1
Yesterday once more:
Short selling and two banking crises
Dien Giau Bui*
Department of Finance
National Taiwan University
Chih-Yung Lin**
College of Management &
Innovation Center for Big Data and Digital Convergence
Yuan Ze University
Tse-Chun Lin***
Faculty of Business and Economics
University of Hong Kong
Abstract
We find that change of short interest predicts banks’ stock returns during the two
recent banking crises. More strikingly, before the 2007-2009 crisis, short interests
increase more for the banks that suffered more in the LTCM crisis. We also find that
change of short interest predicts banks’ loan quality and default risk during the
2007-2009 crisis. The results are stronger for banks with higher risk-taking behavior.
Overall, our findings suggest that short sellers are informed about the persistent risk
culture or risky business models of banks documented in Fahlenbrach, Prilmeier, and
Stulz (2012) and short these banks before the crisis.
JEL classification: G01, G14, G21, G32
Keywords: Short selling; short interest; financial crisis; predictability; risk culture
______________________________
We would like to thank David Lucca, Rüdiger Fahlenbrach, and Gerard Hoberg for the comments.
Tse-Chun Lin gratefully acknowledges research support from the Faculty of Business and Economics
at the University of Hong Kong and the Research Grant Council of the Hong Kong SAR government.
Chih-Yung Lin appreciates the financial support from Taiwan Ministry of Science and Technology.
Any remaining errors are ours. * E-mail: [email protected].
** Corresponding author. E-mail: [email protected]
*** E-mail: [email protected]
2
Yesterday once more:
Short selling and two banking crises
Abstract
We find that change of short interest predicts banks’ stock returns during the two
recent banking crises. More strikingly, before the 2007-2009 crisis, short interests
increase more for the banks that suffered more in the LTCM crisis. We also find that
change of short interest predicts banks’ loan quality and default risk during the
2007-2009 crisis. The results are stronger for banks with higher risk-taking behavior.
Overall, our findings suggest that short sellers are informed about the persistent risk
culture or risky business models of banks documented in Fahlenbrach, Prilmeier, and
Stulz (2012) and short these banks before the crisis.
JEL classification: G01, G14, G20, G32
Keywords: Short selling; short interest; financial crisis; predictability; risk culture
3
1. Introduction
A burgeoning literature has been investigating why some banks underperformed,
particularly during financial crises. For example, Fahlenbrach, Prilmeier, and Stulz
(2012) suggest that the persistent culture in a financial institution regarding
risk-taking behavior plays an important role. They find that a bank’s performance in
the past crisis could be a proxy for its systemic risk exposure or its sensitivity to the
next crisis. Hence, the performance of a bank in the past crisis could predict its
performance in the next crisis.1 In this paper, we investigate whether short sellers are
informed about the persistent risk culture and risky business models of banks and thus
target the underperforming ones in the LTCM crisis before the 2007-2009 crisis.
We focus on short sellers because they serve as important information
intermediaries in multiple dimensions. The asset pricing literature shows that short
sellers are informed about future stock returns and firm performances (Senchack and
Starks, 1993; Asquith, Pathak, and Ritter, 2005; Nagel, 2005; Pownall and Simko,
2005; Boehmer, Jones, and Zhang, 2008; Engelberg, Reed, and Riggenberg, 2012;
Kecskés, Mansi, and Zhang, 2012). In addition, Karpoff and Lou (2010) show that
short sellers can detect firms that misrepresent their financial statements as early as 19
months before the firms publicly disclose the information. Ho, Lin, and Lin (2016)
find that short selling reduces the information asymmetry and agency costs between
firms and banks, resulting in lower bank loan spreads for firms treated in the Reg
1 Furthermore, recent evidence indicates that poorly-performed banks includes those with CEOs who
had better incentives in terms of the dollar value of their stakes (Fahlenbrach and Stulz, 2011); banks
with more shareholder-friendly boards and more fragile financing (Beltratti and Stulz, 2012); banks
with lower-quality regulatory capital such as Tier 1 ratios (Berger and Bouwman, 2013); banks with
worse risk management mechanism (Ellul and Yerramilli, 2013); banks with more highly rated
tranches of securitization (Erel, Nadauld, and Stulz, 2014). Ho, Huang, Lin, and Yen (2016) find that
banks with overconfident chief executive officers (CEOs) tend to take more risk prior to the financial
crisis, thereby these banks experienced lower operating performance and stock returns in the crisis
period.
4
SHO PILOT program. Ljungqvist and Qian (2016) demonstrate that unlike the reports
of sell-side analysts, a vast majority of short sellers’ reports disclose new and hard
information.
Given the aforementioned evidence that short sellers are informed about the poor
performance and agency problems of industrial firms, we conjecture that some short
sellers might also be informed about the bleak outlook of some banks before the
occurrence of a financial crisis and capitalize on that private information. If this is the
case, we expect that short interest should increase in the pre-crisis periods and predicts
negatively the crisis returns of those banks. Accordingly, we propose our first
hypothesis that short sellers are informed about the poor performance of some banks
during financial crises, and thus pre-crisis short selling activities could predict bank
stock returns in the crisis periods.
Our empirical tests rely on the two recent banking crises: LTCM crisis and
2007-2009 financial crisis.2 In 1998, the Russian Default led to the collapse of the
hedge fund managed by LTCM with $5 billion of capital and $125 billion of debt.
Federal Reserve Bank of New York induced 14 large banks to provide $3.6 billion US
dollar to rescue the LTCM.3 Alan Greenspan said that “I’ve watching the US markets
for fifty years and I never seen anything like this.” Later on, Federal Reserve System
(FED) lowered interest rate three times in rapid succession between September 29 and
November 17, 1998.4 Similarly, the 2007-2009 financial crisis is considered as the
second largest crisis in history after the Great Depression of the 1930s. Financial
2 Following Fahlenbrach, Prilmeier, and Stulz (2012), the recent financial crisis refers to the period
from July 2007 to December 2009, and the 1998 LTCM crisis refers to the period from August 1998 to
December 1998. 3 Kindleberger, Charles P., and Robert Z. Aliber (2005). Manias, Panics and Crashes. Palgrave
Macmillan. 4 Greenspan, Alan (2008). The age of turbulence: Adventures in a new world. Penguin.
5
institutions like Lehman Brothers, Bear Stearns, Merrill Lynch, Fannie Mae, Freddie
Mac, Citigroup, and AIG went bankrupt. American families’ wealth fell totally by
$11 trillion in 2008, equal to the combined output of Germany, Japan, and the UK.5
The finance literature also pays an increasing attention to the role of culture in various
aspects among firms and banks (e.g., Hilary and Hui, 2009; McGuire, Omer, and
Sharp, 2012; Fahlenbrach, Prilmeier, and Stulz, 2012; Bereskin, Campbell, and Kedia,
2014; Guiso, Sapienza, and Zingales, 2015; Liu, 2016). In particular, in a speech
delivered on October 14, 2014, Williams C. Dudley, the president and CEO of the
New York Fed, pointed out that: “I will focus on how incentives could be improved
within the financial services industry to encourage better culture and behavior.”6
Dudley cautioned against an overly risky business model for banks that makes them
more vulnerable during the crises. However, from the short sellers’ perspective, the
weakness of banks is an opportunity for them to make sizable profits if they can
identify the banks with overly risky business models. Then, the natural question to ask
is how do short sellers identify the banks with weakness before crises? The persistent
bank risk culture or risky business models seem to be a good starting point.
Therefore, based on the argument in Fahlenbrach, Prilmeier, and Stulz (2012)
and our first hypothesis, we propose our second hypothesis that short sellers tend to
target the banks that had high-risk exposures in the LTCM crisis when they anticipate
an imminent financial crisis. Following Fahlenbrach, Prilmeier, and Stulz (2012), we
use banks’ stock returns during the 1998 LTCM crisis to represent the potential risk
exposure in the 2007-2009 financial crisis. If short sellers indeed target these banks
with overly risky business models, we should find that the LTCM crisis returns of
5 S. Mitra Kalita (2009). Americans See 18% of Wealth Vanish. The Wall Street Journal.
6 The full text of Dudley’s speech is on the website of the New York Fed (http://www.newyorkfed.org/
newsevents/speeches/2014/dud141020a.html).
6
banks predict negatively the change of short interest for these banks before the
2007-2009 financial crisis.
To test our two hypotheses, we collect the short interest data from the New York
Stock Exchange (NYSE), American Stock Exchange (AMEX), and NASDAQ. The
sample comprises 212 banks from 1998 LTCM crisis and 683 banks from the
2007-2009 financial crisis.7 Based on Karpoff and Lou (2010), we use the short
interest level two years before the crisis periods as the benchmark to calculate the
change in short interest. Our results show that the change of short interest in the
pre-crisis period is negatively correlated to the banks’ stock returns during both
banking crises. The result is robust to using one-year short interest change. The banks’
stock returns in the financial crisis, on average, decrease by 5.72% (4.84×1.1830) for a
one-standard-deviation increase in the pre-crisis change of short interest. For the
LTCM crisis, the stock returns decrease by 2.02% for a one-standard-deviation increase
in the pre-crisis change of short interest. These results provide supportive evidence to
our first hypothesis.8
In terms of economic magnitude, the 5.72% for short selling effect represents
approximately 70% of the risk culture effect in Fahlenbrach, Prilmeier, and Stulz
(2012). They find that a one-standard-deviation decrease in banks’ stock returns during
the LTCM crisis is associated with 8.2% lower stock returns during the 2007-2009
financial crisis. Our findings indicate that the predictability of short interest on banks’
stock returns during the crisis periods is both statistically significant and economically
meaningful.
7 We exclude banks without short selling data.
8 We find similar results when using characteristic-adjusted stock returns (Daniel, Grinblatt, Titman,
and Wermers (1997).
7
For our second hypothesis, we also find consistent evidence that short sellers
establish larger short positions before the 2007-2009 crisis on the banks that performed
poorly during the LTCM crisis. The result is in line with that of Fahlenbrach, Prilmeier,
and Stulz (2012) who show that past crisis performance can predict bank performance
in the next crisis. Our results not only provide a validity check to their claim on the
existence of bank persistent risk culture but also further indicates that short sellers are
able to comprehend the banks’ risk culture and target those with overly risky business
models before the forthcoming crisis. Overall, the evidence regarding our two
hypotheses provide convincing evidence that short sellers are informed about the
bleak outlook of some banks for the two financial crises and target those with a
persistent culture for risky business models.
We conduct several tests and robustness checks to corroborate our two
hypotheses. First, we repeat our main analysis for the non-financial firms to mitigate
the concern that our previous finding is just a re-documentation of return predictability
of short selling shown in the literature. The results indicate that the magnitude of return
predictability is much larger for the banking industry than that for the non-financial
industries. For example, a one-standard-deviation increase in pre-crisis short selling is
associated with 1.5% lower stock returns for the non-financial firms but 5.72% lower
stock returns for the banks during the financial crisis. Second, we randomly choose
pseudo-events for each bank in our sample to test whether the predictability of short
selling we documented is much weaker during non-crisis periods. This exercise is to
mitigate the concern that we just re-document the stock return predictability of short
interest for the bank sample in the two crisis periods. We find that predictability of short
selling in the financial crisis is around three times stronger than that in the pseudo-event
periods. Thus, this evidence supports our arguments that the stronger predictability of
8
short interest on banks’ stock returns during the crisis periods comes at least partially
from their private information on the excessive risk-taking business models.
Third, instead of using change of short interest, we construct three alternative
measures to capture realized short selling activities before crises. For example, we use
the two-stage least squares (2SLS) method by examining the determinants of short
interest in the first stage regression and use the abnormal short interest as the
independent variable for the second stage regression. We find consistent results that
pre-crisis change in abnormal short interest is negatively related to the banks’ stock
returns in the crisis periods. Fourth, we find that banks with larger pre-crisis change of
short interest have lower loan quality and higher default risk in the financial crisis.
These results provide economic channels through which short sellers can detect
worse-performing banks in the crisis periods.
Fifth, by dividing the banks into subsamples based on the level of risk-taking
behavior (i.e., leverage, short-term funding, tangible common equity ratio, and beta),
we find that the predictability of short interest for the banks’ stock returns during a
crisis is stronger in higher risk-taking subsamples. These results are in line with our
second hypothesis that short sellers mainly target on the banks with a high-risk culture
before the crisis. Sixth, we adopt a quantile regression analysis based on the returns
during the financial crisis to see whether our results concentrate on the
worst-performing banks. We find that the return predictability of the pre-crisis short
interest decreases with the return quintile. That is, the predictability is stronger for the
lower return quantiles than for the higher quantiles. This result is consistent with the
findings in Fahlenbrach, Prilmeier, and Stulz (2012) who show that the
9
worst-performing banks (bottom 20%) have the highest stock return correlation in the
two crises.
Finally, we use costs of borrowing stocks (equity lending fees) as an alternative
measure for informed short selling (Drechsler and Drechsler, 2016). Our results show
that the pre-crisis stock borrowing cost for banks is inversely correlated to their stock
returns in the financial crisis. This result supports our first hypothesis because the
borrowing costs would be higher if there is a high short selling demand for
poorly-performing banks. Further, we find consistent results when using (1) a different
benchmark period to calculate the change in short interest, (2) a different period to
define crisis stock returns, (3) for two bank size subsamples, and (4) for two industry
subsamples (commercial and investment banks vs. insurance). Collectively, these
additional tests and robustness checks substantiate our main findings.
Our paper contributes to the literature in three ways. First, we shed light on the
debate whether there are any market participants aware of the imminent financial crises.
The public criticizes economists because they were not capable of predicting the
crises and might even contribute to it.9 The US government regulators such as FED
and the Treasury also failed to detect financial bubbles. After retiring as FED
chairman, Alan Greenspan admitted that he did not anticipate the speculative bubble
in the mortgage lending market.10
Moreover, Fahlenbrach and Stulz (2011) find that
CEOs do not anticipate upcoming crises even though they are supposed to have more
private information. In contrast, our results indicate that some short sellers are informed
and establish short positions before the 2007-2009 financial crisis, particularly among
banks that performed rather poorly in the LTCM crisis period. In this regard, our paper
9 See the study of Colander, Goldberg, Haas, Juselius, Kirman, Lux, and Sloth (2009) for details.
10 Edmund L.Andrews, 2008. Greenspan Concedes Error on Regulation in The New York Times
(http://www.nytimes.com/2008/10/24/business/economy/24panel.html?_r=0).
10
is related to Hanley and Hoberg (2016) who analyze risk factors in bank 10-Ks based on
computational linguistics and find these factors predicting financial instability in 2008.
Second, our paper adds to the literature on corporate culture, with a focus on the
bank risk culture. Besides the aforementioned Fahlenbrach, Prilmeier, and Stulz
(2012), Ellul and Yerramilli (2013) find that banks with an aggressive risk culture are
associated with weaker risk management functions. Ho, Huang, Lin, and Yen (2016)
find that a bank’s risk culture reflects the character traits of its CEO. We complement
this line of research by showing that short sellers tend to target banks with an overly
risk-taking culture, as exhibited by poor stock performance in the previous crisis.
Third, our research also adds to the short selling literature. A plethora of asset
pricing studies show that short sellers are informed traders and their trading can
predict the future performances of industrial firms. But few studies explore the
interaction between short sellers and banks. One exception is Hasan, Massoud,
Saunders, and Song (2015) who find that banks with subprime assets have higher short
interest.11
Our paper differs from theirs in two ways. First, we demonstrate that short
sellers are able to identify the banks’ risk culture and target overly risk-taking banks
before the two crises. Second, we find that the predictability of short selling for the
financial crisis is around three times stronger than that for non-financial firms and that
in non-crisis periods, indicating that banks with excessive risk-taking models are also
the potential targets among the short sellers.
The paper is organized as follows. Section 2 discusses the hypotheses. Section 3
describes the data. Section 4 presents the empirical results. Section 5 presents the
additional supporting evidence, and Section 6 concludes.
11
The other one is Ho, Lin, and Lin (2016) who find that information asymmetry and agency costs
between firms and banks are reduced due to the relaxation of short-sale constraints, resulting in lower
bank loan spreads.
11
2. Hypothesis development
2.1. Informed short sellers and banking crises
The literature shows that short sellers are more likely to be informed than other
investors (Diamond and Verrecchia, 1987). Among others, Senchack and Starks
(1993), Asquith, Pathak, and Ritter (2005), Nagel (2005), and Boehmer, Jones, and
Zhang (2008) all show that an increase in short interest negatively predicts future
stock returns. Their findings indicate that short sellers possess private information and
reveal it to the market via their trading activities.
Another stream of short selling studies explores what types of information short
sellers have that enable their trading activities to predict stock returns. Karpoff and
Lou (2010) find that short interest goes up significantly 19 months prior to the initial
public revelation of a firm’s misrepresentation. They argue that short sellers can use
not only publicly available information (i.e., fundamental accounting) but also other
private information. Christophe, Ferri, and Hsieh (2010) find that short interest
predicts recommendation changes via analyst tipping. Further, Kecskés, Mansi, and
Zhang (2012) show that short interest also predicts the bond spreads.12
These studies indicate that short sellers are informed about various aspects of
firms. We thus conjecture that short sellers also pay attention to some banks that
might suffer a lot before an imminent financial crisis. The implication would be that
econometricians are able to observe a lead-lag correlation between change of short
interest and banks’ stock returns during the crisis periods. Accordingly, we propose
our first hypothesis:
12
However, Engelberg, Reed, and Ringgenberg (2012) argue that the information advantage of short
sellers mainly comes from their premium ability to process publicly available information.
12
Hypothesis 1: Short sellers are informed about the poor performance of some banks
during financial crises, and thus pre-crisis short selling activities could predict bank
stock returns in the crisis periods.
2.2. Informed short sellers and the persistent risk culture of banks
Several recent studies explore whether corporate culture affects a bank’s
performance. For example, Fahlenbrach, Prilmeier, and Stulz (2012) use banks’
performance in the LTCM crisis and the financial crisis to test two conflicting
hypotheses: the learning hypothesis and the risk culture hypothesis. If a bank learns
from the past experience, the correlation between banks’ stock returns in two crises
should be negative. However, if the risk culture or business model is persistent, a
bank with poor performance in the LTCM crisis would continue to perform relatively
poorly in the 2007-2009 financial crisis. Their results support the risk culture
hypothesis.
Ellul and Yerramilli (2013) study a sample of 74 US bank holding companies
and find that banks with an aggressive risk culture are associated with weaker risk
management. Cheng, Hong, and Scheinkman (2015) examine the relation between
executive compensation and several risk measures in banks. They find that more
excessive compensation is associated with more risk-taking behavior. Ho, Huang, Lin,
and Yen (2016) find that aggressive banks tend to hire overconfident managers who
are willing to take greater risks.
If banks’ risk culture is persistent, as suggested by Fahlenbrach, Prilmeier, and
Stulz (2012), the poorly-performing banks in the LTCM crisis would continue to
underperform in the 2007-2009 crisis. Thus, sensing an asset bubble in the banking
industry, it might be a good strategy for short sellers to establish short positions on the
13
banks that severely underperformed in the LTCM crisis. As a result, we expect a
negative correlation between the banks’ stock returns during the LTCM crisis period
and the change of short interest before the 2007-2009 financial crisis. We thus
propose our second hypothesis:
Hypothesis 2: Short sellers tend to target the banks that had high-risk exposures in
the LTCM crisis when they anticipate an imminent financial crisis.
3. Data
In this section, we provide information on our data sources and summary
statistics of the variables in interests.
3.1. Sample
Our sample comprises all of the financial institutions with SIC codes between
6000 and 6399. These institutions consist of four groups: depository institutions (SIC
6000-6099), non-depository credit institutions (SIC 6100-6199), investment
intermediaries (SIC 6200-6299), and insurance (SIC 6300-6399).13
Following
Fahlenbrach, Prilmeier, and Stulz (2012), the recent financial crisis refers to the
period from July 2007 to December 2009, while the 1998 LTCM crisis refers to the
period from August 1998 to December 1998. The pre-crisis period we use to analyze
the short selling activities refers to the 24-month period prior to the trigger events.
The trigger events occurred in August 1998 and July 2007 for LTCM crisis and
2007-2009 financial crisis, respectively.
We collect the short selling data from two main sources. First, we use the short
interest to measure the trading activities of short sellers. Short interest is the open short
13
For brevity, we use the term “banks” to represent “financial institutions” in this paper.
14
position in the NYSE, AMEX, or the NASDAQ. Second, we collect the borrowing cost
data of short selling from the Markit Data Explorer (DXL). Recent studies such as
Beneish, Lee, and Nichols (2015), Drechsler and Drechsler (2016), Engelberg, Reed,
and Ringgenberg (2014), and Chang, Lin, and Ma (2016) also use this database to
gauge the short selling market condition.14
Then, we match the short selling data with
the stock return data from the Center for Research in Security Prices (CRSP) and the
accounting data from Compustat. In total, the sample comprises 212 banks during the
LTCM crisis and 643 banks during the financial crisis.
We use the change of short interest as the primary independent variable, whereas
borrowing cost as the robustness check in Subsection 5.7. We scale the short interest
(SI) by the percentage of the total shares outstanding as in Asquith, Pathak, and Ritter
(2005). They argue that using outstanding stocks rather than the trading volume to
scale the short interest is more appropriate for testing whether short selling discloses
private information. Based on Karpoff and Lou (2010), we use the change of the short
interest 24 months before the crisis period as the benchmark measure (e.g,
1t tSI SI SI , where 1tSI is 24 months prior to the trigger event). We denote the
two primary independent variables as SI for the 2007-2009 financial crisis and
LTCMSI for the LTCM crisis.
The borrowing cost is the Daily Cost of Borrow Score (DCBS) in the DXL. The
DCBS is a cost index that ranges from one (cheapest) to ten (most expensive) that the
DXL assigns to every stock. Similar to the change in the short interest, the change in
borrowing costs is calculated as 1t tCost DCBS DCBS where
1tDCBS is the
DCBS 24 months prior to the trigger events.
14
The DXL consists of data from more than 100 institutional lenders that cover more than 90% of the
US markets’ capitalization (Beneish, Lee, and Nichols, 2015).
15
The dependent variables that capture the banks’ crisis performance are RE09 (the
annualized buy-and-hold returns from July 1, 2007 through December 31, 2009),
RE08 (the annualized buy-and-hold returns from July 1, 2007 through December 31,
2009), RE98 (the annualized buy-and-hold returns from August 3, 1998 until the day
in 1998 on which the bank’s stock attains its lowest price), ∆EDF (the change in
expected default frequency (EDF) between crisis years (2007-2009) and year 2006),
and ∆NPL (change in the ratio of nonperforming loans (NPL) to total gross loans
between crisis years (2007-2009) and year 2006).15
The controls variables are BHAR06 (the buy-and-hold returns from July 1, 2006,
through June 30, 2007), LnAssets (log of total assets on December 31, 2006), BM
(book value of common equity divided by market value of common equity on
December 31, 2006), Leverage (ratio of assets to book value of equity on December
31, 2006), TCE ratio (tangible common equity ratio: tangible common equity divided
by tangible assets and multiplied by 100), Beta (banks’ equity beta from a market
model of daily returns in excess of three-month T-bills from January 2004 to
December 2006, where the market is represented by the value-weighted CRSP index),
Idiosyncratic volatility (IDIORISK, standard deviations of the residuals obtained from
a market model of daily returns in excess of three-month T-bills from January 2004 to
December 2006), MES (%) (marginal expected shortfall as defined in Acharya,
Pedersen, Philippon, and Richardson, 2010), measured using the 5% worst days for
the value-weighted CRSP market return during 2004–2006).
3.2. Descriptive statistics
15
The EDF is the percentile ranking of a firm’s default risk based on its distance to default (Bharath
and Shumway, 2008).
16
Table 1 presents the summary statistics which include mean, standard deviation,
and quartiles. The table shows that the banks’ stock returns are quite negative during
the financial crisis. The annualized buy-and-hold returns for the periods of July 2007
to December 2008 and July 2007 to December 2009 (RE08 and RE09) are -29.7% and
-24.5%, respectively. In the LTCM crisis, the annualized buy-and-hold return (RE98)
is -82.29% on average. The average ∆EDF and ∆NPL are 0.28 and 2.53%,
respectively.
In particular, there is a considerable increase in the short interest. One and
two-year prior to the financial crisis, the changes of short interest (∆SI12m and ∆SI) are
about 1.24% and 1.98% on average. Likewise, there is an increase of 1.02% in short
interest in the corresponding period of the LTCM crisis (∆LTCMSI). The stock
borrowing costs are greater in the pre-crisis period as well.
[Insert Table 1]
Next, following the study of Fahlenbrach, Prilmeier, and Stulz (2012), we split
our sample into two groups based on the stock returns from July 2007 to December
2009 (i.e., RE09): Bottom Quintile (banks in the lowest RE09 quintile) and Other
Quintiles (other banks).
Table 2 presents the comparisons of the variables between the two groups. The
change of short interest of the bottom performers is 3.1% in the 24-month pre-crisis
period, while the change of short interest for the other banks is only 1.7%. The t-test
of the difference is statistically significant at the 1% level. This result suggests that
some short sellers have established larger short positions for the worst-performing
banks before the crisis.
[Insert Table 2]
17
4. Empirical results
In this section, we first test the relation between pre-crisis change of the short
interest and the banks’ stock returns in the two crises. We then examine whether short
sellers target the banks that performed poorly in the LTCM crisis.
4.1. Short selling and stock returns in crises
We use the following ordinary least-squares (OLS) regression to investigate
whether change of the short interest can predict the bank’s stock returns during a
crisis:
,crisis , ,t 109 i i pre crisis i iRE SI Z (01)
,crisis , ,t 198 i i pre crisis i iRE LTCMSI Z (02)
where RE09i,crisis and RE98i,crisis represent stock returns for bank i in the 2007-2009
financial crisis and the LTCM crisis, respectively; ,i pre crisisSI and
,i pre crisisLTCMSI are the changes of the short interest for bank i in the pre-crisis
periods of the two crises, respectively; and , 1i tZ is a vector of control variables for
bank i in year 2006 and year 1997, which are the last full fiscal years prior to the two
crises. The definitions of these control variables are presented in Appendix. In all
regressions, we report the t-values based on the standard errors adjusted for
heteroskedasticity (White, 1980) and industry clustering (Petersen, 2009).
Our first hypothesis argues that short sellers anticipate the imminent banking
crises and establish large short positions for some banks in the pre-crisis periods. The
banks that are heavily shorted should perform worse in the crisis periods. We thus
expect the signs of coefficients in Equations (0)(1) and (2) to be negative.
18
We control a number of important bank characteristics: BHAR06, LnAssets, BM,
Beta, Leverage, TCE, MES, and IDIORISK, as those in Fahlenbrach, Prilmeier, and
Stulz (2012). Table 3 presents the regression results. The first specification controls
for BHAR06, LnAssets, BM, Beta, and Leverage. The second specification substitutes
TCE for Leverage. The third specification controls for Leverage, MES, and IDIORISK
but omits the TCE. The fourth specification contains all of the control variables.
Across all specifications, the coefficients of ∆SI are significantly negative. For
example, in Model (1), the coefficient of ∆SI is -0.9982 and statistically significant. A
one-standard-deviation increase in the ∆SI is associated with a 4.83% (4.84×0.9982)
lower stock return during the 2007-2009 financial crisis. After controlling for more
bank characteristics in Models (2) to (4), the economic magnitude of the coefficients
becomes even larger. For example, in Model (4), the stock returns decrease by 5.72%
(4.84×1.1830) during the crisis for a one-standard-deviation increase in the pre-crisis
change of short interest.16
Furthermore, this 5.72% is approximately 70% of the risk culture effect in
Fahlenbrach, Prilmeier, and Stulz (2012), where they find that a one-standard-
deviation lower return during the LTCM crisis is associated with an 8.2% lower return
during the financial crisis. Thus, the predictability of short interest for the banks’ stock
returns during the financial crisis is both statistically significant and economically
meaningful.
[Insert Table 3]
16
We find similar results when we use 125 (5×5×5) or 27 (3×3×3) characteristic-adjusted returns in
Equation (1) based on Daniel, Grinblatt, Titman, and Wermers (1997). The results are available upon
request.
19
Next, we test the stock return predictability using LTCM crisis data (Equation
(2)). The control variables, BHAR97, LnAssets1997, BM97, Leverage97, and
TCE1997 ratio are measured on December 31, 1997. The Beta1997, IDIORISK1997,
and MES1997 are estimated using the 1995–1997 period while the BHAR1997 is the
buy-and-hold returns from July 1, 1997, through June 30, 1998.
Table 4 presents the results, which are consistent with those in Table 3. In all
models, there is a significantly negative correlation between LTCMSI and RE98. In
terms of economic significance, Model (4) shows that a one-standard-deviation
increase in LTCMSI is associated with a 2.02% (3.84×0.5268) lower stock return
during the LTCM crisis.
[Insert Table 4]
In sum, the results in both Tables 3 and 4 provide supportive evidence to our first
hypothesis that there is a negative correlation between the change of short interest
prior to the crises and the banks’ stock performances during the crises. That is, short
sellers are informed that some banks are going to be in trouble, and they try to profit
from it.
4.2. Short selling and banks’ risk culture
To test our second hypothesis, we perform the following regression:
, crisis , ,t 1 98i pre i crisis i iSI RE Z (03)
where , crisisi preSI is the change in short interest for bank i in the pre-crisis period of
the financial crisis, while RE98i,crisis is the stock returns of bank i during the LTCM
crisis; and , 1i tZ is a vector of control variables in the year 2004. The Beta2004,
IDIORISK2004, and MES2004 are estimated using the 2002–2004 period while
20
BHAR2004 is the buy-and-hold returns from July 1, 2004, through June 30, 2005. The
other control variables are measured on December 31, 2004.
Our second hypothesis states that short sellers target the banks that severely
underperformed during the LTCM crisis before the 2007-2009 financial crisis. These
banks are the obvious targets of short sellers as long as the culture of taking excessive
risks does not change. We thus expect a negative sign for the coefficient in
Equation (0)(3).
Table 5 presents the results. Across all four specifications, we can observe a
negative correlation between RE98 and ∆SI. In Model (4), a one-standard-deviation
decrease in the bank stock return during the LTCM crisis is associated with a 1.64%
(18.55×0.0884) higher short interest in the pre-crisis period of the financial crisis.
[Insert Table 5]
These results not only support our second hypothesis but also provide a validity
check to the risk culture hypothesis in Fahlenbrach, Prilmeier, and Stulz (2012). They
argue that banks with a persistent risk culture and that performed poorly in the LTCM
crisis do not learn from that experience. These banks continue to take on higher
leverage, riskier funding, and higher asset growth than the other banks. Consequently,
these banks continue to underperform in the next crisis. Therefore, for short sellers
who predict that the asset bubble might be bringing down the banking industry, they
could target these banks, resulting in the documented negative correlation between the
returns during the LTCM crisis and the ∆SI before the financial crisis.
21
5. Additional supporting evidence
In this section, we provide further supporting evidence that aligns our findings to
the two proposed hypotheses. First, we repeat the analysis for non-financial firms.
Second, we compare the predictabilities of short interest on bank stock returns during
the crisis periods and pseudo-event periods. These two exercises help to rule out the
concern that we just re-document the stock return predictability of short sales. Third,
we adopt alternative measures of abnormal short interest and re-do the main analysis.
Fourth, we investigate whether the change of short interest also predicts alternative
measures of bank performance in the financial crisis. Fifth, we conduct a subsample
analysis for our first hypothesis by dividing the sample into high/low risk-taking banks.
Sixth, we adopt a quantile regression analysis to see whether the predictability of short
interest for bank returns concentrates on the lower performing quantiles. Seventh, we
use the borrowing cost of short selling as an alternative measure for informed short
selling. Finally, we repeat the main analysis by using alternative time periods of short
interest and crisis returns as well as using bank size/industries subsamples for
establishing the robustness.
5.1 Returns predictability for non-financial firms is much lower in crises
In this subsection, we repeat our analysis for the non-financial sample. We collect
all firms with available short selling data before the financial crisis, after excluding the
financial industry. Then, we merge the short selling data with stock returns and
accounting data. We perform a regression as specified in Equation (1) and present the
results in Table 6.
In all specifications, we find a negative relation between RE09 and ∆SI. This result
is consistent with the existing literature that short sellers are informed, and short
22
interest can predict stock returns. However, the economic significance for the
non-financial sample is much smaller than that for bank sample during the crisis
periods. For example, in Model (4), a one-standard-deviation increase in pre-crisis
short interest is associated with about 1.5% lower stock returns for non-financial firms
in the financial crisis, while stock return decreases about 5.72% for banks (in Table 3).
Thus, short selling predictability is much stronger for banks than that for non-financial
industries.
[Insert Table 6]
5.2. Pseudo-events predictability of short selling is much lower
Following the approach in Chan, Ge, and Lin (2015), we conduct a simulation to
select pseudo-events for banks in non-crisis periods. For each bank in our sample of the
2007-2009 financial crisis, we randomly choose a month in non-crisis periods from
1990 to 2014 and consider that month is the actual trigger event of the financial crisis.
We then regress the annualized buy-and-hold stock returns of 30-month pseudo-events
(from month t to month t+29, which matches the duration of 2007-2009 crisis) on
change in short interest of 24-month before the pseudo-events (from month t-24 to
month t-1). We control for the same pre-year bank characteristics as those in Table 3.
We repeat the process for 1,000 times and report the average coefficient for the change
of short interest (∆SI) and its associated p-value. We compute the p-value for the
change of short interest as the fraction of the number of times that the coefficient of the
simulated sample is much larger than that of the actual sample (in Table 3). We present
the results of this simulation exercise in Table 7. For easier comparison, we also report
the coefficients of the actual sample in the first row of the table.
23
The negative sign of the average coefficient from 1,000 simulations indicates that
short sellers are in general informed about future banks’ stock returns, which is again
consistent with the short selling literature. However, the coefficients of the actual bank
sample are around three times larger in magnitude than those in the simulations. In
addition, the small p-values indicate that there are very few simulations in which the
short sellers have stronger predictability for banks’ stock returns in the non-crisis
periods. Like results in Section 5.1, these findings indicate that the predictability of
short interest for banks’ stock returns is much stronger in the financial crisis,
supporting our argument that short sellers have incremental information regarding
bank stock performance in the crisis. As we argued in our second hypothesis, this
incremental information is likely to be rooted in the persistent bank risk culture and
business models.
[Insert Table 7]
5.3. Alternative measures of short interest yield consistent results
In this subsection, we construct several alternative measures of abnormal short
interests and test whether our findings are robust to these measures.
Following Dechow, Hutton, Meulbroek, and Sloan (2001), Asquith, Pathak, and
Ritter (2005), and Karpoff and Lou (2010), we construct the first measure of abnormal
short interest ABSI(1) that adjusts for size (market value of equity), book-to-market
ratio (BM), and momentum (the prior year return of the stock). For each month, each
stock is assigned to one of 27 portfolios which are constructed by sorting stocks based
on size, book-to-market ratio, and momentum. We run a first-stage regression as
follows:
24
1 2 3 4
5 61
(4)
it it it it it
K
it it itk iktk
SI LowSize MedSize LowBM MedBM
LowMom MedMom Ind u
SIit is the number of shares shorted divided by the shares outstanding. The first six
explanatory variables are used to jointly define 27 portfolios based on size,
book-to-market, and momentum. For example, if bank i is assigned to the portfolio with
lowest market value in month t, then LowSizeit = 1 and MedSizeit = 0. Indikt are industry
dummies based on first two-digit SIC code. If bank i belongs to industry k, then Indikt =
1 and otherwise is 0. After running the regression, abnormal short interest is defined as
the difference between raw short interest and fitted short interest:
( ) ( ) , 1,2 it it itABSI j SI SI j j
(04)
where SIit is raw short interest and ( )itSI j
is fitted short interest from the above
regression.
With the same process, we construct our second abnormal short interest measure,
ABSI(2), from 243 size-, BM-, momentum-, share turnover-, and institutional
ownership- based portfolios with share turnover (i.e., share trading volume over
number of share outstanding) and institutional ownership (i.e., number of share owned
by institutional investors divided by number of share outstanding) as additional
controls.
Accordingly, we have two measures of abnormal short interest: ABSI(1) and
ABSI(2). Then, we define changes in abnormal short interest in the 24-month period
before the financial crisis: , , 1( ) ( ) ( )i i t i tABSI j ABSI j ABSI j . We then use these two
measures as our main explanatory variables to re-perform our main analysis in the 2nd
stage regression as follows:
,crisis , ,t 109 ( ) , 1,2i i pre crisis i iRE ABSI j Z j (6)
25
Table 8 presents the regression results. Consistently, we find a negative
correlation between change of abnormal short interest and crisis returns, supporting our
first hypothesis that short sellers have private information regarding the imminent
financial crisis.
[Insert Table 8]
We construct a third measure of abnormal short interest based on a 2SLS
approach. We first regress the change of short interest on the risk-taking measures and
bank characteristics in the following equation:
, , 1 i pre crisis ii tSI Z (7)
where , 1i tZ is a vector of control variables in the year 2004, including BHAR04,
LnAssets04, BM04, Leverage04, TCE04, Beta04, MES04, and IDIORISK04.
Then, our third measure of abnormal short interest is calculated as:
,,(3) i pre crisisi pre crisisi SISIABSI
, where ,i pre crisisSI
is fitted value that
derived from the 1st-stage regression in Equation (7). We then replace ,i pre crisisSI
with (3)iABSI into the regression Equation (1) as follows:
,crisis , 1,09 ABSI(3)i i t ii pre crisisRE Z (8)
where , 1i tZ is a vector of control variables in the year 2006.
Table 9 presents the regression results and shows that the negative correlation
between abnormal short interest and crisis returns remains.
[Insert Table 9]
5.4. Short selling predicts loan quality and default risk
Ho, Huang, Lin, and Yen (2016) find that risky banks (banks with overconfident
CEOs) suffer more in terms of more non-performing loans (NPL) and higher expected
26
default frequency (EDF) during the financial crisis. Thus, in this subsection, we
examine whether the pre-crisis change of short interest also predicts these operating
performance measures of banks during the 2007-2009 financial crisis as additional
supportive evidence for our first hypothesis. We perform the following regression:
,crisis , ,t 1 i i pre crisis i iNPL SI Z (9)
,crisis , ,t 1 i i pre crisis i iEDF SI Z (10)
where ∆NPLi,crisis and ∆EDFi,crisis represent the change in the NPL ratio and the
change in EDF for bank i in the financial crisis, respectively; ,i pre crisisSI is the
change in short interest for bank i in the pre-crisis period of the financial crisis as in
Equation (1); and , 1i tZ is a vector of control variables for bank i in year 2006. We
expect the signs of the coefficients in Equations (9) and (10) to be positive.
Table 10 presents the results. First, there is a significantly positive correlation
between pre-crisis short interest and the NPL ratio during the financial crisis. For
example, in Model (4), a one-standard-deviation increase in the pre-crisis short
interest is associated with a 0.49% (4.84×0.1023) increase in the NPL ratio, indicating
that short sellers can identify banks that have poor loan quality.
[Insert Table 10]
Second, the coefficients of ∆SI are also significantly positive across columns (5)
to (8). Banks with a higher change of short interest prior to the financial crisis are
more likely to default during the crisis. In terms of economic significance, the EDF
increases by 0.045 (4.84×0.0094) when there is a one-standard-deviation increase in
the change in short interest. These results provide consistent evidence to our first
hypothesis that short sellers are informed about the poor performance of some banks
27
before the financial crisis. These results also provide potential channels in what way
the targeted banks will underperform in the financial crisis.
5.5. Return predictability concentrates on the risky subsample
Fahlenbrach, Prilmeier, and Stulz (2012) show that worse-performing banks in
the LTCM crisis tend to have greater asset growth, more reliance on short-term
funding, and higher leverage in year 2006. Therefore, we should observe that short
sellers target these banks before the financial crisis. We thus expect that our previous
findings in Equation (1) should be stronger for the banks with higher Leverage, TCE,
and Beta. In each risk-taking proxy, we divide the sample into three groups: low,
medium, and high-risk.
We present the results in Table 11. As expected, the predictability of short
interest indeed concentrates on the banks at a higher risk-taking level (i.e., medium
and high-risk groups). The ∆SI is inversely correlated to RE09 in high-risk groups for
Leverage (medium and high), TCE (medium and high), and Beta (medium and high).
In contrast, the coefficients of ∆SI are negative but insignificant in the low-risk groups.
This evidence supports our argument that short sellers mainly target the banks with
high risk-taking business models.
[Insert Table 11]
5.6. Quantile regression: Predictability is stronger for worse-performing banks
Fahlenbrach, Prilmeier, and Stulz (2012) find that bottom quantile (banks in the
lowest RE09 quintile) has the strongest stock return correlations between the two
crises, compared with the other quantiles. In this subsection, we adopt a quantile
regression analysis to examine whether the stock return predictability of short interest
is the stronger in lower quantile banks. The quantile regression framework overcomes
28
several disadvantages of the standard linear regression such as a partial relation based
on a conditional mean function, sensitivity to outliers, and restrictive assumptions.
Thus, we use it as a methodological robustness check. We re-estimate the correlation
between the changes of short interest and the bank crisis returns as specified in
Equation (1). We use a full set of control variables as in Model (4) of Table 3.
Table 12 presents the estimated coefficients, and we also report the OLS
estimation from Model 4 in Table 3 in the last column for easier comparison. In the
lower quantiles such as 0.2 and 0.4, there is a significantly negative correlation
between ∆SI and the annualized buy-and-hold returns in the financial crisis. For
example, at quantile 0.2 , a one-standard-deviation increase in ∆SI is associated
with a 6.38% (4.84×1.3181) lower return during the financial crisis. However, in the
higher quintiles, the effect becomes weaker. At quantile 0.8 , the coefficient for
∆SI remains negative but statistically insignificant.
[Insert Table 12]
To sum up, these findings indicate that the crisis return predictability of short
interest is indeed stronger for worse-performing banks.
5.7. Using borrowing costs of short selling yield similar results
In this subsection, we use stock borrowing cost as an alternative measure for
informed short selling. We perform the following regression to test whether our main
results hold when replacing the change of short interest with the change of borrowing
costs:
, ,t 1,crisis09 i pre crisis iiiRE COST Z (11)
29
where RE09i,crisis represents the stock returns for bank i in the financial crisis;
,i pre crisisCOST is the change of stock borrowing costs for bank i in the pre-crisis
period; and , 1i tZ is a vector of control variables for bank i in the year 2006. We
expect the sign of the coefficient in Equation (11) to be negative.
Table 13 presents the regression results. The coefficients of COST are
significantly negative in all specifications, consistent with our first hypothesis. The
economic magnitude is also meaningful. For example, in Model (4), a
one-standard-deviation increase in COST is associated with a 4.24%
(0.904×4.6930) lower return during the financial crisis.
[Insert Table 13]
5.8. Other robustness checks provide consistent results
This subsection provides several other robustness checks for our main results.
First, we adopt alternative time periods to calculate the crisis returns and the changes of
short interest. This is to check the sensitivity of the regression results in Equation (1).
Following Fahlenbrach, Prilmeier, and Stulz (2012), the definition of crisis period for
our main results in Table 3 is from July 2007 to December 2009, while the change in
short interest in the pre-crisis period is calculated from June 2005 to June 2007. Here,
we consider the crisis period as July 2007 to December 2008 for the dependent variable
(RE08) and the change of short interest from June 2006 to June 2007 for the
independent variable (12mSI ). We then consider three different model specifications:
Model (1) includes the alternative crisis return (RE08) and the original change of short
interest ∆SI; Model (2) includes the original crisis return RE09 and the alternative
change of short interest (∆SI12m); and Model (3) includes the alternative crisis return
(RE08) and the alternative change of short interest (∆SI12m). In all specifications, we
30
control for a full set of bank characteristics. Models (1) to (3) of Table 14 report the
results. In all models, the change of the short interest predicts the crisis returns of the
banks. This is line with our previous findings and indicates that our main results are
robust to the definitions of crisis and pre-crisis periods.
Second, we redo the analysis for two subsamples based on market capitalization:
small banks (smaller than median) and large banks (higher than median) in Models (4)
and (5). The results show that short selling predictability are robust across bank size.
The coefficients of ∆SI are significantly negative in both subsamples but are larger for
small banks. This evidence is consistent with the literature that return predictability is
larger among smaller firms.
Third, to test if our results are robust among financial industries, we repeat the
analysis with two industry subsamples: (1) Commercial and investment banks, and (2)
Insurance in Models (6) and (7) of Table 14. The coefficients of SI are statistically
significant and negative in both industry subsamples.
[Insert Table 14]
Collectively, these results provide evidence that our results are not driven by a
particular definition of a crisis period, change of short interest, bank size, or a
particular type of financial institution.
6. Conclusion
The existing studies show that short sellers are informed, and their trading can
predict various aspects of firm performance. Our study sheds light to this literature by
focusing on whether the change of short interest before a crisis predicts the banks’
stock returns during the crisis. More intriguingly, we further explore whether short
31
sellers target the banks that performed worse in the previous LTCM crisis, which
indicates banks excessive risk-taking culture could serve as a red flag to the short
sellers.
Our results provide convincing evidence that there is a negative correlation
between the change of short interest before the crisis and the bank stock performance
during the two crises. We also find that before the 2007-2009 financial crisis, the short
selling concentrates on the banks that performed relatively poorly in the LTCM crisis.
This evidence not only provides the validity check to the finding in Fahlenbrach,
Prilmeier, and Stulz, (2012) who argue that there is a persistent risk culture among
banks, but also indicates that the very culture of taking overly high risks makes these
banks the targets of short sellers before the 2007-2009 financial crisis.
We provide a set of robustness checks to support our main findings. First, short
sellers’ predictability is stronger for banks than non-financial industries. Second, short
sellers’ predictability is stronger in the financial crisis than in non-crisis
pseudo-periods. Third, our main results are robust to alternative constructions of short
interests. Fourth, banks that are shorted more before the crisis have lower loan quality
and higher default risk in the financial crisis. Fifth, the crisis return predictability of
short interest is stronger among the riskier banks. Sixth, our quantile regression
analysis indicates that the crisis return predictability of short interest is stronger for the
worse-performing banks. Seventh, we find similar results when using the borrowing
costs faced by short sellers as the informed short selling measure. Finally, our results
are robust to using alternative definitions of pre-crisis and crisis periods, subsamples
of bank size, and subsamples of financial industries.
32
Collectively, these results provide strong evidence that short selling predicts the
performance of banks in the crisis periods, and that short sellers are able to identify
the banks with a persistent culture of high-risk business models.
33
Appendix Variable definitions
Variable Definition Data Source
Panel A: Short selling variables
SI Change in total number of stocks that are borrowed divided by
the stocks outstanding from June 2005 through June 2007.
NYSE, AMEX,
NASDAQ
12mSI Change in total number of stocks that are borrowed divided by
the stocks outstanding from June 2006 through June 2007.
NYSE, AMEX,
NASDAQ
LTCMSI Change in total number of stocks that are borrowed divided by
the stocks outstanding from August 1996 through July 1998.
NYSE, AMEX,
NASDAQ
COST Change in stock borrowing costs (Daily Cost of Borrow
Score—a relative measure of borrowing costs, constructed by
DXL. It ranges from 1- cheap to borrow- to 10- expensive to
borrow) from June 2005 through June 2007.
DXL
Panel B: Crisis performance variables
RE09 The annualized buy-and-hold returns from July 1, 2007 through
December 31, 2009.
CRSP
RE08 The annualized buy-and-hold returns from July 1, 2007 through
December 31, 2008.
CRSP
RE98 Following (Fahlenbrach et al., 2012), RE98 is the annualized
buy-and-hold returns from August 3, 1998, until the day in 1998
on which the bank’s stock attains its lowest price. If the lowest
price occurs more than once, then the return is calculated using
the first date on which it occurs.
CRSP
∆EDF Change in expected default frequency (EDF) between crisis years
(2007-2009) and year 2006. The EDF is the percentile ranking of
a firm’s default risk based on its distance to default (constructed
from Bharath and Shumway, 2008).
Compustat and
CRSP
∆NPL Change in ratio of nonperforming loans (NPL) to total gross
loans between crisis years (2007-2009) and year 2006.
Nonperforming loans are defined as loans with interest payments
and principal more than 90 days overdue.
Compustat
Panel C: Bank characteristics
BHAR06 The buy-and-hold returns from July 1, 2006, through June 30,
2007.
CRSP
LnAssets Log of total assets (U.S. billion) on December 31, 2006. Compustat
BM Book value of common equity divided by market value of
common equity on December 31, 2006.
Compustat and
CRSP
Leverage Ratio of assets to book value of equity on December 31, 2006. Compustat
TCE ratio Tangible common equity ratio: tangible common equity divided
by tangible assets and multiplied by 100.
Compustat
Beta Bank’s equity beta from a market model of daily returns in
excess of three-month T-bills from January 2004 to December
2006, where the market is represented by the value-weighted
CRSP index.
CRSP
Idiosyncratic
volatility
Standard deviation of the residuals obtained from a market
model of daily returns in excess of three-month T-bills from
January 2004 to December 2006, where the market is
CRSP
34
(IDIORISK) represented by the value-weighted CRSP index.
MES (%) Marginal expected shortfall as defined in Acharya, Pedersen,
Philippon, and Richardson (2010), measured using the 5% worst
days for the value-weighted CRSP market return during 2004–
2006.
CRSP
35
References
Acharya, V. V., Pedersen, L. H., Philippon, T., & Richardson, M. (2010). Measuring
Systemic Risk. Unpublished Working Paper, New York University.
Asquith, P., Pathak, P. A., & Ritter, J. R. (2005). Short interest, institutional
ownership, and stock returns. Journal of Financial Economics, 78, 243–276.
Beltratti, A., & Stulz, R. M. (2012). The credit crisis around the globe: Why did some
banks perform better? Journal of Financial Economics, 105, 1–17.
Beneish, M. D., Lee, C. M. C., & Nichols, D. C. (2015). In short supply: Short-sellers
and stock returns. Journal of Accounting and Economics, 60, 33–57.
Bereskin, F. L., Campbell, T. L., & Kedia, S. (2014). Philanthropy, corporate culture
and misconduct. Available at SSRN 2370482.
Berger, A. N., & Bouwman, C. H. (2013). How does capital affect bank performance
during financial crises? Journal of Financial Economics, 109, 146–176.
Berger, A. N., & Hannan, T. (1998). The efficiency cost of market power in the
banking industry: A test of the quiet life and related hypothesis. Review of
Economics and Statistics 80, 454–465.
Biddle, G. C., & Hilary, G. (2006). Accounting quality and firm‐level capital
investment. Accounting Review 81, 963–982.
Biorn, E. (2000). Panel data with measurement errors: Instrumental variables and
GMM procedures combining levels and differences. Econometric Reviews 19,
391–424.
Bharath, S. T., & Shumway, T. (2008). Forecasting default with the Merton distance
to default model. Review of Financial Studies, 21, 1339–1369.
Boehmer, E., Jones, C. M., & Zhang, X. (2008). Which Shorts Are Informed? The
Journal of Finance, 63, 491–527.
Chan, K., Ge, L., & Lin, T.-C. (2015). Informational Content of Options Trading on
Acquirer Announcement Return. Journal of Financial and Quantitative
Analysis, 50, 1057–1082.
Chang, E. C., Lin, T. C., & Ma, X. (2016a). Does short-selling threat discipline
managers in M&A decisions? Unpublished Working Paper. University of Hong
Kong.
Cheng, I. H., Hong, H., & Scheinkman, J. A. (2015). Yesterday’s Heroes:
Compensation and Risk at Financial Firms. The Journal of Finance, 70, 839–
879.
Christophe, S. E., Ferri, M. G., & Hsieh, J. (2010). Informed trading before analyst
downgrades: Evidence from short sellers. Journal of Financial Economics, 95,
85–106.
36
Colander, D., Goldberg, M., Haas, A., Juselius, K., Kirman, A., Lux, T., & Sloth, B.
(2009). The financial crisis and the systemic failure of the economics
profession. Critical Review, 21, 249–267.
Daniel, K., Grinblatt, M., Titman, S., & Wermers, R. (1997). Measuring Mutual Fund
Performance with Characteristic-Based Benchmarks. The Journal of Finance,
52(3), 1035–1058.
Dechow, P. M., Hutton, A. P., Meulbroek, L., & Sloan, R. G. (2001). Short-sellers,
fundamental analysis, and stock returns. Journal of Financial Economics, 61,
77–106.
Diamond, D. W., & Verrecchia, R. E. (1987). Constraints on short-selling and asset
price adjustment to private information. Journal of Financial Economics, 18,
277–311.
Drechsler, I., & Drechsler, Q. (2016), The Shorting Premium and Asset Pricing
Anomalies, Unpublished Working Paper. New York University.
Ellul, A., & Yerramilli, V. (2013). Stronger risk controls, lower risk: Evidence from
US bank holding companies. The Journal of Finance, 68, 1757–1803.
Engelberg, J. E., Reed, A. V., & Ringgenberg, M. C. (2012). How are shorts
informed?: Short sellers, news, and information processing. Journal of
Financial Economics, 105, 260–278.
Erel, I., Nadauld, T., & Stulz, R. M. (2014). Why Did Holdings of Highly Rated
Securitization Tranches Differ So Much across Banks? Review of Financial
Studies, 27, 404–453.
Fahlenbrach, R., Prilmeier, R., & Stulz, R. M. (2012). This Time Is the Same: Using
Bank Performance in 1998 to Explain Bank Performance during the Recent
Financial Crisis. The Journal of Finance, 67, 2139–2185.
Fahlenbrach, R., & Stulz, R. M. (2011). Bank CEO incentives and the credit crisis.
Journal of Financial Economics, 99, 11–26.
Griliches, Z., & Hausman, J. A. (1986). Errors in variables in panel data. Journal of
Econometrics 31, 93–118.
Guiso, L., Sapienza, P., & Zingales, L. (2015). The value of corporate culture.
Journal of Financial Economics, 117, 60–76.
Hanley, K.W., & Hoberg, G. (2016). Dynamic interpretation of emerging systemic
risks. Unpublished Working Paper. University of Southern California.
Hasan, I., Massoud, N., Saunders, A., & Song, K. (2015). Which financial stocks did
short sellers target in the subprime crisis? Journal of Banking & Finance, 54,
87–103.
Hilary, G., & Hui, K. W. (2009). Does religion matter in corporate decision making in
America? Journal of Financial Economics, 93, 455–473.
37
Ho, P.-H., Huang, C.-W., Lin, C.-Y., & Yen, J.-F. (2016). CEO overconfidence and
financial crisis: Evidence from bank lending and leverage. Journal of
Financial Economics, 120, 194–209.
Ho, P.-H., Lin, C.-Y., & Lin, T.-C. (2016). Equity Short Selling and Bank Loan
Market: A Controlled Experiment. Unpublished Working Paper. University of
Hong Kong.
Karpoff, J. M., & Lou, X. (2010). Short Sellers and Financial Misconduct. The
Journal of Finance, 65, 1879–1913.
Kecskés, A., Mansi, S. A., & Zhang, A. J. (2012). Are short sellers informed?
Evidence from the bond market. The Accounting Review, 88, 611–639.
Liu, X. (2016). Corruption culture and corporate misconduct. Journal of Financial
Economics, 122, 307–327.
Ljungqvist, A., & Qian, W. (2016). How Constraining Are Limits to Arbitrage?
Evidence from a Recent Financial Innovation. Forthcoming in Review of
Financial Studies.
McGuire, S. T., Omer, T. C., & Sharp, N. Y. (2011). The impact of religion on
financial reporting irregularities. The Accounting Review, 87, 645–673.
Nagel, S. (2005). Short sales, institutional investors and the cross-section of stock
returns. Journal of Financial Economics, 78, 277–309.
Petersen, M. A. (2009). Estimating standard errors in finance panel data sets:
Comparing approaches. Review of Financial Studies, 22, 435–480.
Pownall, G., & Simko, P. J. (2005). The information intermediary role of short sellers.
The Accounting Review, 80, 941–966.
Rajgopal, S., & Shevlin, T. (2002). Empirical evidence on the relation between stock
option compensation and risk taking. Journal of Accounting and Economics 33,
145–171.
Senchack, A. J., & Starks, L. T. (1993). Short-Sale Restrictions and Market Reaction
to Short-Interest Announcements. The Journal of Financial and Quantitative
Analysis, 28, 177–194.
White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a
direct test for heteroskedasticity. Econometrica, 48, 817–838.
38
Table 1 Summary statistics
The table presents the summary statistics for all of the variables used in this study. The financial crisis
is from July 1, 2007, through December 31, 2009, for RE09; and from July 1, 2007, through December
31, 2008, for RE08. The RE98 is from August 3, 1998, until the day in 1998 on which the bank’s stock
attains its lowest price during the LTCM crisis. The ∆SI and ∆COST are changes in the short interest
and the stock borrowing cost from June 2005 through June 2007, respectively. ∆SI12m and ∆LTCMSI
are similar measures of short interest from June 2006 through June 2007 and from August 1996
through July 1998. The other variables are bank characteristics in the year 2006. The variable
definitions are in the appendix.
(1) (2) (3) (4) (5) (6)
Obs Mean SD p25 p50 p75
RE09 683 -24.50 28.76 -42.55 -20.88 -4.84
RE08 683 -29.70 32.30 -51.94 -30.52 -7.08
RE98 249 -82.29 18.55 -96.16 -88.66 -75.91
∆SI 643 1.98 4.84 -0.03 0.30 3.82
∆SI12m 688 1.24 3.42 -0.02 0.12 2.20
∆LTCMSI 328 1.02 3.84 0.00 0.18 0.90
∆COST 400 0.15 0.90 0.00 0.00 0.14
BHAR06 696 0.35 24.48 -12.84 -2.84 10.47
LnAssets 695 13.06 1.83 11.68 12.88 14.24
BM 691 0.93 2.20 0.45 0.61 0.78
Leverage 638 9.45 5.14 5.41 9.78 12.55
TCE 583 13.83 27.83 3.57 5.23 10.85
Beta 696 0.94 0.18 0.87 0.96 1.04
IDIORISK 696 0.02 0.01 0.01 0.02 0.03
MES 696 -1.09 0.86 -1.64 -1.09 -0.35
∆EDF 515 0.28 0.21 0.12 0.27 0.44
∆NPL 436 2.53 2.75 0.73 1.75 3.51
39
Table 2 Comparison of bank characteristics
This table presents the differences in characteristics between two groups: Bottom Quintile (i.e., lowest
RE09 quintile) and Other Quintiles (i.e., other banks). The crisis period is from July 1, 2007, through
December 31, 2009, for RE09; and from July 1, 2007, through December 31, 2008, for RE08. The ∆SI
and ∆COST are changes in the short interest and the stock borrowing cost from June 2005 through June
2007, respectively. The ∆SI12m is similar measure of short interest from June 2006 through June 2007.
The other variables are firm characteristics in the year 2006. The variable definitions are in the
appendix. The superscripts *, **, and *** denote significance at 10%, 5%, and 1% levels, respectively.
Bottom
Quintile
Obs
Bottom
Quintile
Mean
Other
Quintiles
Obs
Other
Quintiles
Mean
Difference
Bottom-Other
t-statistics
RE09 136 -68.1106 547 -13.6570 -54.4536*** -30.2008
RE08 136 -68.9451 547 -19.9404 -49.0046*** -19.9002
∆SI 122 3.1305 521 1.7141 1.4165*** 2.9280
∆SI12m 135 1.7882 553 1.1112 0.6770** 2.0679
∆COST 61 0.4153 339 0.1001 0.3152** 2.5229
BHAR06 136 -9.0278 560 2.6314 -11.6592*** -5.0700
LnAssets 135 13.0488 560 13.0619 -0.0132 -0.0749
BM 135 0.9845 556 0.9206 0.0638 0.3019
Leverage 129 11.3651 509 8.9596 2.4054*** 4.8352
TCE 123 11.3835 460 14.4790 -3.0955 -1.0958
Beta 136 0.9225 560 0.9418 -0.0193 -1.1311
IDIORISK 136 0.0209 560 0.0219 -0.0010 -0.8284
MES 136 -1.0405 560 -1.1069 0.0664 0.8099
∆EDF 115 0.4889 380 0.2245 0.2644*** 13.5995
∆NPL 107 5.326 311 1.652 3.6740*** 14.4165
40
Table 3 Short selling and financial crisis returns
This table presents OLS regression results for the short selling and financial crisis returns. The crisis
period is from July 1, 2007, through December 31, 2009.
,crisis , , t 109
i i pre crisis i iRE SI Z
where RE09i,crisis represent stock returns for bank i in the financial crisis; ,i pre crisis
SI
is the change in
short interest for bank i in the pre-crisis period of the crisis; , 1i t
Z
is a vector of control variables for
bank i in the year 2006. The variable definitions are in the appendix. The t-statistics are in parentheses
and are based on standard errors adjusted for heteroskedasticity (White, 1980) and industry clustering
(Petersen, 2009). The superscripts *, **, and *** denote significance at the 10%, 5%, and 1% levels,
respectively.
(1) (2) (3) (4)
RE09 RE09 RE09 RE09
Constant -33.4786*** -53.8432*** -0.5101 -11.0910
(-3.22) (-5.38) (-0.05) (-0.70)
∆SI -0.9982*** -1.0844*** -1.0612*** -1.1830***
(-3.57) (-3.56) (-3.75) (-4.04)
BHAR06 0.2088*** 0.2331*** 0.1897*** 0.1816***
(3.64) (4.08) (3.38) (3.11)
LnAssets -1.1034 -0.4164 -1.3483* -0.9473
(-1.50) (-0.52) (-1.81) (-1.15)
BM -0.1494 -0.8885* -0.3250 -0.5397
(-0.28) (-1.83) (-0.61) (-0.92)
Beta 38.2027*** 37.7829*** 4.6405
(3.20) (3.18) (0.33)
Leverage -1.1629*** -1.1645*** -1.0100***
(-4.57) (-4.41) (-3.33)
TCE 0.0712** 0.0283
(2.14) (0.81)
MES -7.0877*** -6.6086***
(-4.11) (-2.96)
IDIORISK -25.8571 -47.9235
(-0.20) (-0.39)
Obs. 575 525 575 525
Adj-R2 0.1003 0.0653 0.1078 0.0979
41
Table 4 Short selling and LTCM crisis returns
This table presents OLS regression results for the short selling and LTCM crisis returns. The crisis
period is from August 3, 1998, through December 31, 1998.
,crisis , , t 198
i i pre crisis i iRE LTCMSI Z
where RE98i,crisis represent stock returns for bank i in the LTCM crisis; ,i pre crisis
LTCMSI
is the
change in short interest for bank i in the pre-crisis period of the LTCM crisis; , 1i t
Z
is a vector of
control variables for bank i in the year 1997. The variable definitions are in the appendix. The
t-statistics are in parentheses and are based on standard errors adjusted for heteroskedasticity (White,
1980) and industry clustering (Petersen, 2009). The superscripts *, **, and *** denote significance at
the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4)
RE98 RE98 RE98 RE98
Constant -83.5167*** -93.3388*** -80.7967*** -90.1619***
(-12.79) (-12.01) (-10.02) (-10.68)
∆LTCMSI -0.4865** -0.5916** -0.4971** -0.5268*
(-2.56) (-1.99) (-2.50) (-1.83)
BHAR97 -0.0195 0.0194 -0.0106 0.0257
(-0.93) (0.83) (-0.51) (0.99)
LnAssets97 1.4420* 0.6129 0.6023 0.7292
(1.90) (0.75) (0.81) (0.79)
BM97 11.2706* 21.0396*** 12.8849** 19.2531***
(1.80) (2.99) (2.03) (2.86)
Leverage97 -0.4248*** -0.2973** -0.2107
(-3.14) (-2.08) (-0.79)
Beta97 -15.8279*** -8.3232* 0.3063
(-4.15) (-1.83) (0.04)
TCE97 0.0372 0.0359
(0.54) (0.52)
MES97 5.6706*** 4.2989
(3.70) (1.28)
IDIORISK97 -63.4524 -83.6520
(-0.52) (-0.87)
Obs. 212 124 212 124
Adj-R2 0.1319 0.0963 0.1180 0.0908
42
Table 5 LTCM crisis returns and short selling in pre-crisis period
This table presents OLS regression results for the LTCM crisis return and short selling in pre-crisis
period.
, crisis , , t 1 98
i pre i crisis i iSI RE Z
where
,i pre crisisSI
is the change in short interest for bank i in the pre-crisis period of the financial
crisis while RE98i,crisis is the stock returns for bank i in the LTCM crisis; , 1i t
Z
is a vector of control
variables for bank i in the year 2004. The variable definitions are in the appendix. The t-statistics are in
parentheses and are based on standard errors adjusted for heteroskedasticity (White, 1980) and industry
clustering (Petersen, 2009). The superscripts *, **, and *** denote significance at the 10%, 5%, and
1% levels, respectively.
(1) (2) (3) (4)
∆SI ∆SI ∆SI ∆SI
Constant 2.4188 -0.4563 6.9827*** 5.2138
(0.71) (-0.13) (3.44) (1.55)
RE98 -0.0628*** -0.0802*** -0.0653*** -0.0884***
(-6.67) (-12.85) (-3.59) (-4.14)
BHAR04 0.0240 0.0200 0.0210 0.0207
(0.44) (0.37) (0.41) (0.39)
LnAssets04 -1.0285*** -0.6109*** -1.3358*** -1.5273***
(-5.41) (-4.46) (-3.37) (-2.63)
BM04 4.1995*** 4.7528*** 4.9020*** 6.7421**
(5.24) (3.62) (3.34) (2.10)
Beta04 0.2991 -0.5327 0.6078
(0.09) (-0.17) (0.15)
Leverage04 0.2055** 0.2340* 0.2871
(2.10) (1.80) (1.57)
TCE04 0.0134 0.0179
(0.36) (0.62)
MES04 -0.3336 -0.3859
(-0.31) (-0.42)
IDIORISK04 -160.1702** -188.8326*
(-2.21) (-1.88)
Obs. 109 90 109 90
Adj-R2 0.0344 0.0034 0.0602 0.0434
43
Table 6 Short selling and financial crisis returns: Non-financial firms
This table presents OLS regression results for the short selling and financial crisis returns for
non-financial firms. The crisis period is from July 1, 2007, through December 31, 2009.
,crisis , , t 109
i i pre crisis i iRE SI Z
where RE09i,crisis represent stock returns for bank i in the financial crisis; ,i pre crisis
SI
is the change in
short interest for bank i in the pre-crisis period of the crisis; , 1i t
Z
is a vector of control variables for
bank i in the year 2006. The variable definitions are in the appendix. The t-statistics are in parentheses
and are based on standard errors adjusted for heteroskedasticity (White, 1980) and industry clustering
(Petersen, 2009). The superscripts *, **, and *** denote significance at the 10%, 5%, and 1% levels,
respectively.
(1) (2) (3) (4)
RE09 RE09 RE09 RE09
Constant -17.1186***
-17.3212***
-12.4984***
-13.6439***
(-7.78) (-7.25) (-4.26) (-4.42)
∆SI -0.1347***
-0.1304***
-0.1437***
-0.1383***
(-3.18) (-3.07) (-3.35) (-3.22)
BHAR06 -0.0187***
-0.0185***
-0.0176***
-0.0173***
(-2.78) (-2.75) (-2.72) (-2.67)
LnAssets 0.7667***
0.7926***
0.3621 0.4183
(3.69) (3.59) (1.36) (1.47)
BM -4.7471* -5.3901
** -4.8799
* -5.3285
**
(-1.81) (-2.08) (-1.88) (-2.06)
Beta 1.0383 0.9685 0.6172
(1.10) (1.01) (0.45)
Leverage 0.0324***
0.0321***
0.0326***
(5.53) (5.56) (5.69)
TCE 0.0135 0.0122
(1.29) (1.15)
MES -91.8474 -62.8853
(-1.49) (-0.72)
IDIORISK -0.9005**
-0.7417*
(-2.01) (-1.66)
Obs. 2,624 2,571 2,624 2,571
Adj-R2 0.0181 0.0166 0.0195 0.0185
44
Table 7 Pseudo-events predictability of short selling
This table shows the coefficients of change in short interest in regressions of the financial crisis returns
based on simulation. For comparison, the coefficients in the first row are our original sample of the
financial crisis (i.e., Table 3). The second row presents the coefficients from the simulation. For each
bank in our sample, we randomly choose a non-crisis month as its pseudo-event month. We regress the
annualized buy-and-hold stock returns of 30-month pseudo-events (from month t to month t+29) on
change in short interest of 24-month before the pseudo-events (from month t-24 to month t-1), and
control for the pre-year bank characteristics as those in Table 3. We repeat the process for 1,000 times
and report the average coefficient of the change in short interest (∆SI) and its associated p-value in
parentheses. p-value is the fraction of the number of times that the simulated coefficient is larger than the
coefficient of the actual sample (in Table 3).
Model 1 Model 2 Model 3 Model 4
Original sample on actual
Crisis periods
-0.9982 -1.0844 -1.0612 -1.1830
Sample banks on Non-Crisis
periods pseudo-events
-0.4032
(0.068)
-0.3667
(0.054)
-0.3493
(0.047)
-0.3502
(0.027)
45
Table 8 Abnormal short interest and stock returns (I)
This table presents OLS regression results for the short selling and financial crisis returns. The crisis period is from July 1, 2007, through December 31, 2009.
,crisis , , t 109 ( ) 1, 2,
i i pre crisis i iRE ABSI j Z j
where RE09i,crisis represent stock returns for bank i in the financial crisis; ,( )
i pre crisisABSI j
is the change in abnormal short interest for bank i in the pre-crisis period of the
crisis; , 1i t
Z
is a vector of control variables for bank i in the year 2006. To construct ABSI(1), we regress short interest (as percentage of the number of share outstanding) on
explanatory variables (size, book-to-market, momentum, and industry dummies). To construct ABSI(2), besides size, book-to-market, momentum, and industry dummies, we
add share turnover and institutional ownership as explanatory variables in the short interest regression. Following Karpoff and Lou (2010), abnormal short interest is
calculated by subtracting raw short interest from the fitted short interest of the short interest regression. Industry is defined as two-digit SIC code from CRSP, share turnover
is share trading volume divided by the number of share outstanding, and institutional ownership is the number of shares owned by institutional investors divided by the
number of share outstanding. The variable definitions are in the appendix. The t-statistics are in parentheses and are based on standard errors adjusted for heteroskedasticity
(White, 1980). The superscripts *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4) (5) (6) (7) (8)
RE09 RE09 RE09 RE09 RE09 RE09 RE09 RE09
Constant -20.7047** -54.0617*** 13.2073 -15.8855 -19.4389** -52.4897*** 11.6412 -16.9220
(-2.34) (-5.75) (1.51) (-1.15) (-2.19) (-5.60) (1.33) (-1.19)
∆ABSI(1) -0.8669*** -0.8299** -0.9128*** -0.9396***
(-2.97) (-2.41) (-2.95) (-2.70)
∆ABSI(2) -0.7098*** -0.6796** -0.7489*** -0.7651**
(-2.67) (-2.14) (-2.71) (-2.45)
BHAR06 0.1893*** 0.2418*** 0.1720*** 0.2055*** 0.1963*** 0.2481*** 0.1809*** 0.2136***
(3.24) (4.18) (3.00) (3.53) (3.37) (4.32) (3.17) (3.70)
LnAssets -1.3619** 0.4100 -1.7885*** 0.1531 -1.3496** 0.4494 -1.7344*** 0.2614
(-2.33) (0.63) (-3.01) (0.22) (-2.30) (0.69) (-2.89) (0.38)
BM 0.2694 -0.5489 0.1503 -0.3024 0.3208 -0.4870 0.2033 -0.2463
(0.38) (-1.14) (0.22) (-0.54) (0.48) (-1.10) (0.31) (-0.48)
Beta 33.2917*** 23.8190** -6.4520 31.1316*** 21.3427** -8.0062
(3.07) (2.20) (-0.50) (2.88) (2.00) (-0.60)
46
Leverage -1.8929*** -1.9193*** -0.9849*** -1.8664*** -1.8861*** -0.9409***
(-11.71) (-11.83) (-4.44) (-11.46) (-11.56) (-4.21)
TCE 0.1046*** 0.0607* 0.0993*** 0.0561*
(3.55) (1.94) (3.30) (1.74)
MES -6.4437*** -5.9391*** -6.0970*** -5.6696***
(-4.02) (-2.72) (-3.84) (-2.59)
IDIORISK -145.1415 -91.2912 -120.0952 -60.5015
(-1.10) (-0.80) (-0.92) (-0.53)
Obs. 638 534 638 534 638 534 638 534
Adj. R2
0.2292 0.0926 0.2363 0.1252 0.2209 0.0821 0.2272 0.1119
47
Table 9 Abnormal short interest and stock returns (II)
This table presents two-stage least squares (2SLS) regression results for the short selling and financial
crisis returns. The empirical model is as follows.
Stage 1: Regress ,i pre crisis
SI
on the all control variables of the model:
, , 1
i pre crisis i t iSI Z
where , 1i t
Z
is a vector of control variables in the year 2004, including BHAR04, LnAssets04, BM04,
Leverage04, TCE04, Beta04, MES04, and IDIORISK04. .
Stage 2: Replace ,i pre crisis
SI
with the (3)i
ABSI derived from the 1st-stage into the regression
equation (1):
, crisis , 1,09 (3)
i i t ii pre crisisRE ZABSI
where RE09i,crisis represent stock returns for bank i in the financial crisis;
,,(3) i pre crisisi pre crisisi SISIABSI
, ,i pre crisisSI
is fitted value that derived from the 1st-stage
regression, and , 1i t
Z
is a vector of control variables in the year 2006. The variable definitions are in
the appendix. The t-statistics are in parentheses and are based on standard errors adjusted for
heteroskedasticity (White, 1980). The superscripts *, **, and *** denote significance at the 10%, 5%,
and 1% levels, respectively.
(1) (2) (3) (4)
RE09 RE09 RE09 RE09
Constant -40.1009*** -62.5638*** 10.0340 -10.0593
(-3.11) (-4.92) (0.85) (-0.36)
∆ABSI(3) -1.1360*** -1.3494*** -1.3702*** -1.5130***
(-4.05) (-4.62) (-4.78) (-5.22)
BHAR06 0.2687*** 0.2574*** 0.2568*** 0.2261***
(4.36) (4.11) (4.06) (3.19)
LnAssets -1.3276* -1.0921 -2.1722*** -1.8289**
(-1.71) (-1.26) (-2.67) (-2.01)
BM -0.2713 -1.0395* -0.2943 -0.5565
(-0.46) (-1.81) (-0.46) (-0.74)
Beta 46.2697*** 53.8245*** 12.9563
(2.98) (3.23) (0.41)
Leverage -1.1389*** -1.2378*** -0.9356***
(-4.32) (-4.57) (-2.81)
TCE 0.0589 0.0049
(1.50) (0.11)
MES -8.5299*** -6.8125*
(-4.34) (-1.93)
IDIORISK -157.4509 -82.8259
(-1.09) (-0.57)
Obs. 513 468 476 436
Adj. R2
0.1214 0.0816 0.1473 0.1197
48
Table 10 Short selling, loan quality, and default risk in the 2007-2009 financial crisis
This table presents OLS regression results for short selling, loan quality, and default risk. The crisis period is from July 1, 2007, through December 31, 2009.
,crisis , , t 1
i i pre crisis i iNPL SI Z
,crisis , , t 1
i i pre crisis i iEDF SI Z
where ∆NPLi,crisis and ∆EDFi,crisis represent the change in the nonperforming loan ratio and the change in the expected default frequency, respectively, of bank i in the
financial crisis; ,i pre crisisSI
is the change in short interest for bank i in the pre-crisis period of the financial crisis;
, 1i tZ
is a vector of control variables for bank i in the year
2006. The variable definitions are in the appendix. The t-statistics are in parentheses and are based on standard errors adjusted for heteroskedasticity (White, 1980) and
industry clustering (Petersen, 2009). The superscripts *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4) (5) (6) (7) (8)
∆NPL ∆NPL ∆NPL ∆NPL ∆EDF ∆EDF ∆EDF ∆EDF
Constant 2.4155* 2.1603* 1.7114 0.1658 0.2444** 0.3745*** 0.0621 0.1810
(1.88) (1.65) (1.40) (0.08) (2.03) (3.18) (0.50) (1.11)
∆SI 0.1120* 0.0958* 0.1242** 0.1023* 0.0085*** 0.0087*** 0.0086*** 0.0094***
(1.91) (1.66) (2.08) (1.73) (3.38) (3.48) (3.38) (3.68)
BHAR06 -0.0458*** -0.0436*** -0.0461*** -0.0425*** -0.0023*** -0.0023*** -0.0022*** -0.0021***
(-3.57) (-3.57) (-3.59) (-3.44) (-3.48) (-3.58) (-3.41) (-3.13)
LnAssets -0.0488 -0.0557 0.0216 -0.0056 0.0181** 0.0150** 0.0167** 0.0174**
(-0.68) (-0.77) (0.26) (-0.07) (2.51) (2.04) (2.18) (2.25)
BM 0.0251 0.0272 0.0321 0.0320 -0.0041 0.0128 -0.0173 0.0038
(0.35) (0.41) (0.45) (0.47) (-0.07) (0.20) (-0.29) (0.06)
Beta 0.2235 0.4094 1.8663 -0.3237*** -0.3224*** -0.1468
(0.16) (0.30) (0.93) (-3.02) (-3.02) (-1.14)
Leverage 0.0002 0.0001 0.0081 0.0092*** 0.0086*** 0.0076***
(0.01) (0.00) (0.18) (3.39) (3.15) (2.81)
TCE 0.0316 0.0297 -0.0006 -0.0002
(0.96) (0.92) (-0.82) (-0.26)
MES 0.1868 0.3289 0.0565*** 0.0391*
(0.79) (1.01) (3.29) (1.75)
IDIORISK 9.4594 11.5440 -1.8568 -2.1097*
49
(0.72) (0.79) (-1.57) (-1.74)
Obs. 354 354 354 354 423 407 423 407
Adj-R2 0.0789 0.0843 0.0786 0.0802 0.0881 0.0512 0.1026 0.0847
50
Table 11 Short selling and financial crisis: Controlling banks’ risk-taking in pre-crisis period
This table presents OLS regression results for short selling and financial crisis returns by considering the risk-taking levels in the pre-crisis period. The crisis period is from
July 1, 2007, through December 31, 2009. Firms are sorted into three groups based on their level of risk-taking. We run the regression for each group:
,crisis , , t 109
i i pre crisis i iRE SI Z
where RE09i,crisis represents stock returns for bank i in the financial crisis; ,i pre crisis
SI
is the change in short interest for bank i in the pre-crisis period of the financial crisis;
, 1i tZ
is a vector of control variables for bank i in the year 2006. The variable definitions are in the appendix. The t-statistics are in parentheses and are based on the standard
errors adjusted for heteroskedasticity (White, 1980) and industry clustering (Petersen, 2009). The superscripts *, **, and *** denote significance at the 10%, 5%, and 1%
levels, respectively.
Panel A: High risk-taking subsamples Panel B: Medium risk-taking subsamples Panel C: Low risk-taking subsamples
Leverage TCE Beta Leverage TCE Beta Leverage TCE Beta
(1)
RE09
(2)
RE09
(3)
RE09
(4)
RE09
(5)
RE09
(6)
RE09
(7)
RE09
(8)
RE09
(9)
RE09
Constant -27.7526 8.6891 -23.7053 5.8778 -6.6851 5.9402 -5.5207 -40.4838*** -110.7346***
(-1.61) (0.31) (-0.32) (0.47) (-0.22) (0.18) (-0.16) (-2.76) (-3.65)
∆SI -1.0625*** -1.5566*** -1.0560*** -1.2860*** -1.1223*** -2.0453* -1.3224 -0.7618 -1.8667
(-5.25) (-2.59) (-3.56) (-4.74) (-2.95) (-1.96) (-1.27) (-1.52) (-1.08)
BHAR06 0.0653 0.1289 0.2084*** 0.2643*** 0.2243* 0.1836* 0.1191*** 0.2085*** -0.0227
(0.34) (1.55) (6.57) (3.25) (1.79) (1.83) (4.84) (3.70) (-0.19)
LnAssets -2.8759** -0.4687 -0.1867 -0.5834 -2.3574* -0.4294 -0.7523 -0.2875 6.3255***
(-2.08) (-0.31) (-0.07) (-0.96) (-2.26) (-0.48) (-0.30) (-0.44) (3.05)
BM -0.2744 -2.2835 -1.3041* -1.9615 -0.5348 0.2852 -2.9819*** -0.0047 21.4483
(-0.82) (-2.13) (-1.99) (-0.90) (-1.01) (0.45) (-4.72) (-0.05) (2.27)
Beta 49.5262*** -13.3221 11.5529 -24.9839*** -9.8909 -24.3137 -7.7702 21.6491** 11.9305
(5.55) (-0.25) (0.37) (-5.75) (-0.67) (-0.48) (-0.75) (2.26) (1.02)
TCE -0.1382 0.0012 0.0401 -0.0429 3.0893** 0.0523 0.0498*** 1.1736 -0.0297
(-1.63) (0.03) (1.52) (-0.71) (3.73) (1.09) (5.09) (0.76) (-0.44)
Leverage -0.5734 -1.4529*** -0.7155 -1.2174 -0.6216 -1.0134*** 0.7360 -0.9948*** -1.7119***
(-0.92) (-5.13) (-1.28) (-1.50) (-1.00) (-5.46) (0.62) (-7.18) (-3.74)
MES -4.8976*** -9.1358** -6.8018* -9.6410*** -10.7605*** -6.0857 -8.6454** -2.8486* -1.9208
51
(-5.78) (-2.02) (-1.82) (-9.98) (-5.90) (-1.28) (-2.18) (-1.94) (-0.67)
IDIORISK -310.0927** -207.5354** -376.0092 276.9265*** 37.9354 65.7739 -48.1068 133.2917 238.7698*
(-2.15) (-2.36) (-1.45) (3.32) (0.18) (0.21) (-0.24) (1.17) (1.68)
Obs. 163 151 175 235 217 212 126 156 137
Adj-R2 0.0540 0.1825 0.0829 0.0474 0.0506 0.1058 0.0765 0.0425 0.0678
52
Table 12 Quantile regression: short selling and financial crisis returns
This table presents quantile regression results for short selling and financial crisis returns. The crisis
period is from July 1, 2007, through December 31, 2009.
,crisis , , t 109
i i pre crisis i iRE SI Z
where RE09i,crisis represents stock returns for bank i in the recent financial crisis; ,i pre crisisSI
is the
change in the short interest for bank i in the pre-crisis period of the financial crisis; , 1i t
Z
is a vector
of control variables for bank i in the year 2006. The variable definitions are in the appendix. The
t-statistics are in parentheses and are based on standard errors adjusted for heteroskedasticity (White,
1980) and industry clustering (Petersen, 2009). The superscripts *, **, and *** denote significance at
the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4) (5)
RE09 RE09 RE09 RE09 RE09
Quantile =0.2 Quantile =0.4 Quantile =0.6 Quantile =0.8 OLS
Constant -1.1086 -17.6811 -15.2842 -7.5817 -11.0910
(-0.04) (-0.99) (-0.83) (-0.35) (-0.70)
∆SI -1.3181*** -1.3726*** -0.5524* -0.5842 -1.1830***
(-3.49) (-4.86) (-1.86) (-1.46) (-4.04)
BHAR06 0.2641*** 0.1999*** 0.0997 0.1187 0.1816***
(2.65) (3.35) (1.62) (1.62) (3.11)
LnAssets -2.2869 0.2799 0.1256 0.0952 -0.9473
(-1.58) (0.30) (0.13) (0.09) (-1.15)
BM -0.4790 -0.6764 -0.3948 -0.4106 -0.5397
(-0.53) (-1.20) (-0.65) (-0.59) (-0.92)
Beta -2.3795 -5.7098 -2.7440 5.2273 4.6405
(-0.09) (-0.32) (-0.16) (0.26) (0.33)
TCE 0.0265 0.0260 0.0261 0.0162 0.0283
(0.39) (0.55) (0.56) (0.26) (0.81)
Leverage -1.7042*** -1.2696*** -0.8340*** -0.5313 -1.0100***
(-3.67) (-4.27) (-2.73) (-1.43) (-3.33)
MES -5.1348 -5.8992** -7.8874*** -6.2156** -6.6086***
(-1.19) (-2.25) (-2.96) (-2.06) (-2.96)
IDIORISK -27.4731 -93.5415 8.9415 -103.3147 -47.9235
(-0.12) (-0.72) (0.07) (-0.58) (-0.39)
Pseudo R2 0.0997 0.0759 0.0586 0.0429
Adj-R2 0.0979
Obs. 525 525 525 525 525
53
Table 13 Stock borrowing costs and financial crisis returns
This table presents OLS regression results for stock borrowing costs and financial crisis returns. The
crisis period is from July 1, 2007, through December 31, 2009.
,crisis , , t 109
i i pre crisis i iRE COST Z
where RE09i,crisis represents the stock returns for bank i in the financial crisis; ,i pre crisisCOST
is the
change in stock borrowing costs for bank i in the pre-crisis period of the financial crisis; , 1i t
Z
is a
vector of control variables for bank i in the year 2006. The variable definitions are in the appendix. The
t-statistics are in parentheses and are based on standard errors adjusted for heteroskedasticity (White,
1980) and industry clustering (Petersen, 2009). The superscripts *, **, and *** denote significance at
the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4)
RE09 RE09 RE09 RE09
Constant -38.1137* -52.1133** -10.6749 -10.5196
(-1.78) (-2.46) (-0.60) (-0.35)
∆COST -4.3369* -5.4595** -4.1433* -4.6930**
(-1.90) (-2.34) (-1.80) (-2.05)
BHAR06 0.1798*** 0.2077*** 0.1600** 0.1613**
(2.61) (3.01) (2.39) (2.27)
LnAssets -1.2370 -0.7026 -0.9912 -0.9209
(-1.31) (-0.67) (-1.02) (-0.84)
BM 3.7104 -3.0361 3.6991 -4.2848
(0.60) (-0.40) (0.59) (-0.54)
Beta 40.5145** 40.9150** 2.9419
(2.26) (2.32) (0.11)
Leverage -0.9470*** -0.9168*** -0.7412*
(-3.10) (-2.79) (-1.94)
TCE 0.0548 0.0272
(1.40) (0.66)
MES -5.0405** -5.3054
(-2.52) (-1.54)
IDIORISK 144.6500 11.3416
(0.70) (0.06)
Obs. 366 329 366 329
Adj-R2 0.0844 0.0622 0.0836 0.0731
54
Table 14 Robustness checks
This table presents the robustness checks for our main hypothesis. The first three columns present OLS regression results for the crisis returns and the change in short interest
by using different time period definitions. The next two columns present OLS regression results for RE09 and ∆SI in two subsamples based on bank size. The last two
columns present the OLS regression results for RE09 and ∆SI in two subsamples based on different industries: Commercial&Investment and Insurance.
,crisis , , t 1
i i pre crisis i iCrisis Return SI Z
where Crisis Returni,crisis represents the stock returns for bank i in the financial crisis; ,i pre crisisSI
is the change in short interest for bank i in the pre-crisis period of the
financial crisis; , 1i t
Z
is a vector of control variables for bank i in the year 2006. The variable definitions are in the appendix. The t-statistics are in parentheses and are based
on standard errors adjusted for heteroskedasticity (White, 1980) and industry clustering (Petersen, 2009). The superscripts *, **, and *** denote significance at the 10%, 5%,
and 1% levels, respectively.
(1) (2) (3) (4) (5) (6) (7)
RE08 RE09 RE08 RE09 RE09 RE09 RE09
Subsamples=
Small Banks
Subsamples=
Large Banks
Subsamples=
Commercial
&Investment
Subsamples=
Insurance
Constant -23.1243 -17.6000 -11.6565 25.6189 5.4330 -1.1371 27.1387
(-1.27) (-1.31) (-0.78) (0.76) (0.15) (-0.06) (0.57)
∆SI -1.0630*** -2.2441*** -1.2317*** -0.9357*** -3.2733***
(-3.21) (-3.06) (-3.76) (-2.82) (-4.77)
∆SI12m -1.2877*** -1.0466**
(-3.25) (-2.31)
BHAR06 0.0357 0.2104*** 0.0426 0.0683 0.2707***
0.1825*** 0.1672
(0.52) (3.78) (0.65) (0.82) (3.55) (2.74) (1.30)
LnAssets -2.1559** -0.8318 -1.5583* -2.8066 -3.3554**
-1.0433 -2.8115
(-2.30) (-1.08) (-1.73) (-1.11) (-2.13) (-0.95) (-1.45)
BM -1.8934*** -0.1413 -1.4053*** -1.5730 -0.3011 0.0665 -3.9131**
(-3.47) (-0.29) (-3.29) (-1.36) (-0.48) (0.14) (-2.58)
Beta 20.3376 15.4932 8.2962 -9.6702 19.9563 -7.7260 -3.1034
55
(1.19) (1.59) (0.77) (-0.52) (0.76) (-0.48) (-0.06)
TCE 0.0047 0.0217 -0.0090 0.0881**
0.0088 -0.0511 0.0551
(0.09) (0.63) (-0.18) (2.14) (0.16) (-0.62) (1.02)
Leverage -0.5814* -1.1254*** -0.7920*** -1.1614**
-0.6897* -0.9389*** -1.4975**
(-1.75) (-4.15) (-2.64) (-2.59) (-1.79) (-2.82) (-2.25)
MES -10.6934*** -3.0167 -7.7052*** -5.9839* -5.6240 -7.2515*** -13.3807**
(-3.97) (-1.59) (-3.18) (-1.83) (-1.36) (-2.63) (-2.04)
IDIORISK 4.6237 -121.8329 -182.4181 -132.0328 139.9965 -19.8892 -98.6838
(0.03) (-1.06) (-1.45) (-0.77) (0.71) (-0.13) (-0.47)
Obs. 525 564 564 265 260 422 79
Adj-R2 0.0809 0.0984 0.0569 0.0838 0.1119 0.0648 0.2188