Professional Documents
Culture Documents
CEO Overconfidence and Financial Crises
CEO Overconfidence and Financial Crises
a r t i c l e i n f o a b s t r a c t
Article history: Over a period that includes the 1998 Russian crisis and 20 07–20 09 financial crisis, banks
Received 13 May 2014 with overconfident chief executive officers (CEOs) were more likely to weaken lending
Revised 5 March 2015
standards and increase leverage than other banks in advance of a crisis, making them more
Accepted 1 April 2015
vulnerable to the shock of the crisis. During crisis years, they generally experienced more
Available online 22 January 2016
increases in loan defaults, greater drops in operating and stock return performance, greater
JEL classification : increases in expected default probability, and higher likelihood of CEO turnover or failure
G01 than other banks. CEO overconfidence thus can explain the cross-sectional heterogeneity in
G21 risk-taking behavior among banks.
G31 © 2016 Elsevier B.V. All rights reserved.
G32
Keywords:
CEO overconfidence
Financial crisis
Bank lending
Bank leverage
Risk culture
http://dx.doi.org/10.1016/j.jfineco.2015.04.007
S0304-405X(16)0 0 022-2/© 2016 Elsevier B.V. All rights reserved.
P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209 195
Because CEOs are the primary influence on bank financ- 100% in the money at least twice during their tenure
ing and investment decisions, CEO attitudes toward bor- as overconfident (from the first time a CEO is seen to
rower prospects could affect their banks’ lending standards postpone exercise). The rationale is that a manager who
and leverage levels, which then can affect a bank’s sensitiv- chooses to keep holding deep-in-the-money stock options
ity to a crisis. We specify that banks with an overconfident after the vesting period is likely to be overconfident about
attitude are banks with overconfident CEOs, who generally the firm’s future prospects.4
think they are better than they are in terms of skill and Our empirical results show that, over 1994–2009, over-
judgment or in gauging the prospects of a successful out- confident banks were more aggressive in lending than non-
come. We thus examine whether CEO overconfidence can overconfident banks in the noncrisis years, particularly in
help explain a bank’s risk taking before a crisis and the lending to real estate borrowers. Overconfident banks on
poor performance of a bank during a crisis.2 average increased their loans (real estate loans) by about
Managerial overconfidence can influence bank risk- 14.98% annually (18.06%), which is 4.60 (11.42) percent-
taking behaviors in several ways. Hirshleifer and Luo age points higher than the increase for non-overconfident
(2001), Malmendier and Tate (2008), and Gervais, Heaton, banks. We also find that overconfident banks showed
and Odean (2011) show that overconfident CEOs overesti- greater growth in leverage in the noncrisis years. For ex-
mate the probability of a positive state and the likelihood ample, the annual rate of change in market leverage for
of returns to be generated from an investment project. overconfident banks was on average about 5.37 percent-
They underestimate the downside risk of a project and age points higher than for non-overconfident banks dur-
tend to choose a project with higher true risks than opti- ing the noncrisis years, again indicating greater risk-taking
mal.3 During an economic upswing, an overconfident CEO behavior.
who is more bullish than others on prospects for the econ- After a crisis developed, our results show that many
omy could relax lending standards and increase bank lever- loans extended by overconfident banks in noncrisis years
age more than other banks while they take exposures be- were in default or near default, creating large capi-
lieved to be the most profitable for current shareholders. tal losses. Such unanticipated losses for overconfident
Yet, by taking greater risk, overconfident CEOs make their banks, accompanied with high leverage, diminished their
banks more vulnerable to an external shock such as a fi- net worth considerably, prompting some depositor with-
nancial crisis. drawals and fire sales and thereby reducing these banks’
We collect CEO overconfidence data from publicly listed net worth still further.5
US banks over 1994–2009, a period that includes the 1998 We find that overconfident banks generally experienced
Russian financial crisis and the most recent worldwide fi- more severe worsening of operating and stock return per-
nancial crisis of 20 07–20 09. Both crises followed a period formance, along with greater increases in expected default
of lending growth (Ivashina and Scharfstein, 2010; Becker probability. Many of these banks replaced their CEOs and
and Ivashina, 2014), and both are among the worst fi- even failed during the crisis years.6 Our results thus indi-
nancial crises in the last 50 years (Fahlenbrach, Prilmeier, cate that overconfidence can lead risk-averse CEOs to take
and Stulz, 2012). Fahlenbrach, Prilmeier, and Stulz (2012) exposures that they perceive are the most profitable for
find that some bank characteristics connoting risk-taking current shareholders ex ante but that could harm their
behaviors are significantly associated with poor perfor- banks and themselves ex post.
mance in both crises. Hence, if overconfidence as a man- We further explore whether a bank CEO’s attitude to-
agerial trait can explain heterogeneity in risk-taking be- ward risk before the 1998 crisis could help predict the
havior among banks, an overconfident bank should have bank CEO’s attitude before the most recent crisis. We find
suffered more than other banks in the two crises. Follow- that, instead of learning from their experience in the 1998
ing Fahlenbrach, Prilmeier, and Stulz (2012), we take 1998 crisis, banks with overconfident CEOs in 1997 significantly
as the start of the Russian crisis and 2007 as the start tended to have overconfident CEOs in 2006. This is consis-
of the most recent financial crisis. We then define 1998 tent with the finding of Fahlenbrach, Prilmeier, and Stulz
and 20 07–20 09 as crisis years and other years as noncri- (2012) that the same firms that suffered significant losses
sis years.
We use a stock options-based proxy for CEO overconfi-
4
dence and construct our measure using Standard & Poor’s The options-based CEO overconfidence measure has become widely
used in recent empirical research (see, e.g., Malmendier and Tate,
ExecuComp database. Following Campbell, Gallmeyer,
20 05, 20 08; Campbell, Gallmeyer, Johnson, Rutherford, and Stanley, 2011;
Johnson, Rutherford, and Stanley (2011), we classify CEOs Malmendier, Tate, and Yan, 2011; Hirshleifer, Low, and Teoh, 2012).
who postpone exercising stock options that are more than Malmendier and Tate (2008) and Hirshleifer, Low, and Teoh (2012) note
several alternative reasons for such behavior, including positive inside in-
formation, signaling, board pressure, risk tolerance, and taxes, that fail to
2 sufficiently explain the delay in exercise behavior among overconfident
For simplicity, we call banks with overconfident CEOs “overconfident
banks” and those whose CEOs are not overconfident “non-overconfident CEOs.
5
banks.” Laeven (2011) shows that, after the global financial crisis started in
3 the summer of 2007, a large number of US banks experienced depositor
Several researchers have investigated the effects of managerial over-
confidence on firm performance. Many of them show that overconfident runs, forced fire sales, reductions in the value of their assets, or increases
CEOs reduce the value of the firm as a result of overinvestment (see, e.g., in the possibility of failure to meet their obligations.
6
Goel and Thakor, 2008; Malmendier and Tate, 2008; Campbell, Gallmeyer, In our sample during crisis years, 25.93% of overconfident banks ex-
Johnson, Rutherford, and Stanley, 2011). On the other side of the coin, perienced CEO turnover, compared with only 15.83% of non-overconfident
Hirshleifer, Low, and Teoh (2012) show that overconfident CEOs can in- banks, and 10.37% of overconfident banks failed, compared with 4.17% of
crease firm value through effective innovation. non-overconfident banks during the period.
196 P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209
in the 1998 crisis then incurred the heaviest losses in the studies by investigating for two financial crises two further
most recent crisis. This result provides evidence to support questions: (1) How does CEO overconfidence affect banks’
persistence in a bank’s risk culture. lending standards, including lending amount and lending
Our work contributes to the literature in three ways. quality, and leverage decisions? (2) How did overconfident
First, it complements research on the kinds of banks that banks that assumed higher risk than other banks perform
perform poorly during a financial crisis by relating man- during crisis years?
agerial overconfidence to bank misfortunes during crisis This paper is organized as follows. We develop our
years. Previous studies have shown that the banks that main hypotheses in Section 2. Section 3 describes our
fared poorly during the most recent financial crisis are data sources and methodology, including the overconfi-
banks with CEOs who had better incentives in terms of the dence measure and the identification strategy. Descriptive
dollar value of their stake (Fahlenbrach and Stulz, 2011), statistics and the primary empirical results are presented
banks with more fragile financing and more shareholder- in Sections 4 and 5, respectively. Section 6 discusses equi-
friendly boards (Beltratti and Stulz, 2012), banks entering librium for our results. Section 7 concludes.
the crisis with lower-quality regulatory capital such as Tier
1 ratios (Berger and Bouwman, 2013), banks with weak-
ening risk management (Ellul and Yerramilli, 2013), and 2. Hypothesis development
banks with greater increases in real estate loans prior to
the crisis (Ma, 2014).7 Palumbo and Parker (2009) and This section provides literature review and presents hy-
Fahlenbrach, Prilmeier, and Stulz (2012) show that banks pothesis development of this study. First, we investigate
that perform poorly in the 1998 crisis relied more heav- how CEO overconfidence affects a bank’s lending standards
ily on short-term funding before the crisis. CEO overconfi- and leverage decisions before a crisis. Second, we examine
dence, prompting greater lending and higher leverage dur- the consequences for overconfident banks during a crisis.
ing the boom preceding a crisis, could have contributed to
banks’ woes during the ensuing crisis years.
Second, our paper complements the risk culture hy- 2.1. Impact of CEO overconfidence on bank lending and
pothesis proposed by Fahlenbrach, Prilmeier, and Stulz leverage
(2012). In examination of a sample of 347 publicly listed
US banks in both the 1998 crisis and the more recent In the psychology literature, the theory on manage-
financial crisis, Fahlenbrach, Prilmeier, and Stulz (2012) rial overconfidence is based on the better-than-average ef-
find that differences in postcrisis stock performance could fect. Psychological studies suggest that overconfident peo-
be attributed to differences in the risk culture of banks. ple tend to overestimate their wisdom or skills beyond an
Ellul and Yerramilli (2013), in a study of 74 US bank average benchmark.8
holding companies, show that banks with an aggressive Overconfident CEOs tend to think they have more pre-
risk culture are associated with weaker risk management cise knowledge about future events than they actually have
functions. In our work, we find that a bank’s risk cul- and that they are more likely to experience favorable fu-
ture reflects the character traits of its CEOs. Aggressive ture outcomes than they actually are. The research shows,
(conservative) banks tend to have an overconfident (non- both empirically and theoretically, that this managerial
overconfident) manager who is willing to take greater character trait has a notable influence on corporate poli-
(less) risk. cies. Overconfident managers are convinced of their abil-
Third, our paper relates managerial character traits to ity to generate higher returns on their investment projects,
risk-taking heterogeneity among banks. Malmendier and often resulting in overinvestment (see, e.g., Malmendier
Tate (20 05, 20 08), Hackbarth (20 08), Malmendier, Tate, and Tate, 20 05, 20 08; Goel and Thakor, 20 08; Campbell,
and Yan (2011), and Graham, Harvey, and Puri (2013) have Gallmeyer, Johnson, Rutherford, and Stanley, 2011; Ben-
drawn attention to how CEOs and their character traits af- David, Graham, and Harvey, 2013).
fect investment policy in industrial firms. Few studies have Extrapolating from the psychology literature and the
looked at the influence of managerial character traits in overinvestment view, we predict that overconfident CEOs
the banking industry. Compared with non-overconfident affect their banks’ lending before a financial crisis in two
CEOs, banks with overconfident CEOs have higher stan- ways. First, when the economy is good, overconfident man-
dard deviations in stock returns (Niu, 2010), experience agers overestimate the probability that a positive state will
greater default risk (Burg, Scheinert, and Streitz, 2012), and continue because they often overestimate the precision
write more real estate loans (Ma, 2014). We extend these of exogenous noisy signals (Malmendier and Tate, 2008;
Gervais, Heaton, and Odean, 2011). Second, overconfident
managers overvalue the prospects of borrowers and under-
7
estimate the riskiness of their investments (Hirshleifer and
Our paper is distinguishable from Ma (2014) in four major respects.
First, we find that overconfident banks can explain the cross section of Luo, 2001). Both of these biased perceptions cause over-
bank risk taking in both the 1998 Russian crisis and the most recent cri- confident banks to place less weight on downside risk,
sis. Second, we provide a possible argument for why banks hire overcon- leading to an easing of lending standards and, in turn, an
fident CEOs to take risk. Third, we further examine the impact of CEO
overconfidence on bank leverage and lending quality. Finally, beyond look-
ing at stock performance, we investigate five consequences of risk taking
8
by overconfident banks during the crisis: nonperforming loans, operating See Larwood and Whittaker (1977), Svenson (1981), and Alicke (1985)
performance, expected default risk, CEO turnover, and bank failures. for theory development and discussion.
P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209 197
increase in lending amounts.9 We thus propose our first usually result in operating performance declines, as non-
hypothesis. performing loans become a capital loss and are clas-
sified as an expense for banks. Then, higher leverage
Hypothesis 1. Overconfident banks have a greater increase in erodes earnings because of high interest payments. These
loans than non-overconfident banks in noncrisis years. two stresses on net earnings produce a slump in stock
market capitalizations of overconfident banks.10 This fur-
CEO overconfidence also appears to have an impact on
ther increases the expected default probability according
corporate financing policy. When overconfident CEOs over-
to models proposed by Merton (1974) and Bharath and
estimate the return on an investment but are unable to
Shumway (2008). Therefore, we infer that overconfident
fully finance the investment with their own capital, they
banks that relax lending standards and assume more lever-
need to borrow external funds and prefer debt to equity
age should experience greater declines in bank perfor-
(Malmendier, Tate, and Yan, 2011). While Adrian and Shin
mance and greater increases in expected default probabil-
(2009, 2010) find that banks like to increase leverage when
ity than non-overconfident banks in crisis years. We thus
they see strong growth opportunities during a credit boom,
propose two hypotheses.
we infer that overconfident banks would have greater in-
creases in leverage in the noncrisis years because these Hypothesis 4. Overconfident banks are likely to experi-
banks perceive business opportunities as stronger than ence greater reduction in bank performance than non-
they really are. Hence, managerial overconfidence, which overconfident banks in crisis years.
leads CEOs to tend toward overinvestment, also could lead
these CEOs to incur additional debt. We thus propose our Hypothesis 5. CEO overconfidence makes their banks prone to
second hypothesis. a greater increase in expected default probability than non-
overconfident banks in crisis years.
Hypothesis 2. Overconfident banks have a greater increase in
leverage than non-overconfident banks in noncrisis years. In the crisis years, many financial institutions replaced
CEOs because of poor performance. Considerable empiri-
cal evidence suggests that the likelihood of CEO turnover
2.2. Consequences for overconfident banks in crisis years
is associated with poor corporate performance (see, e.g.,
Coughlan and Schmidt, 1985; Warner, Watts, and Wruck,
In the period leading up to the recent financial cri-
1988; Parrino, 1997; Farrell and Whidbee, 2003). While
sis, overconfident banks relaxed lending standards and
we infer that banks with overconfident CEOs performed
increased leverage. Assuming greater risk would make
more poorly than banks with non-overconfident CEOs in
these banks more vulnerable to external shocks. Once
crisis years, we further conjecture that these overconfident
an economy weakens, banks with loosened lending stan-
banks should be more likely to replace their CEOs than
dards or increased leverage, or both, often suffer poorer
non-overconfident banks at that time. We thus propose an-
stock performance (Beltratti and Stulz, 2012; Fahlenbrach,
other hypothesis.
Prilmeier, and Stulz, 2012) and experience reduced profits
and greater likelihood of financial distress (Dell’Ariccia and Hypothesis 6. Overconfident banks are more likely to replace
Marquez, 2006). Thus, overconfident banks can be more their CEOs than non-overconfident banks in crisis years.
likely to suffer more unfavorable consequences than non-
overconfident banks after the busting of a credit boom. Evidence also indicates that banks with poorer per-
We compare nonperforming loans, defined as loans formance have more of a chance to fail to meet obliga-
with payments of interest and principal more than 90 days tions and thus have a greater likelihood of failure (see,
overdue, made by overconfident and non-overconfident e.g., Cole and Gunther, 1995; Wheelock and Wilson, 20 0 0;
banks. Relaxation of lending standards by overconfident Dell’Ariccia and Marquez, 2006; Männasoo and Mayes,
CEOs weakens loan quality, and shaky loans are more vul- 2009; Cole and White, 2012). We thus infer that, after
nerable to the influence of macroeconomic shocks. Hence, the credit boom burst, overconfident banks, whose perfor-
when the economy goes into a downturn, overconfident mance suffered more, faced a greater likelihood of bank
banks can face more loan defaults than non-overconfident failure than non-overconfident banks. We propose the fi-
banks. We thus propose our third hypothesis. nal hypothesis.
Hypothesis 3. Overconfident banks suffer more of an increase Hypothesis 7. Overconfident banks are more likely to fail than
in nonperforming loans than non-overconfident banks in crisis non-overconfident banks in crisis years.
years.
3. Data and methodology
Overconfident banks can also perform more poorly
themselves in crisis years. Masses of nonperforming loans
This section provides information on our data sources,
defines the overconfidence measure, and describes the
9
The literature indicates that underwriting criteria were loosened prior
identification strategy.
to the onset of both crises (see, e.g., Mian and Sufi, 2011; Huizinga and
Laeven, 2012; Becker and Ivashina, 2014; and Rajan, Seru, and Vig, 2015).
10
Ivashina and Scharfstein (2010) and Becker and Ivashina (2014) show For example, over 2007 through 2009, the stock market capitalization
lending growth in both periods. We infer that CEO overconfidence can of many major banks dropped by more than half, badly damaging the
account for such lending standard distortions. market value of bank assets.
198 P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209
3.1. Data low overconfident CEO exercises stock options that are less
than 30% in the money and does not maintain any ex-
We start with all companies with standard indus- ercisable options that are more than 30% in the money.
trial classification (SIC) codes between 60 0 0 and 6300 We require that CEOs exhibit this conservative option ex-
that intersect in the Standard & Poor’s Compustat and ercise behavior at least twice during their tenure and as-
ExecuComp databases. Following Fahlenbrach, Prilmeier, sign them to the low overconfident category when they
and Stulz (2012), we exclude firms that are not in the first exhibit this behavior. CEOs who hold or exercise op-
traditional banking industry, such as online brokerages, tions with moneyness between 30% and 100% are classified
payment processors, or investment advisors (SIC 6282). as moderately overconfident. Both Goel and Thakor (2008)
Our sample includes both depository institutions (com- and Campbell, Gallmeyer, Johnson, Rutherford, and Stan-
mercial banks and savings institutions) and investment ley (2011) have theoretically and empirically found that
banks. Given the availability of data from the ExecuComp only highly overconfident CEOs can exhibit overinvestment
database, we choose 1993 as the first year of the sample. behavior. Therefore, we characterize highly overconfident
As we need to calculate the rate of change in our main CEOs as our overconfident group. The low and moderately
variables at year t based on the data at year t – 1, our overconfident CEOs are defined as the non-overconfident
sample is further restricted to include firms that have at group.
least two years of data. Thus, the effective results period We compute option moneyness as follows. Realizable
begins with 1994. As the literature on the most recent fi- values per option are estimated from the total realizable
nancial crisis usually defines it as the period from the 2008 value of exercisable options divided by the number of
liquidation of Lehman Brothers to 2009 (Laeven and Valen- exercisable options. Then, the estimated average exercise
cia, 2013), we end our sample in 2009.11 The final sample prices of the options are computed from the fiscal year-
contains 1,643 bank-year observations from 1994 through end stock price minus the realizable value per option.
2009. Hence, the percentages of average moneyness are obtained
We obtain accounting data for depository institutions from per-option realizable value divided by the estimated
from Compustat Bank, accounting data for investment average exercise price. We employ a similar methodol-
banks from Compustat, and stock returns from the Cen- ogy to measure the percentage of moneyness of exercised
ter for Research in Security Prices (CRSP). To prevent options.
outliers from biasing the results, we winsorize account-
ing data at 1% and 99% in all the analyses. The ex-
pected default frequency is estimated using the CRSP and 3.3. Identification
Compustat databases. CEO turnover is obtained from the
ExecuComp database. We also collect information on bank The process for deducing the effect of CEO overconfi-
failures from the Federal Deposit Insurance Corporation dence on bank lending and leverage can suffer from en-
(FDIC) database. dogeneity problems for three reasons: simultaneity, omit-
ted variables, or measurement errors. First, simultaneity or
reverse causality can confound the results if CEOs whose
3.2. Overconfidence measure past success is built on aggressive lending and high lever-
age tend to be more overconfident. We address this simul-
We use a stock options-based proxy for CEO over- taneity bias by estimating a two-stage least squares (2SLS)
confidence, constructing the measure from Standard & procedure (see, e.g., Griliches and Hausman, 1986; Berger
Poor’s ExecuComp database between 1992 and 2009. We and Hannan, 1998; Biorn, 20 0 0) in Section 5.1.2.
adopt criteria for the CEO overconfidence indicator from Second, unobservable variables that are common to
Campbell, Gallmeyer, Johnson, Rutherford, and Stanley banks can also generate a positive relation between CEO
(2011). This is a modified version of the stock options- overconfidence and bank lending and leverage (the omit-
based overconfidence measure from Malmendier and Tate ted variable bias). If CEO overconfidence is correlated with
(20 05, 20 08), in which overconfident CEOs are those who a variable that has been omitted from the analysis but
delay in the exercise of deep in-the-money exercisable op- that determines, in part, bank lending and leverage, the
tions. regression estimator of CEO overconfidence will be biased
Following the classification of overconfident CEOs in and inconsistent. We eliminate this type of bias from our
Campbell, Gallmeyer, Johnson, Rutherford, and Stanley analysis because both bank and year fixed effects control
(2011), we define three levels of managerial overconfi- for unobservable omitted variables (Roberts and Whited,
dence: high, moderate, and low overconfidence, using 100% 2013).
and 30% moneyness as the cutoff points. To identify highly Third, measurement error in the CEO overconfidence
overconfident CEOs, we require that CEOs have postponed variable can influence the effect of CEO overconfidence
exercise of 100% in-the-money options at least twice dur- on bank lending and leverage. When the CEO overconfi-
ing their tenure. CEOs are assigned to the highly overcon- dence variable of an empirical model is mismeasured, coef-
fident category when they first exhibit this behavior. The ficients estimated via regression are inconsistent. The 2SLS
method also controls for the possibility that measurement
errors can contaminate our results (see, e.g., Rajgopal and
11
Didier, Hevia, and Schmukler (2012) indicate that the recovery phase Shevlin, 20 02; Biddle and Hilary, 20 06; Chen, Kacperczyk,
started in 2010. and Ortiz-Molina, 2011).
P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209 199
Table 1
Summary statistics.
This table presents summary statistics for all the variables used in this study. Over 1994–2009, we define 1998 and
20 07–20 09 as crisis years and others as noncrisis years. OCnoncrisis is a dummy variable that equals one if a bank is an
overconfident bank in the noncrisis years and zero otherwise. OCpre is a dummy variable that equals one if a bank is an
overconfident bank in the year prior to the crisis years and zero otherwise. Variable definitions are in Table A1.
OC measure
OCnoncrisis 0.4697 0.4993 0.0 0 0 0 0.0 0 0 0 1.0 0 0 0
OCpre 0.6157 0.4864 0.0 0 0 0 1.0 0 0 0 1.0 0 0 0
Bank risk-taking measure in noncrisis years
Loans 0.1259 0.1681 0.0341 0.0979 0.1773
Real estate loans 0.1028 0.2474 0.0202 0.0950 0.2063
Book leverage −0.0025 0.1135 −0.0658 −0.0087 0.0534
Market leverage −0.0196 0.2223 −0.1734 −0.0469 0.1030
Regulatory leverage 0.0102 0.1023 −0.0505 0.0060 0.0637
Bank consequence measure in crisis years
NPL/Loancrisis –NPL/Loanpre (%) 1.0276 1.8915 0.0 0 02 0.5706 1.3744
NPL/Equitycrisis –NPL/Equitypre (%) 16.1376 36.3015 −0.0 0 07 2.8030 15.6832
ROEcrisis –ROEpre (%) −10.5029 26.6091 −13.1349 −2.3327 0.8488
Crisis return −0.3241 0.2530 −0.4401 −0.2725 −0.1682
EDFcrisis –EDFpre 0.2954 0.3591 0.0 0 0 0 0.1058 0.5432
TURNOVER 0.2069 0.4059 0.0 0 0 0 0.0 0 0 0 0.0 0 0 0
FAILURE 0.0728 0.2603 0.0 0 0 0 0.0 0 0 0 0.0 0 0 0
Bank characteristics in noncrisis years
Assets 9.8680 1.5533 8.7040 9.6692 10.8672
Book leverage 13.2908 6.1088 10.5197 12.4982 14.6126
ROA (%)∗∗ 1.4187 1.7177 0.9513 1.2078 1.5233
LOANDEP 0.9020 0.2519 0.7869 0.9184 1.0394
Tier 1 (%) 9.7092 2.5912 7.8200 9.2900 11.1350
Bank characteristics in 1997 and 2006
Assets 9.7056 1.6420 8.5643 9.3755 10.8327
Book leverage 13.4633 7.6291 9.6250 11.9898 14.5303
ROA (%) 0.6391 2.0815 0.0733 0.8030 1.2708
LOANDEP 0.9218 0.2280 0.8459 0.9412 1.0342
Tier 1 (%) 10.5759 2.9450 8.50 0 0 10.50 0 0 12.0900
4. Summary statistics and univariate analyses Prilmeier, and Stulz, 2012). We follow Beltratti and Stulz
(2012) and Erel, Nadauld, and Stulz (2014) and construct
In this section, we present the results of descriptive three leverage measures: book, market, and regulatory. The
statistics for our data and the univariate results in which average annual rates of change in book leverage, market
we compare the principle variables of overconfident banks leverage, and regulatory leverage are −0.25%, −1.96%, and
with those of non-overconfident banks using our sample. 1.02%, respectively, in the noncrisis years.
To study how managerial overconfidence affects banks’
4.1. Summary statistics sensitivity to the exogenous shocks in both crises, we focus
on bank loan defaults, bank performance, bank insolvency,
In Table 1, we report summary statistics for the data. CEO turnover, and bank failure. For convenience, we define
The definitions of all variables are shown in Table A1. the amount of change in variable X from noncrisis to cri-
Overconfident banks represent 46.97% of the sample in sis years as the difference between the mean of variable
the noncrisis years and 61.57% of the sample in the year X during the crisis years and the level of variable X in the
prior to the crisis years (i.e., 1997 and 2006). To cap- year before the crisis years, i.e., Xcrisis – Xpre . 12
ture bank risk-taking behavior, we use the annual rate Following Liu and Ryan (2006), our main proxy for loan
of change in bank loans (Loans) and real estate loans defaults is the ratio of nonperforming loans to total gross
(RealEstateLoans). The annual rate of change in variable loans (NPL/Loan), which measures the portion of banks’
X in year t is defined as Xt = [Xt − Xt−1 ]/Xt−1 . Real estate loan portfolios in default or close to default and can also
lending, generally the largest part of bank lending, became capture ex ante lending quality. The higher the ratio of
an important consideration in the recent financial crisis nonperforming loans to total gross loans, the higher the
(see, e.g., Cooper, 2009; Demyanyk and Van Hemert, 2011; loan default rates and the more lower-quality loans that
Mian and Sufi, 2011; Huizinga and Laeven, 2012). In our were made ex ante by banks. For robustness, we use the
sample, the average annual rates of change in bank loans ratio of nonperforming loans to total equity (NPL/Equity)
and real estate loans are 12.59% and 10.28%, respectively,
in the noncrisis years.
12
The liability-side measure of risk is bank leverage. Pre- We calculate the amount of change in variable X from noncrisis to
crisis years using the level of variable X in crisis year 1998 minus the level
vious studies have shown that increases in leverage prior of variable X in 1997 for the 1998 Russian crisis and using the mean of
to a financial crisis played an important role in the ensu- variable X during the crisis years 20 07–20 09 minus the level of variable
ing crisis (see, e.g., Beltratti and Stulz, 2012; Fahlenbrach, X in 2006 for the financial crisis of 2007–2009.
200 P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209
Table 2
Bank characteristics for overconfident and non-overconfident banks.
This table presents value-weighted mean, calculated with weights relating to bank size, and mean dif-
ference between overconfident and non-overconfident banks. Overconfident (non-overconfident) banks
have chief executive officers classified as overconfident (non-overconfident). Over 1994–2009, we de-
fine 1998 and 20 07–20 09 as crisis years and others as noncrisis years. Variable definitions are in
Table A1. ∗, ∗∗ and ∗∗∗ denote significance of the t-test for the difference in the means between the
two subsamples at the 10%, 5%, and 1% level, respectively. + denote significance of Fisher’s exact test for
the difference in means between the two subsamples at the 10% level.
as another lending quality measure, which can represent noncrisis to crisis years (EDFcrisis – EDFpre ) with an average
the ability of shareholders to cover the losses. The higher value of 0.2954.
the ratio, the lower the ability to cover loan losses and Finally, TURNOVER is a dummy variable that equals one
the poorer a bank’s loan standards. The average change if the bank experienced at least one CEO turnover in crisis
in NPL/Loan (NPL/Equity) from noncrisis to crisis years, i.e., years, which represents on average 20.69% of our sample.
NPL/Loancrisis – NPL/Loanpre (NPL/Equitycrisis – NPL/Equitypre ), We also classify a bank as failed, following the definition in
is about 1.03% (16.14%). Fahlenbrach, Prilmeier, and Stulz (2012), if it was acquired
Our proxies for bank performance are return on by the FDIC, merged at a discount relative to the last close
equity (ROE), which describes bank operating perfor- before the merger announcement, or forced to delist by an
mance, and stock return during crisis years (Crisis Return), exchange during August 1998 through December 1998 or
which captures bank stock return performance. Following July 2007 through December 2009. We find that 7.28% of
Fahlenbrach, Prilmeier, and Stulz (2012), the crisis stock re- our sample failed in crisis years.
turns use buy-and-hold returns from July 1, 2007 through Beyond these primary variables, we study five other
December 31, 2008 for the financial crisis of 2007–2009 characteristics of banks as control variables: log of total as-
and from August 3, 1998 (the first trading day in August sets (Assets), ratio of total assets to book value of equity
1998) until the day in 1998 when the bank’s stock reached (Book Leverage), ratio of net income to total assets (ROA),
its lowest price for the 1998 Russian crisis.13 Not surpris- ratio of average balance of loans to average balance of de-
ingly, banks perform poorly during crisis years. The aver- posits (LOANDEP), and Tier 1 capital ratio (Tier 1).14 The
age change in ROE from noncrisis to crisis years (ROEcrisis – summary statistics for these variables are consistent with
ROEpre ) is about −10.50%, and the average Crisis Return is those in other studies examining bank characteristics in
−32.41%. the US. At the end of Table 1 is information about bank
We use the expected default probability (EDF) to proxy characteristics in the year prior to the crisis years.
for bank insolvency. Following Chava and Purnanandam
(2011) and Santos (2011), the EDF is estimated from the 4.2. Univariate analyses
modified Merton model (Bharath and Shumway, 2008). In
our sample, the expected default probability increases from Table 2 presents univariate results using a value-
weighted mean that is calculated with weights relating to
bank size. The last column shows mean differences be-
13
The same conclusion can be drawn from several other specifications tween overconfident and non-overconfident banks.
for the calculation of Crisis Return, including using buy-and-hold returns
from July 1, 2007 until the day during 2007–2009 when the bank’s stock
14
attained its lowest price for the most recent financial crisis and from Au- The selection of these variables is based on Dinç (2005), Ivashina and
gust 3, 1998 through December 31, 1998 for the 1998 Russian crisis. Scharfstein (2010), and Puri, Rocholl, and Steffen (2011).
P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209 201
16
Throughout the paper, WLS regressions are estimated using weights
relating to bank size.
17
The variable overconfidence measure OCi, t is time-varying in one of
15
According to the DuPont identity that ROE could be affected by ROA two ways. First, Campbell, Gallmeyer, Johnson, Rutherford, and Stanley
and financial leverage, we also show ROA and book leverage in Internet (2011) classify CEOs as overconfident from the first time they exhibit
Appendix Figure IA1. The figure reveals that the dramatic decline in ROE the postponed-exercise behavior, and before that time the same CEO had
of overconfident banks in crisis years is attributed both to a reduction been in the non-overconfident category. Second, because banks can re-
in ROA and an increase in book leverage, reflecting a huge loss in share- place their CEOs, the attitude of a new CEO toward project risks could be
holder wealth. different from his or her predecessor’s.
202 P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209
Table 3
Overconfident banks and annual rate of change in bank loans and leverage.
This table presents weighted least squares regression results on the annual rate of change in bank loans and leverage in the noncrisis years. Over 1994–
2009, we define 1998 and 2007–2009 as crisis years and others as noncrisis years.
Loans Real estate loans Regulatory leverage Book leverage Market leverage
Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Assets, Book Leverage, ROA, LOANDEP, and Tier 1.18 In all Hypothesis 1 that overconfident banks made aggressive
equations, we use the White (1980) heteroskedasticity- lending decisions more than non-overconfident banks in
consistent standard errors as well as the Petersen (2009) noncrisis years. For instance, after controlling for bank
approach to adjust for heteroskedasticity and clustering at characteristics, the coefficient on OC of 0.0467 in Model 2
the bank level. shows that overconfident banks, on average, experienced
The main results are reported in Table 3. For each de- a 4.67% higher annual rate of change in bank loans than
pendent variable, we conduct two specifications to test the non-overconfident banks during the crises.
overconfidence effect in the noncrisis years. The first speci- Real estate loan results are reported in Models 3
fication contains the CEO overconfidence variable only. The and 4. Similar to those in Models 1 and 2, the coef-
second specification adds controls for bank characteristics. ficients on OC are statistically significantly positive. For
To save space, we do not report the coefficients of bank instance, in Model 4, the coefficient on OC shows that
and year dummies in the analyses. Because data on lend- overconfident banks, on average, experienced a 6.70%
ing and Tier 1 capital ratio are not available for invest- higher annual rate of change in real estate loans than
ment banks, the regressions in Models 1–6 where the an- non-overconfident banks. This indicates that overconfident
nual rates of change in lending and regulatory leverage are banks made more aggressive lending decisions regarding
the dependent variables include only depository institu- real estate loans than non-overconfident banks before the
tions. We lose 70 bank-year observations in Models 3 and crises.
4, because data on real estate loans are not available in To further test our Hypothesis 2, Models 5–10 use dif-
the FDIC Call Report database. The regressions in Models ferent types of leverage as the dependent variables. No
7–10 where the annual rates of change in book leverage matter what the type of leverage, the coefficients on OC
and market leverage are the dependents variable include are positive and statistically significant at the 5% level or
both depository institutions and investment banks. better. For instance, in Model 10, the coefficient on OC of
In Models 1 and 2, the coefficients on OC are 0.0212 shows that overconfident banks, on average, expe-
both positive and statistically significant, confirming rienced a 2.12% higher annual rate of change in market
leverage than non-overconfident banks. The results provide
evidence that overconfident banks tended to have higher
18
The correlation coefficient matrix of the variables is reported in Ta- increases in leverage than non-overconfident banks in the
ble IA1 of the Internet Appendix. The correlations between all variables noncrisis years.
are very low, making multicollinearity less of a concern. Significant posi- These results overall support our first two hypothe-
tive correlations exist between Loans, Real Estate Loans, Book Lever-
age, Market Leverage, Regulatory Leverage, and the CEO overconfidence
ses that during the noncrisis years overconfident banks
proxy (OC), supporting the contention that overconfident banks tend to were more aggressive in granting loans, particularly real
exhibit more aggressive behavior. estate loans, than non-overconfident banks. Overconfident
P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209 203
Table 4
Overconfident banks and annual rate of change in bank loans and leverage: two-stage least squares (2SLS) regression.
This table presents a 2SLS regression on the annual rate of change in bank loans and leverage in the noncrisis years. Over 1994–2009, we define 1998 and
20 07–20 09 as crisis years and others as noncrisis years.
Stage 1. Regress OC on the instrument and other exogenous variables of the model
Stage 2. Replace the fitted value of OC derived from the first stage with OC in the main regression Eq. (1):
i,t + β Zi,t−1 + νi + μt + εi,t ,
Annual Rat ei,t = α1 + α2 OC
where L is the logistic distribution, OCi, t is a dummy variable that equals one if the bank i in year t is an overconfident bank and zero otherwise, OC is the
predicted value of OC from the first stage regression, CEO Age is the age of CEO, Zi,t−1 is a vector of control variables from bank i in year t – 1, ν i and μt
capture fixed effects of bank and year, respectively, ɛi, t is the random error, and Annual Ratei, t represents five dependent variables of interest: the annual
rate of change in loans (Loansi, t ), real estate loans (Real Estate Loansi, t ), regulatory leverage (Regulatory Leveragei, t ), book leverage (Book Leveragei, t ),
and market leverage (Market Leveragei, t ) for bank i in year t. Variable definitions are in Table A1. In parentheses are t-statistics based on standard errors
adjusted for heteroskedasticity (White, 1980) and bank clustering (Petersen, 2009). ∗ , ∗∗ , and ∗∗∗ denote significance at the 10%, 5%, and 1% level, respectively.
To save space, we do not report coefficients on bank and year dummies.
First stage Second stage Second stage Second stage First stage Second stage Second stage
OC Loans Real estate loans Regulatory leverage OC Book leverage Market leverage
Variable (1) (2) (3) (4) (5) (6) (7)
banks also had greater growth in leverage than non- people show greater overconfidence than younger people.
overconfident banks. Thus, we suggest that the age of a CEO is positively related
to CEO overconfidence.19 We also have no reason to believe
5.1.2. Endogoneity issue that the instrument has a direct economic impact on bank
The effect of CEO overconfidence on bank lending and lending and leverage, so it is unlikely to be correlated with
leverage could suffer from endogeneity problems. Thus, in the error term in the second stage regression.
Table 4, we estimate a 2SLS regression model with bank In the 2SLS model, we treat OC as an endogenous vari-
and year fixed effects to determine whether our regression able that we instrument with CEO Age in the first stage.
results are robust to endogeneity. Stage 1. Regress OC on the instrument and other exoge-
For a 2SLS regression analysis, it is important first to nous variables of the model:
find a good instrument that is an exogenous variable eco- P (OCi,t = 1|CEO Agei,t , Zi,t−1 ) = L(δ1 + δ2CEO Agei,t
nomically related to CEO overconfidence but is uncorre- +θ Zi,t−1 + νi + μt ). (2)
lated with the error term of the regression relating the
bank lending and leverage. To find such an instrument, we Stage 2. Replace the fitted value of OC derived from the
consider a factor that the literature on overconfidence in- first stage with OC in the main regression Eq. (1):
dicates empirically affects the determinants of CEO over- i,t + β Zi,t−1 + vi + μt + εi,t ,
Annual Rat ei,t = α1 + α2 OC
confidence (see, e.g., Crawford and Stankov, 1996; Bruin,
(3)
Parker, and Fischhoff, 2012): age of the CEO (CEO Age).
Bruin, Parker, and Fischhoff (2012) suggest that the re- where L is the logistic distribution, OC is a dummy variable
lation between age and degree of confidence depends on that equals one if the bank is an overconfident bank and
how cognitively demanding a task is. In demanding jobs,
such as the CEO position of a company, they suggest that 19
In our sample during noncrisis years, the average age of overconfident
older adults are more overconfident than younger ones. CEOs is 57.5084, which is significantly older by 1.1347 years than non-
Similarly, Crawford and Stankov (1996) find that older overconfident CEOs.
204 P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209
zero otherwise, OC is the fitted value of OC derived from hypothesized variable to examine the overconfidence ef-
the first stage regression, CEO Age is the age of the CEO, fect on lending quality, bank performance, bank insolvency,
Z is a vector of control variables, ν i and μt capture fixed CEO turnover, and bank failure in the crisis years. Sample
effects of bank and year, and ɛi, t is the residual of the sec- sizes vary for reasons of data availability.
ond stage regressions. Estimation of the second-stage re-
gression uses the same dependent variables as in Eq. (1). 5.2.1. Effects of CEO overconfidence on bank loan quality
Models 1 and 5 of Table 4 report the first stage results To measure the quality of loans, we use NPL/Loan and
relating OC to CEO Age and other exogenous variables of NPL/Equity as the principle variables. Our main focus is on
the model in different samples. Model 1 includes depos- increases in these ratios after the crises for both overconfi-
itory institutions, and Model 5 adds investment banks to dent and non-overconfident banks. If there are greater in-
the sample used in Model 1. Consistent with the hypothe- creases in NPL/Loan and NPL/Equity after the crises for over-
sized economic relation between our instruments and CEO confident banks than for non-overconfident banks, we can
overconfidence, the first stage results show that OC is pos- infer that overconfident banks were more prone to have
itively related to CEO Age, and the coefficient is statistically originated low-quality loans that experienced higher future
significant at the 1% level. default rates than non-overconfident banks.
Models 2, 3, 4, 6, and 7 report the coefficients from the In Table 5, we use a cross-sectional WLS regression to
second stage regression of the annual rates of change in test the effects of overconfidence on lending quality, where
bank loans and leverage-relevant variables on the value of the amount of change in NPL/Loan or NPL/Equity from the
and the corresponding exogenous control variables. The
OC noncrisis years to the crisis years is the dependent vari-
coefficients on OC are all statistically significantly positive, able, and OCpre is the primary independent variable:
similar to the results reported in Table 3. Hence, we con-
Xi,crisis − Xi,pre = α1 + α2 OCi,pre + β Zi,pre + εi , (4)
clude that our findings of the effect of CEO overconfidence
on bank lending and leverage are robust to an endogeneity where X is either NPL/Loan or NPL/Equity, OCpre is a dummy
bias. variable that equals one if the bank is an overconfident
bank in the year prior to the crisis years and zero other-
5.2. Consequences of overconfident banks wise, Zpre represents a vector of control variables that are
measured in the year prior to the crisis years, and ɛi is the
To examine the consequences for overconfident banks random error.
that relaxed lending standards and increased leverage ra- Models 1 and 3 include only OCpre as the indepen-
tios before the crises, we set the CEO overconfidence mea- dent variable, and Models 2 and 4 further control for bank
sure in the year prior to the crisis years, OCpre , as the characteristics. The coefficients on OCpre are positive and
Table 5
Overconfident banks and bank loan quality.
This table presents weighted least squares regression results regarding the effects of banks’ chief executive
officer overconfidence on bank loan quality in the crisis years. Over 1994–2009, we define 1998 and 20 07–20 09
as crisis years and others as noncrisis years. To measure bank loan quality, we use nonperforming loans to total
loans (NPL/Loan) and nonperforming loans to total equity (NPL/Equity). We examine the amount of change in
bank loan quality from noncrisis to crisis years by computing the difference between average bank loan quality
in crisis years and bank loan quality in the year prior to the crisis years, Xcrisis – Xpre .
where X is either NPL/Loan or NPL/Equity for bank i, OCpre is a dummy variable that equals one if a bank is an
overconfident bank in the year prior to the crisis years and zero otherwise, Zpre represents a vector of control
variables measured in the year prior to the crisis years, and ɛi is the random error. Variable definitions are in
Table A1. In parentheses are t−statistics based on standard errors adjusted for heteroskedasticity (White, 1980).
∗, ∗∗, and ∗∗∗ denote significance at the 10%, 5%, and 1% level, respectively.
Table 6
Overconfident banks and bank performances and insolvencies.
This table presents weighted least squares regression results regarding the effects of banks’ chief executive officer overconfidence on bank performances
and insolvencies in crisis years. Over 1994–2009, we define 1998 and 20 07–20 09 as crisis years and others as noncrisis years.
and
E DFi,crisis − E DFi,pre = α1 + α2 OCi,pre + β Zi,pre + εi ,
where BankPerformi,crisis refers to ROEi,crisis –ROEi,pre and Crisis Returni , respectively, for bank i, OCpre is a dummy variable that equals one if a bank is an
overconfident bank in the year prior to the crisis years and zero otherwise, Zpre represents a vector of control variables that are measured in the year prior
to the crisis years, and ɛi is the random error. Variable definitions are in Table A1. In parentheses are t-statistics based on standard errors adjusted for
heteroskedasticity (White, 1980). ∗, ∗∗, and ∗∗∗ denote significance at the 10%, 5%, and 1% level, respectively.
Intercept −14.8128∗∗∗ −40.0778∗∗∗ 1.0314 −0.2858∗∗∗ 0.0170 0.1264 0.4940∗∗∗ 0.1970 0.3980
(−5.03) (−3.13) (0.06) (−9.24) (0.15) (0.55) (13.91) (1.29) (1.03)
OCpre −8.7653∗ −7.5365∗∗ −4.8702∗∗ −0.0786∗ −0.0883∗∗ −0.0738∗ 0.1420∗∗∗ 0.1550∗∗∗ 0.1270∗∗
(−1.92) (−2.51) (−2.47) (−1.85) (−2.14) (−1.82) (3.06) (3.34) (2.29)
Assets 5.6054∗∗∗ 1.0669 −0.0238∗∗ −0.0185 0.0296∗∗ 0.0210
(3.22) (1.09) (−2.08) (−1.35) (1.99) (1.15)
Book leverage −2.0322∗∗∗ −1.7623∗∗∗ −0.0071∗∗ −0.0100 −0.0 0 01 −0.0 0 05
(−3.52) (−4.78) (−2.33) (−1.41) (−0.02) (−0.04)
ROA 4.9112∗∗ 15.8965∗∗∗ 0.0 0 02 0.0795∗ −0.0077 −0.1270∗∗
(2.02) (7.84) (0.03) (1.81) (−0.46) (−2.20)
LOANDEP −0.6918 −0.2873∗∗∗ 0.1020
(−0.12) (−2.95) (0.75)
Tier 1 0.0285 0.0044 −0.0038
(0.06) (0.57) (−0.26)
Number of observations 243 243 201 243 243 201 207 207 165
Adjusted R2 0.0900 0.4903 0.8400 0.0083 0.0068 0.1298 0.3782 0.3873 0.4264
statistically significant at the 5% level or better in all mod- The dependent variable is measured as the amount of
els, implying that overconfident banks were more likely to change in EDF from noncrisis to crisis years, EDFcrisis –
write loans with a high likelihood of becoming nonper- EDFpre . The cross-sectional WLS regression is
forming in a future economic downturn.
E DFi,crisis − E DFi,pre = α1 + α2 OCi,pre + β Zi,pre + εi , (6)
5.2.2. Effects of CEO overconfidence on bank performance where all independent variables are defined as before.
and insolvency In Table 6, Models 7, 8, and 9 show that OCpre is sig-
Our fourth hypothesis is that CEO overconfidence nificantly and positively correlated with the change in ex-
led to poorer bank performance than banks with non- pected default probability, supporting Hypothesis 5 that
overconfident CEOs when the crises occurred. To test this the aggressive lending behavior of overconfident banks ac-
hypothesis, we adopt a cross-sectional WLS regression as companied by high levels of leverage subjected the banks
follows: to more of an increase in expected default probability in
crisis years.
BankPer f or mi,crisis = α1 + α2 OCi,pre + β Zi,pre + εi , (5)
where BankPerformi,crisis refers to the change in ROE from 5.2.3. Effects of CEO overconfidence on CEO turnover and
noncrisis to crisis years (ROEi,crisis –ROEi,pre ) and the bank’s bank failure
annualized returns during crisis years (Crisis Returni ), re- Overconfident CEOs were aggressive in assuming risk
spectively, for bank i. OCpre is the primary independent in the noncrisis years, making them more likely to be re-
variable, and the independent variables are defined as be- placed and pushing their banks closer to failure when the
fore. crises hit. To test these two hypotheses, we use two logis-
Models 1–6 of Table 6 show a significantly negative as- tic regressions:
sociation between OCpre and postcrisis bank performance,
P (T URNOV E Ri = 1|OCi,pre , Zi,pre ) = L(α1 +α2 OCi,pre
supporting our hypothesis that CEO overconfidence led
to poorer bank performance after the crises. The reasons +β Zi,pre ) (7)
could be the mass of nonperforming loans and the pres- and
sure of interest payments facing overconfident banks in
crisis years. P (F AILUREi = 1|OCi,pre , Zi,pre ) = L(α1 +α2 OCi,pre +β Zi,pre ),
While overconfident banks have been shown to suffer
(8)
poorer operating and stock return performance than non-
overconfident banks in the crisis years, did they become where L is the logistic distribution, TURNOVER is a dummy
more insolvent than non-overconfident banks in this pe- variable that equals one if the bank experienced a CEO
riod? To answer this question, we examine the impact of turnover in the crisis years and zero otherwise, and FAIL-
CEO overconfidence on expected default probability (EDF). URE is a dummy variable that equals one if a bank failed
206 P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209
Table 7
Overconfident banks, chief executive officer (CEO) turnover, and bank failure.
This table presents logistic regression analyses of CEO turnover and bank failures in the crisis years:
and
P (F AILUREi = 1|OCi,pre , Zi,pre ) = L(α1 + α2 OCi,pre + β Zi,pre ),
where L is the logistic distribution, TURNOVER is a dummy variable that equals one if the bank experienced a CEO
turnover in the crisis years and zero otherwise, FAILURE is a dummy variable that equals one if a bank failed during the
crisis years and zero otherwise, OCpre is a dummy variable that equals one if a bank is overconfident bank in the year
prior to the crisis years and zero otherwise, and Zpre represents a vector of control variables measured in the year prior
to the crisis years. Variable definitions are in Table A1. In parentheses are t-statistics based on standard errors adjusted
for heteroskedasticity (White, 1980). ∗ , ∗∗ , and ∗∗∗ denote significance at the 10%, 5%, and 1% level, respectively.
TURNOVER FAILURE
Variable (1) (2) (3) (4) (5) (6)
during the crisis years and zero otherwise. All independent tion is a bank’s risk culture. Fahlenbrach, Prilmeier, and
variables are defined as before. Stulz (2012) propose that a bank with an aggressive risk
Models 1–3 of Table 7 report coefficients on OCpre that culture generally chooses to take greater risks, and such
are positive and statistically significant at the 1% level. a risk culture is persistent. While previous studies have
This supports Hypothesis 6 that overconfident banks were shown that overconfident CEOs are more willing to un-
more likely to replace their CEOs than non-overconfident dertake risky projects (Simon and Houghton, 2003; Goel
banks in crisis years. This evidence complements findings and Thakor, 2008; Niu, 2010; Gervais, Heaton, and Odean,
in Goel and Thakor (2008) and Campbell, Gallmeyer, John- 2011; Hirshleifer, Low, and Teoh, 2012), we conjecture that
son, Rutherford, and Stanley (2011), who suggest that over- a bank with an aggressive risk culture prefers an overcon-
confident CEOs are more likely to be replaced in industrial fident CEO, to the extent that a bank with an overconfident
firms. CEO in 1997 was more likely to have an overconfident CEO
The aggressive lending and leveraging activities in over- in 2006 than banks without overconfident CEOs.
confident banks can make them more likely to suffer illiq- To test this, we conduct a logistic regression with
uidity than non-overconfident banks when the economy OC2006 as the dependent variable and OC1997 as the primary
enters a downturn. Illiquidity could have made them less independent variable. OC2006 (OC1997 ) is a dummy variable
able to meet obligations on time and thus subject to a that equals one if the bank is an overconfident bank in
higher likelihood of failure. The results reported in Mod- 2006 (1997) and zero otherwise. The results are reported
els 4–6 of Table 7 show a significantly positive relation be- in Table 8. In Model 1, we find that management over-
tween overconfident banks and the likelihood of bank fail- confidence in 1997 strongly predicts management overcon-
ure during crisis years. This confirms our conjecture that fidence in 2006. When we regress OC2006 on OC1997 as
overconfident banks experienced a greater likelihood of well as on characteristics of banks in 2005, OC1997 remains
failure than non-overconfident banks when facing an ex- highly significant. Model 3 shows that the coefficient on
ogenous shock. OC1997 is 2.0862, statistically significant at the 5% level,
meaning that the odds of overconfidence in 2006 for banks
identified as overconfident in 1997 is 8.0543 (calculated by
6. Equilibrium
e2.0862 = 8.0543) times the odds for other banks. These re-
sults indicate that banks with an aggressive risk culture
If our evidence shows that CEO overconfidence made
in 1997 were still likely to have an overconfident manager
banks more vulnerable to crises, one could ask why a bank
prior to the most recent financial crisis, even though over-
would hire an overconfident CEO.20 One possible explana-
confident managers incurred heavy losses in the 1998 Rus-
sian crisis, a finding that supports persistence in a bank’s
20
We thank the referee for pointing out this question. risk culture.
P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209 207
Table 8
Logit regressions predicting membership in the 2006 overconfident bank group.
This table presents logistic regression analyses to test the association between banks with over-
confident chief executive officers (CEOs) in 1997 and the likelihood of banks with overconfident
CEOs in 2006:
P (OCi,2006 = 1|OCi,1997 , CEO Agei,2005 , Zi,2005 ) = L(α1 + α2 OCi,1997 + α3CEO Agei,2005 + β Zi,2005 ),
where L is the logistic distribution, OC2006 is a dummy variable that equals one if the bank
is an overconfident bank in 2006 and zero otherwise, OC1997 is a dummy variable that equals
one if the bank is an overconfident bank in 1997 and zero otherwise, CEO Age is the age of
CEO in 2005, and Z2005 represents a vector of control variables measured in 2005. Variable
definitions are in Table A1. In parentheses are t-statistics based on standard errors adjusted for
heteroskedasticity (White, 1980). ∗, ∗∗, and ∗∗∗ denote significance at the 10%, 5%, and 1% level,
respectively.
If the risk culture of banks can explain why banks tions rapidly morphed into a crisis for the entire financial
hire overconfident CEOs, why would shareholders prefer system. Shocks to bank health have played a systematic
greater risk taking? Cheng, Hong, and Scheinkman (2015) role in worsening financial stability. Yet not all financial
suggest that institutional investors can push managers to institutions suffered during these crisis years. We propose
take action using a risky business model. This is because a new perspective that manager overconfidence could ex-
the US accounting system generally treats long-term, in- plain the substantial heterogeneity in bank risk-taking be-
tangible investments as an immediate expense and short- haviors during a boom, as well as the performance of these
term performance pressures and fiduciary responsibilities banks during the ensuing crisis years.
lead some institutions to rely on near-term earnings as a Our empirical evidence shows that the aggressive risk
measure of the performance of an investment (see, e.g., culture of some banks makes them more likely to hire
Porter, 1992; Lang and McNichols, 1997; Bushee, 2001). We overconfident CEOs who take risks when the economy is
thus infer that institutional investors owning large blocks in an upturn. Overconfident banks tended to approve more
of shares could be influential in leading banks to hire over- loans and assume more leverage than non-overconfident
confident managers to take risk. banks before crises, which gave them more serious ex-
To investigate this possible explanation, we com- posures. With the onset of a crisis, overconfident banks
pare the institutional ownership of overconfident banks suffered greater capital losses from many nonperforming
and non-overconfident banks. We find higher institu- loans, which, accompanied by higher leverage, caused a
tional ownership for overconfident banks than for non- much greater drop in their net worth than for other banks.
overconfident banks in both 1997 and 2006. Untabulated The unexpected losses preceded sharper drops in banks’
results show that, in 1997 (2006), 50.51% (65.32%) of shares performance, greater increases in expected default proba-
of overconfident banks were held by institutional investors, bility, and a higher likelihood of CEO turnover and failures
which is 5.22 (12.13) percentage points higher than the than for non-overconfident banks in the crisis years.
holdings in non-overconfident banks (statistically signifi- Overall, our results suggest that overconfidence, includ-
cant at the 5% level). This result confirms our conjecture ing underestimation of project risks, can lead a risk-averse
and provides further support for the risk culture hypothe- bank CEO to take exposures perceived to be the most prof-
sis proposed by Fahlenbrach, Prilmeier, and Stulz (2012). itable for current shareholders ex ante but that could harm
their banks and themselves ex post.
7. Conclusion
Appendix
Bank financial contagion is cited as a key feature of the
financial crises, as problems at specific financial institu- Table A1
208 P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209
Table A1
Variable definitions.
Overconfidence variables
OC Dummy variable that equals one if the bank is an overconfident bank and zero otherwise. In ExecuComp
overconfident banks, chief executive officers (CEOs) postpone exercising highly in-the-money (EC)
options at least twice during their tenure. CEOs are called highly overconfident CEOs the first
time they start to exhibit this behavior. Following Campbell, Gallmeyer, Johnson, Rutherford,
and Stanley (2011), we choose 100% moneyness as the cutoff point to identify CEOs as highly
overconfident.
Bank level variables
Loans Annual change in bank loans normalized by the previous bank loan value. Compustat
Bank (CB)
Real estate loans Annual change in real estate loans normalized by the previous real estate loan value. Call Report
Book leverage Annual change in book leverage normalized by the previous book leverage value. Book leverage is CB
calculated as total assets / book value of equity.
Market leverage Annual change in market leverage normalized by the previous market leverage value. Market CRSP and
leverage is calculated as (market value of equity + book value of liabilities) / market value of CB
equity. Market value of equity is calculated as year-end stock price × shares outstanding.
Regulatory leverage Annual change in regulatory leverage normalized by the previous regulatory leverage value. CB
Regulatory leverage is calculated as total assets divided by Tier 1 capital.
NPL/Loan Ratio of nonperforming loans to total gross loans. CB
NPL/Equity Ratio of nonperforming loans to total equity. CB
ROE Ratio of net income to total equity. CB
Crisis return Bank stock return in crisis years using buy-and-hold returns. The holding period is from July 1, CRSP
2007 through December 31, 2009 for the financial crisis of 2007–2009 and from August 3, 1998
(the first trading day in August 1998) until the day in 1998 when the bank’s stock attained its
lowest price for the 1998 Russian crisis.
EDF Expected default frequency measure of the firm. It is the percentile ranking of a firm’s default risk CRSP and
based on its distance to default (constructed from Bharath and Shumway, 2008). Compustat
TURNOVER Dummy variable that equals one if the bank experienced a CEO turnover and zero otherwise. EC
FAILURE Dummy variable that equals one if the bank failed and zero otherwise. Banks are classified as FDIC
failed if the bank was acquired by the Federal Deposit Insurance Corporation (FDIC), merged at
a discount relative to the last close before the merger announcement, or was forced to delist by
an exchange during August 1998 through December 1998 or July 2007 through December 2009.
Control variables: Bank and
CEO characteristics
Assets Log of total assets (billions of us dollars). CB
Book leverage Ratio of assets to book value of equity. CB
ROA Ratio of net income to total assets. CB
LOANDEP Ratio of average balance of loans to average balance of deposits. CB
Tier 1 Tier 1 capital ratio is the ratio of a bank’s core equity capital to its total risk-weighted assets, CB
which is calculated according to the Basel Accord for reporting risk-adjusted capital adequacy.
CEO Age Age of CEO EC
References Biorn, E., 20 0 0. Panel data with measurement errors: instrumental vari-
ables and GMM procedures combining levels and differences. Econo-
Adrian, T., Shin, H.S., 2009. Money, liquidity and monetary policy. Ameri- metric Review 19, 391–424.
can Economic Review 99, 600–605. Bruin, W., Parker, A., Fischhoff, B., 2012. Explaining adult age differences
Adrian, T., Shin, H.S., 2010. Liquidity and leverage. Journal of Financial In- in decision-making competence. Journal of Behavioral Decision Mak-
termediation 19, 418–437. ing 25, 352–360.
Alicke, M., 1985. Global self-evaluation as determined by the desirability Burg, V., Scheinert, T., Streitz, D., 2012. Bank CEO overconfidence and risk-
and controllability of trait adjectives. Journal of Personality and Social taking Unpublished working paper. Humboldt University of Berlin,
Psychology 49, 1621–1630. Berlin, Germany.
Becker, B., Ivashina, V., 2014. Cyclicality of credit supply: firm level evi- Bushee, B., 2001. Do institutional investors prefer near-term earnings over
dence. Journal of Monetary Economics 62, 76–93. long-run value? Contemporary Accounting Research 18, 207–246.
Beltratti, A., Stulz, R., 2012. The credit crisis around the globe: why did Campbell, T.C., Gallmeyer, M.F., Johnson, S.A., Rutherford, J., Stanley, B.,
some banks perform better? Journal of Financial Economics 105, 1– 2011. CEO optimism and forced turnover. Journal of Financial Eco-
17. nomics 101, 695–712.
Ben-David, I., Graham, J.R., Harvey, C.R., 2013. Managerial miscalibration. Chava, S., Purnanandam, A., 2011. The effect of banking crisis on bank-
Quarterly Journal of Economics 128, 1547–1584. dependent borrowers. Journal of Financial Economics 99, 116–135.
Berger, A.N., Bouwman, C.H.S., 2013. How does capital affect bank per- Chen, H., Kacperczyk, M., Ortiz-Molina, H., 2011. Labor unions, operating
formance during financial crises? Journal of Financial Economics 109, flexibility, and the cost of equity. Journal of Financial and Quantitative
146–176. Analysis 46, 25–58.
Berger, A.N., Hannan, T., 1998. The efficiency cost of market power in the Cheng, I.H., Hong, H., Scheinkman, J.A., 2015. Yesterday’s heroes: compen-
banking industry: a test of the quiet life and related hypotheses. Re- sation and risk at financial firms. Journal of Finance, forthcoming.
view of Economics and Statistics 80, 454–465. Cole, R.A., Gunther, J.W., 1995. Separating the likelihood and timing of
Bharath, S., Shumway, T., 2008. Forecasting default with the Merton dis- bank failure. Journal of Banking and Finance 19, 1073–1089.
tance to default model. Review of Financial Studies 21, 1339–1369. Cole, R.A., White, L.J., 2012. Déjà vu all over again: the causes of US com-
Biddle, G.C., Hilary, G., 2006. Accounting quality and firm-level capital in- mercial bank failures this time around. Journal of Financial Services
vestment. Accounting Review 81, 963–982. Research 42, 5–29.
P.-H. Ho et al. / Journal of Financial Economics 120 (2016) 194–209 209
Cooper, D., 2009. Impending US spending bust? The role of housing Liu, C., Ryan, S.G., 2006. Income smoothing over the business cycle:
wealth as borrowing collateral Public Policy Discussion Paper No. 09- changes in banks’ coordinated management of provisions for loan
9. Federal Reserve Bank, Boston, MA. losses and loan charge-offs from the pre-1990 bust to the 1990s
Coughlan, A.T., Schmidt, R.M., 1985. Executive compensation, management boom. Accounting Review 81, 421–441.
turnover, and firm performance. Journal of Accounting and Economics Ma, Y., 2014. Bank CEO optimism and the financial crisis Unpublished
7, 43–66. working paper. Harvard University, Cambridge, MA.
Crawford, J.D., Stankov, L., 1996. Age differences in the realism of confi- Malmendier, U., Tate, G., 2005. CEO overconfidence and corporate invest-
dence judgments: a calibration study using tests of fluid and crystal- ment. Journal of Finance 60, 2661–2700.
lized intelligence. Learning and Individual Differences 8, 83–103. Malmendier, U., Tate, G., 2008. Who makes acquisitions? CEO overconfi-
Dell’Ariccia, G., Marquez, R., 2006. Lending booms and lending standards. dence and the market’s reaction. Journal of Financial Economics 89,
Journal of Finance 61, 2511–2546. 20–43.
Demyanyk, Y., Van Hemert, O., 2011. Understanding the subprime mort- Malmendier, U., Tate, G., Yan, J., 2011. Overconfidence and early-life expe-
gage crisis. Review of Financial Studies 24, 1848–1880. riences: the effect of managerial traits on corporate financial policies.
Didier, T., Hevia, C., Schmukler, S., 2012. How resilient and countercyclical Journal of Finance 66, 1687–1733.
were emerging economies during the global financial crisis? Journal Männasoo, K., Mayes, D.G., 2009. Explaining bank distress in Eastern Eu-
of International Money and Finance 31, 1971–1975. ropean transition economies. Journal of Banking and Finance 33, 244–
Dinç, I.S., 2005. Politicians and banks: political influences on government- 253.
owned banks in emerging countries. Journal of Financial Economics Merton, R.C., 1974. On the pricing of corporate debt: the risk structure of
77, 453–459. interest rates. Journal of Finance 29, 449–470.
Ellul, A., Yerramilli, V., 2013. Stronger risk controls, lower risk: evidence Mian, A., Sufi, A., 2011. House prices, home equity-based borrowing, and
from US bank holding companies. Journal of Finance 68, 1757–1803. the US household leverage crisis. American Economic Review 101,
Erel, I., Nadauld, T., Stulz, R., 2014. Why did holdings of highly rated secu- 2132–2156.
ritization tranches differ so much across banks? Review of Financial Niu, J., 2010. The effect of CEO overconfidence on bank risk taking. Eco-
Studies 27, 404–453. nomics Bulletin 30, 3288–3299.
Fahlenbrach, R., Prilmeier, R., Stulz, R., 2012. This time is the same: using Palumbo, M.G., Parker, J.A., 2009. The integrated financial and real system
bank performance in 1998 to explain bank performance during the of national accounts for the United States: does it presage the finan-
recent financial crisis. Journal of Finance 67, 2139–2185. cial crisis? American Economic Review 99, 80–86.
Fahlenbrach, R., Stulz, R., 2011. Bank CEO incentives and the credit crisis. Parrino, R., 1997. CEO turnover and outside succession a cross-sectional
Journal of Financial Economics 99, 11–26. analysis. Journal of Financial Economics 46, 165–197.
Farrell, K.A., Whidbee, D.A., 2003. The impact of firm performance expec- Petersen, M., 2009. Estimating standard errors in finance panel data sets:
tations on the CEO turnover and replacement decisions. Journal of Ac- comparing approaches. Review of Financial Studies 22, 435–480.
counting and Economics 36, 165–196. Porter, M., 1992. Capital choices: changing the way America invests in in-
Gervais, S., Heaton, J.B., Odean, T., 2011. Overconfidence, compensation dustry. Journal of Applied Corporate Finance 5, 4–16.
contracts, and capital budgeting. Journal of Finance 66, 1735–1777. Puri, M., Rocholl, J., Steffen, S., 2011. Global retail lending in the aftermath
Goel, A.M., Thakor, A.V., 2008. Overconfidence, CEO selection, and corpo- of the US financial crisis: distinguishing between supply and demand
rate governance. Journal of Finance 63, 2737–2784. effects. Journal of Financial Economics 100, 556–578.
Graham, J.R., Harvey, C.R., Puri, M., 2013. Managerial attitudes and corpo- Rajan, U., Seru, A., Vig, V., 2015. The failure of models that predict failure:
rate actions. Journal of Financial Economics 109, 103–121. distance, incentives and defaults. Journal of Financial Economics 115,
Griliches, Z., Hausman, J.A., 1986. Errors in variables in panel data. Journal 237–260.
of Econometrics 31, 93–118. Rajgopal, S., Shevlin, T., 2002. Empirical evidence on the relation between
Hackbarth, D., 2008. Managerial traits and capital structure decisions. stock option compensation and risk taking. Journal of Accounting and
Journal of Financial and Quantitative Analysis 43, 843–882. Economics 33, 145–171.
Hirshleifer, D.A., Low, A., Teoh, S.H., 2012. Are overconfident CEOs better Roberts, M.R., Whited, T.M., 2013. Endogeneity in empirical corporate fi-
innovators? Journal of Finance 67, 1735–1777. nance. In: Constantinides, G., Harris, M., Stulz, R. (Eds.), Handbook of
Hirshleifer, D.A., Luo, D.Y., 2001. On the survival of overconfident traders the Economics of Finance: Empirical Corporate Finance 2A. Elsevier,
in a competitive securities market. Journal of Financial Markets 4, 73– North Holland, Amsterdam, pp. 493–572.
84. Santos, J.A.C., 2011. Bank corporate loan pricing following the subprime
Huizinga, H., Laeven, L., 2012. Bank valuation and accounting discretion crisis. Review of Financial Studies 24, 1916–1943.
during a financial crisis. Journal of Financial Economics 106, 614–634. Simon, M., Houghton, S.M., 2003. The relationship between overconfi-
Ivashina, V., Scharfstein, D., 2010. Bank lending during the financial crisis dence and the introduction of risky products: evidence from a field
of 2008. Journal of Financial Economics 97, 319–338. study. Academy of Management Journal 46, 139–149.
Laeven, L., 2011. Banking crises: a review. Annual Review of Financial Eco- Svenson, O., 1981. Are we all less risky and more skillful than our fellow
nomics 3, 17–40. drivers? Acta Psychologica 47, 143–148.
Laeven, L., Valencia, F., 2013. The real effects of financial sector interven- Warner, J.B., Watts, R., Wruck, K., 1988. Stock prices and top management
tions during crises. Journal of Money, Credit and Banking 45, 147–177. changes. Journal of Financial Economics 20, 461–492.
Lang, M., McNichols, M., 1997. Institutional trading and corporate perfor- Wheelock, D.C., Wilson, P.W., 20 0 0. Why do banks disappear? The deter-
mance Unpublished working paper. Stanford University, Stanford, CA. minants of US bank failures and acquisitions. Review of Economics
Larwood, L., Whittaker, W., 1977. Managerial myopia: self-serving biases and Statistics 82, 127–138.
in organizational planning. Journal of Applied Psychology 62, 194– White, H., 1980. A heteroskedasticity-consistent covariance matrix estima-
198. tor and a direct test for heteroskedasticity. Econometrica 48, 817–838.