environment
where banks operate. The interactions between problems in the
banking sector and the real economy highlight the particular
importance of the banks building up their capital buffer during
business cycle upturns. In downturns when risks materialize,
banks can then rely on these higher buffers to withstand unexpected capital shocks. The establishment of new proposal raises
an important question: How would banks actually adjust their capital buffers and risk over the business cycle? It is difcult to predict
how banks respond to this regulatory change without empirical re-

search. In the present paper, we provide some empirical evidence

on the cyclical behavior of capital buffers and bank risk using an
unbalanced panel of U.S. bank holding companies for the period
1992:Q12011:Q3. Since the literature suggests that bank capital
and risk behavior is simultaneously determined, we use the simultaneous equations model to examine the relationship between
capital buffer and risk movements and business cycle.
The results suggest that the economic cycle plays an important
role in the determination of capital holdings and risk levels. Specifically, we nd a negative relationship between the business cycle
and capital buffer, implying that banks may increase capital buffers
by shrinking their risk-weighted assets during cyclical downturns.
It might be more efcient to improve capital ratios through loan
contraction rather than capital addition since raising capital is extremely costly in recessions. This lending restraint and subsequent
credit squeeze would make the recessions deeper and ultimately
exacerbate the stability of banking systems. Our results might offer
support for the Basel III agreements that countercyclical capital
buffer above the minimum requirement is necessary at individual
banks and the banking sector. This buffer of capital can be used in
recessionary phases and individual banks remain solvent through
this period of recession. Furthermore, the banking sector in aggregate can help the performance of the real economy if the banking

J. Shim / Journal of Banking & Finance 37 (2013) 761772

sector holds sufcient buffers and maintains the ow of credit in

the economy without its solvency being questioned during recessions. We provide evidence that bank risk varies anticyclically with
business cycle. We nd more convincing and robust evidence of inverse relationship between business cycle and bank default risk
when using the Z-score as an alternative measure of bank risk.
We nd evidence that banks correspond their capital buffers to
perceived risk exposures. That is, banks with lower default risk
maintain higher capital buffers and vice versa.
The U.S. banking holding companies are steadily offering a
growing range of nancial services that generate non-interest income. Understanding whether banks can reduce risk and achieve
capital savings from a diversied revenue portfolio is critical information for individual banks as well as regulators for successful
capital and risk management. We examine whether the shift toward non-interest income and the diversied revenue portfolio offer potential diversication benets. Our analysis provides some

771

evidence that there exist the benets of revenue diversication.

For diversied banks that have broader sources of operating revenue, the probability of insolvency risk decreases possibly because
more diversied revenue portfolios are associated with less volatile prots. In addition, banks can attain capital savings by reducing
portfolio risks through revenue diversication. We nd that the increase in non-interest share may reduce capital buffer requirement. We also report that changes in capital buffer and risk are
associated with variations in fundamental bank characteristics
such size, protability, liquidity, loan loss reserve and asset
growth.
Acknowledgements
The author is grateful to an anonymous referee and Ike Mathur,
the editor, for their helpful comments. The author would like to
thank Robert Kearney and David Marvin for useful discussions

Table A
The sources of total interest income. Source: The Federal Reserve Board (www.federalreserve.gov).
Interest and fee income on loans
It includes all interest, fees, and similar charges levied against or associated with all loans. Also included are all yield-related fees on loans held in the banks
portfolio, investigation and service charges, fees representing a reimbursement of loan processing costs, renewal and past-due charges, and fees charged for the
execution of mortgages.
Income from leases
It includes amortized income relating to direct and leveraged nancing leases reportable as lease nancing receivables.
Interest income on balances due from depository institutions
It includes interest income received or accrued during the reporting period on balances carried with domestic and foreign banks and other depository institutions. It
also includes premiums received or discounts paid on foreign exchanges contracts related to nancial swap transactions involving interest bearing balances due
from depository institutions.
Interest and dividend income on securities
Included are interest and dividends on securities held in the banks held-to-maturity and available-for-sale portfolios, even if such securities have been lent, sold
under agreements to repurchase, or pledged as collateral for any purpose. The held-to-maturity securities include U.S. government bonds and notes, securities issued
by federal agencies, state and local government bond, and corporate bonds and notes. The available-for-sale portfolios include holdings of short-term government
securities and privately issued money market securities. Also included is interest received at the sale of securities to the extent that such interest had not already
been accrued on the banks books.
Interest income from assets held trading accounts
It includes the interest income earned on all trading account assets. It also includes accretion of discount on assets held in trading accounts that have been issued on
a discount basis, such as U.S. Treasury bills and commercial paper.
Interest income on federal funds sold
It includes gross revenue from assets of federal funds sold and securities purchased under agreements to resell in domestic ofces of the bank. This item includes
primarily temporary (for 1 day or overnight) loans made to other depository institutions. The funds for these temporary loans often come from the reserves a bank
has on deposit with the Federal Reserve Bank in its district.
Other interest income
It includes income on real estate sales contracts, interest received on other assets not specied elsewhere and interest attributed to transactions not directly
associated with a balance sheet asset, such as the interest attributed to interest rate swaps or to foreign exchange transactions.

Table B
The sources of non-interest income. Source: The Federal Reserve Board (www.federalreserve.gov).
Fiduciary activities
It includes gross income from services rendered by the banks trust department or by any of its consolidated subsidiaries acting in any duciary capacity, such as
commissions and fees on the sale of annuities by these entities.
Service charges on deposit accounts
It includes charges for maintenance of deposit accounts, failure to maintain specied minimum deposit balances, writing excessive checks, withdrawals from
nontransaction accounts, closing of savings accounts, dormant accounts, use of ATM or remote service units, processing of checks with insufcient funds, and other
fees.
Trading revenue
It includes the net gain or loss from trading cash instruments and off-balance sheet derivative contracts (including commodity contracts). It also includes revaluation
adjustments to the carrying value of assets and liabilities resulting from the periodic marking to market, revaluation adjustments from the periodic marking to
market of interest rate, foreign exchange rate, commodity, and equity derivative contracts, and incidental income and expense related to the purchase and sale of
cash instruments reportable in trading assets and liabilities.
Other non-interest income
It includes fees, commissions, and all other service charges that cannot properly be included elsewhere. The sources of other non-interest income include rental of
safe deposit boxes; safekeeping of securities; sale of bank drafts, money orders, and travelers checks; collection of utility bills, redemption of savings bonds;
handling of food stamps; execution of acceptances and issuance of letters of credit; notarizing of forms; consulting and advisory services; credit card fees; charges to
merchants for handling of credit card; data processing services; loan commitment fees; rental fees; interest on tax refunds; and life insurance proceeds on policies. It
also includes net gains (losses) from the sale of branches, transactions of foreign currency, or nonhedging derivative instruments.

772

J. Shim / Journal of Banking & Finance 37 (2013) 761772

and their encouragement. The research support provided by the

Illinois Wesleyan University Faculty Development Committee is
gratefully acknowledged. All errors are my own.
Appendix A
Tables A and B.
References
Aggarwal, R., Jacques, K.T., 1998. Assessing the impact of prompt corrective action
on bank capital and risk. Federal Reserve Bank of New York Economic Policy
Review October, 2332.
Aggarwal, R., Jacques, K.T., 2001. The impact of FDICIA and prompt corrective action
on bank capital and risk: estimates using a simultaneous equations model.
Journal of Banking and Finance 25, 11391160.
Allen, L., Saunders, A., 2003. A Survey of Cyclical Effects in Credit Risk Measurement
Models. Bank for International Settlements Working Paper No. 126 (January).
Ayuso, J., Perez, D., Saurina, J., 2004. Are capital buffers pro-cyclical? Evidence from
Spanish panel data. Journal of Financial Intermediation 13, 249264.
Barry, T.A., Lepetit, L., Tarazi, A., 2011. Ownership structure and risk in publicly held
and privately owned banks. Journal of Banking & Finance 35, 13271340.
Berger, P.G., Ofek, E., 1995. Diversications effect on rm value. Journal of Financial
Economics 37, 3965.
Berger, A.N., Herring, R.J., Szego, G.P., 1995. The role of capital in nancial
institutions. Journal of Banking & Finance 19, 393430.
Berger, A.N., DeYoung, R., Flannery, M., Oztekin, O., 2008. How do large banking
organizations manage their capital ratios? Journal of Financial Services
Research 34, 123149.
Campa, J.M., Kedia, S., 2002. Explaining the diversication discount. Journal of
Finance 57, 17311762.
Crockett, A., 2001. Market discipline and nancial stability. In: Financial Stability
Review, Bank of England, pp. 166173.
Curry, T.J., Fissel, G.S., Hanweck, G.A., 2008. Is there cyclical bias in bank holdings
company risk ratings? Journal of Banking & Finance 32, 12971309.
DeYoung, R., Roland, K.P., 2001. Product mix and earnings volatility at commercial
banks: evidence from a degree of total leverage model. Journal of Financial
Intermediation 10, 5484.
Drehmann, M., Borio, C., Gambacorta, L., Jimnez, G., Trucharte, C., 2010.
Countercyclical Capital Buffers: Exploring Options. Bank for International
Settlements, Working Paper No. 317 (July).
Fiordelisi, F., Marques-Ibanez, D., Molyneux, P., 2011. Efciency and risk in
European banking. Journal of Banking & Finance 35, 13151326.
Flannery, M.J., Rangan, K.P., 2006. Partial adjustment toward target capital
structures. Journal of Financial Economics 79, 469506.
Francis, W.B., Osborne, M., 2012. Capital requirements and bank behavior in the UK:
are there lessons for international capital standards? Journal of Banking &
Finance 36, 803816.
Furne, C., 2001. Bank portfolio allocation: the impact of capital requirements,
regulatory monitoring and economic conditions. Journal of Financial Services
Research 20, 3356.
Gallo, J.G., Apilado, V.P., Kolari, J.W., 1996. Commercial bank mutual fund activities:
implications for bank risk and protability. Journal of Banking and Finance 20,
17751791.

Gilson, S.C., 1997. Transactions costs and capital structure choice: evidence form
nancially distressed rms. Journal of Finance 52, 161196.
Gonzalez-Hermosillo, B., 1999. Determinants of Ex-ante Banking System Distress: A
Macromicro Empirical Exploration of Some Recent Episodes. IMF Working
Papers 33, Washington, DC.
Hannan, T.H., Hanweck, G.A., 1988. Bank insolvency risk and the market for large
certicates of deposit. Journal of Money, Credit and Banking 20, 203211.
Hyun, J., Rhee, B., 2011. Bank capital regulation and credit supply. Journal of
Banking & Finance 35, 323330.
Jackson, P., Furne, C., Groeneveld, C., Hancock, D., Jones, D., Perraudin, W., Radecki,
L., Yoneyama, M., 1999. Capital Requirements and Bank Behaviour: The Impact
of the Basel Accord. Basel Committee on Banking Supervision, Working Paper
No. 1.
Jacques, K., Nigro, P., 1997. Risk-based capital, portfolio risk and bank capital: a
simultaneous equations approach. Journal of Economics and Business 49, 533
547.
Jensen, M., 1986. Agency costs of free cash ow, corporate nance, and takeovers.
The American Economic Review 76, 323329.
Jokipii, T., Milne, A., 2008. The cyclical behaviour of European bank capital buffers.
Journal of Banking & Finance 32, 14401451.
Jokipii, T., Milne, A., 2011. Bank capital buffer and risk adjustment decisions. Journal
of Financial Stability 7, 165178.
Keeley, M., 1990. Deposit insurance, risk and market power in banking. American
Economic Review 80, 11831200.
Laeven, L., Levine, R., 2009. Bank governance, regulation and risk taking. Journal of
Financial Economics 93, 259275.
Leary, M., Roberts, M., 2005. Do rms rebalance their capital structure? Journal of
Finance 60, 25752619.
Louzis, D.P., Vouldis, A.T., Metaxas, V.L., 2012. Macroeconomic and bank-specic
determinants of non-performing loans in Greece: a comparative study of
mortgage, business and consumer loan portfolios. Journal of Banking & Finance
36, 10121027.
Marcus, A., 1984. Deregulation and bank nancial policy. Journal of Banking &
Finance 8, 557565.
Milne, A., Whalley, A.E., 2001. Bank Capital Regulation and Incentives for Risktaking. SSRN.
Myers, S.C., Majluf, N., 1984. Corporate nancing and investment decisions when
rms have information investors do not have. Journal of Financial Economics
13, 187221.
Perold, A., 2001. Capital Allocation in Financial Firms. Harvard Business School.
Working Paper 98-072, Cambridge, MA.
Reinhart, C.M., Rogoff, K.S., 2011. From nancial crash to debt crisis. American
Economic Review 101, 16761706.
Rime, B., 2001. Capital requirements and bank behavior: empirical evidence for
Switzerland. Journal of Banking & Finance 25, 789805.
Shim, J., 2010. Capital-based regulation, portfolio risk and capital determination:
empirical evidence from the U.S. property-liability insurers. Journal of Banking
and Finance 34, 24502461.
Shrieves, R.E., Dahl, D., 1992. The relationship between risk and capital in
commercial banks. Journal of Banking & Finance 16, 439457.
Shyam-Sunder, L., Myers, S.C., 1999. Testing static tradeoff against pecking order
models of capital structure. Journal of Financial Economics 51, 219244.
Stiroh, K.J., 2004. Diversication in banking: is noninterest income the answer?
Journal of Money, Credit and Banking 36, 853882.
Stiroh, K.J., Rumble, A., 2006. The dark side of diversication: the case of US nancial
holding companies. Journal of Banking and Finance 30, 21312161.
Wickens, M.R., 1982. The efcient estimation of econometric models with rational
expectations. Review of Economic Studies 49, 5568.