Flying Under the Radar: The E ects of Short-Sale Disclosure Rules on Investor Behavior ... ·...
Transcript of Flying Under the Radar: The E ects of Short-Sale Disclosure Rules on Investor Behavior ... ·...
Flying Under the Radar:
The Effects of Short-Sale Disclosure Rules on
Investor Behavior and Stock Prices∗
Stephan Jank1, Christoph Roling2, and Esad Smajlbegovic3
March 9, 2018
∗We thank the German Federal Financial Supervisory Authority (Bundesanstalt fur Finanzdien-stleistungsaufsicht, BaFin) for providing the short position notification data. We are grateful to PuriyaAbbassi, Vikas Agarwal, Karl Diether (discussant), Zsuzsa Huszar (discussant), Rajkamal Iyer, CharlesJones (discussant), Alexander Klos (discussant), Philipp Konig, Christian Leuz, Christoph Meinerding,Emanuel Monch, Esteban Prieto, David Rakowski (discussant), Adam Reed (discussant), ZachariasSautner, Christoph Schneider, Gunter Strobl, Pedro Saffi, Patrick Verwijmeren, Verena Weick-Ludewig,as well as seminar and conference participants at the Deutsche Bundesbank, the University of Mannheim,the University of Washington 2016 Summer Finance Conference, the 2016 Annual Meeting of the Eu-ropean Finance Association (EFA), the Fall 2016 NBER Asset Pricing Meeting, and the 2017 AnnualMeeting of the American Finance Association (AFA) for their helpful comments and suggestions. Wetake responsibility for all remaining errors. Financial support from the German Research Foundation(Deutsche Forschungsgemeinschaft, DFG), Grant Number: JA-2396/1-1, is gratefully acknowledged. Thispaper represents the authors’ personal opinions and does not necessarily reflect the views of the DeutscheBundesbank, its staff, or the Eurosystem.
1Deutsche Bundesbank and Centre for Financial Research (CFR), Cologne. Deutsche Bundesbank,Wilhelm-Epstein-Str. 14, 60431 Frankfurt am Main, Germany. Phone: +49-69-9566-2854, E-mail:[email protected]
2Deutsche Bundesbank, Wilhelm-Epstein-Str. 14, 60431 Frankfurt am Main, Germany. Phone: +49-69-9566-3044, E-mail: [email protected]
3Erasmus University Rotterdam, Erasmus School of Economics, Burgemeester Oudlaan 50, PO Box 1738,30000 DR Rotterdam, The Netherlands. Phone: +31-10-40-81834, E-mail: [email protected]
Flying Under the Radar:
The Effects of Short-Sale Disclosure Rules on
Investor Behavior and Stock Prices
March 9, 2018
Abstract
We study how disclosure requirements for large short positions affect investor behaviorand security prices. Drawing on confidential notifications below the applicabledisclosure threshold, we show that a considerable share of short sellers avoid disclosureby accumulating their positions just below the threshold. This behavior appears tobe part of investors’ general strategy to conceal private information. Studying theperformance of their short positions, we find that secretive investors are particularlyinformed and trade around fundamental news releases. Finally, we document thatshort sellers’ evasive behavior in response to transparency regulation hampers pricediscovery.
Keywords: information disclosure, short selling, investor behavior,limits to arbitrage, stock market efficiency
JEL: G14, G15, G23
Disclosure requirements for investors’ holdings are a prevalent feature of financial market
regulation. Since the Securities Act of 1933 and the Securities Exchange Act of 1934,
significant efforts have been made to increase the transparency of investors’ long posi-
tions.1 Conversely, less regulatory attention has been given to their short positions. This
asymmetry of disclosure requirements has been debated extensively in the aftermath of
the global financial crisis, with regulators on both sides of the Atlantic contemplating
new transparency measures for short positions. Proponents of short-selling disclosure
rules argue that greater transparency would level the playing field for investors, expose
potentially abusive short selling, and help to improve price discovery in the market (NYSE,
2015; NASDAQ, 2015). However, critics raise the concern that a timely publication of short
positions may pose a threat to proprietary investment strategies. In response, investors
may reduce their short-selling activities to protect their private information, engaging less
in arbitrage trading, and thereby hampering price discovery (U.S. SEC, 2014). While
this debate is still ongoing in the United States, in 2012 the European Union adopted a
uniform rule on short position disclosure.2 According to the European regulation, investors
are required to publish any net short position that reaches the threshold of 0.5% of the
shorted stocks’ issued share capital one day after that position arises. Disclosures contain
the identity of the short seller, name and identifier of the target stock, as well as the date
and magnitude of the position.
1In the United States, for example, several disclosure rules apply to long positions. First, anyone whoacquires beneficial ownership of more than 5% of a voting class of a publicly traded company has to file a13D or 13G filing with the Securities and Exchange Commission (SEC). Second, institutional investmentmanagers of a certain size must report their quarterly holdings in 13F filings. Mutual funds also mustregularly report their portfolio holdings to their shareholders (SEC forms N-CSR and N-Q.).
2Among the EU countries, Spain and the United Kingdom had already implemented short-sale disclosurerules in 2008; France followed in 2011. Japan also introduced disclosure requirements in 2008. In theUnited States similar measures have been debated. The Dodd–Frank Act required the U.S. Securities andExchange Commission (SEC) to conduct a study of the feasibility, benefits and costs of real-time disclosuresof shorting. However, the real-time disclosure of shorting was not adopted (U.S. SEC, 2014). Recently,the debate on short-sale disclosure resurfaced, with both large stock exchanges, the NSYE and NASDAQ,filing rulemaking petitions for short-sale disclosure with the SEC (petition number: 4-689, October 7,2015; petition number: 4-691, December 7, 2015, see https://www.sec.gov/rules/petitions.shtml).
1
A central question in this regard is whether investors trade strategically around the
disclosure threshold, and, if so, to what extent does such behavior affect stock prices. To
study these questions we exploit a unique setting of the EU short selling regulation, which
allows us to observe not only public positions above the disclosure threshold but also
confidential positions below the threshold. Specifically, the regulation requires a two-tier
reporting system for short positions in shares: First, investors must file a confidential
notification with the regulator if their short position reaches 0.2% of the shorted stocks’
issued share capital. Second, the short position must be publicly disclosed if it reaches
0.5% of the issued share capital. For our analysis we obtained both confidential and
public short-sale notifications covering the German stock market, offering us a rare glimpse
behind the veil of the disclosure threshold.
A priori, it is unclear how disclosure requirements for short positions affect investors’
behavior. On the one hand, investors may prefer a disclosure regime because it enables them
to better promote their short positions and induce other investors to sell the target stock.
With this type of publication strategy, the aim is to decrease noise trader risk or to reduce
the time period for maintaining the short positions, effectively reducing securities borrowing
costs (e.g., Kovbasyuk and Pagano, 2015; Ljungqvist and Qian, 2016). Disclosure can also
be used to coordinate predatory short selling and to manipulate the price of the target
stock (e.g., Goldstein and Guembel, 2008; Brunnermeier and Oehmke, 2014). In such cases,
a short seller may prefer a short position above rather than below the publication threshold.
On the other hand, investors may prefer a position below the disclosure threshold for
several possible reasons. They may want to avoid an increase in security borrowing fees
induced by a higher demand of copy-cat investors. Alternatively, they may seek to protect
their private information from such investors, who might mimic their strategies without
incurring the cost of research themselves (Frank, Poterba, Shackelford and Shoven, 2004).
Lastly, some investors have expressed the concern that a public short position may harm
their relationship with the target’s corporate management (AIMA/MFA, 2013), which
2
represents an important source of information when they are conducting fundamental
research. Given that there are reasons both for and against disclosing a short position, it
is ultimately an empirical question as to how investors behave in response to the newly
introduced mandatory disclosure requirement.
Our empirical findings provide strong evidence that a significant share of investors
avoid crossing the disclosure threshold, resulting in positions accumulating just below the
threshold. Specifically, in the reporting interval just below the threshold, the probability
of increasing a short position is lower, and the duration for which such a position is held is
longer, than in all other reporting intervals. Compared to the neighboring intervals, the
duration is 22–55% longer, and the probability of increasing the short position is 20–34%
lower. These patterns emerge only when investors approach the publication threshold for
the first time from below, not for positions that were already public, suggesting that the
pattern observed in the data is the result of a strategy to avoid crossing the disclosure
threshold.
As previously discussed, there are different reasons for investors to remain ‘below the
radar’. To shed more light on the rationale of the short sellers’ behavior, we study a series of
investor and stock characteristics around the disclosure threshold as far as data availability
allows. Interestingly, we find no evidence for the ‘security borrowing cost hypothesis’:
Positions just below and just above the disclosure threshold do not significantly differ
with respect to either ex-ante risk of rising borrowing fees, level of securities lending fees,
supply of stocks to borrow, inventory concentration, or stock liquidity. Instead, disclosure
avoidance appears to be persistent over time, with investors adhering to their secretive
behavior. Moreover, avoidance is associated predominantly with certain investor-specific
characteristics. Secretive positions tend to be held by hedge funds, private funds, investors
located in tax haven countries and investors for whom we are unable to find any public
ownership filing. These findings suggest that ensuring that a short position does not exceed
the publication threshold is part of a general strategy of investors to conceal company- or
3
security-related information from the public and to remain largely out of the spotlight.
Moreover, we find that stocks with secretive short positions exhibit stronger negative
returns than stocks with non-secretive positions, which suggests that the concealment of
short positions is also associated with superior information. Furthermore, approximately
half of the performance of secretive positions is generated on firm-specific news release days,
indicating that these investors rely largely on fundamental trading strategies. These findings
are consistent with two possible explanations. Short sellers may be protecting their private
information from copy-cat investors to maintain their competitive advantage. Alternatively,
investors may fear that disclosing a short position would harm their relationship with
the target’s management, which they perceive as vital in enabling them to collect and
produce fundamental information. By its very nature, the exact process by which private
information is produced is hardly ever observable to researchers. Therefore, the data at
hand do not allow us to disentangle these two non-mutually exclusive explanations.
Lastly, we investigate how investors’ evasive behavior with respect to the disclosure
threshold affects asset prices. If the disclosure threshold discourages informed investors
from increasing their positions any further, then negative information may be incorporated
more slowly into prices. To test this hypothesis, we study stock returns when a secretive
investor is likely to be “constrained” by the threshold. Specifically, when these investors
are just below the threshold, we find subsequent negative risk-adjusted returns of around 6
bp per day. To confirm that the return predictability originates from disclosure avoidance,
we perform two different placebo tests. First, the return predictability does not occur
when we choose various hypothetical publication thresholds above and below the true
one. Second, within secretive positions, subsequent negative returns occur mainly when
the position is just below the publication threshold and less so when the position is not
directly below it.
Our findings contribute to the longstanding debate on the effects of information
disclosure in financial markets. While conventional wisdom suggests that disclosure
4
improves market quality, some theories indicate that it can also have potentially damaging
welfare implications, such as crowding out the acquisition of private information (Verrecchia,
1982; Diamond, 1985), reduction of risk-sharing opportunities (Hirshleifer, 1971) or investors
overweighting public signals (Morris and Shin, 2002). In this context, our paper documents
an unintended consequence of holder disclosure: Position disclosure rules discourage
informed market participants from fully exploiting their information, thereby reducing
price discovery in the market. This mechanism is closely related to the literature on the
trade-off between insider regulation and market quality (Leland, 1992; DeMarzo, Fishman
and Hagerty, 1998; Huddart, Hughes and Levine, 2001). Two features make the disclosure
rule we are examining particularly interesting for our understanding of holder disclosure
and informational efficiency. First, the rule likely conveys highly relevant information to
other market participants because positions must be publicized as early as the next trading
day. Second, the rule applies to short sellers, who are typically perceived as informed
investors.3 Changes in the behavior of these informed investors can significantly affect
the ability of financial markets to produce and aggregate private information in the price
discovery process.
Furthermore, our study adds to the literature on limits to arbitrage (e.g., Shleifer
and Vishny, 1997; Gromb and Vayanos, 2002), in particular short-sale constraints. A
widely studied source of short-sale constraints are frictions in the securities lending market
(e.g., Jones and Lamont, 2002; Asquith et al., 2005; Nagel, 2005; Cohen, Diether and
Malloy, 2007; Prado, Saffi and Sturgess, 2016). In this respect, the novel finding in our
paper is that requirements for timely disclosure also constitute a sizable impediment to a
considerable share of short sellers, and, in particular, to well-informed investors. Short
sellers’ disclosure avoidance is unrelated to previously documented frictions in the securities
3For example, the model by Diamond and Verrecchia (1987) predicts that short sellers are more likelyto be informed than uninformed because shorting is costly. A robust finding of the empirical literatureis that short sales are followed by negative returns, suggesting that short sellers are generally informedinvestors (e.g., Seneca, 1967; Aitken, Frino, McCorry and Swan, 1998; Asquith, Pathak and Ritter, 2005;Boehmer, Jones and Zhang, 2008; Diether, Lee and Werner, 2009). For a recent survey on this topic, seeReed (2013).
5
lending market, such as insufficient lendable supply, concentrated ownership structure, or
the risk of increasing lending fees (Cohen et al., 2007; Kaplan, Moskowitz and Sensoy,
2013; Prado et al., 2016; Engelberg, Reed and Ringgenberg, 2017).
Lastly, our paper contributes to previous work analyzing the effects of short-sale regula-
tion (e.g., Diether et al., 2009; Battalio and Schultz, 2011; Grundy, Lim and Verwijmeren,
2012; Boehmer, Jones and Zhang, 2013; Beber and Pagano, 2013; Fang, Huang and Karpoff,
2016). The consequences of short-sale bans following the financial crisis of 2007–2008
have generated a great deal of interest among researchers. For instance, Boehmer et al.
(2013) and Beber and Pagano (2013) document a deterioration in market quality following
short-sale bans. More recently, Duong, Huszar and Yamada (2015) and Jones, Reed and
Waller (2016) have studied the effects of short-sale disclosure rules. Jones et al. (2016), the
paper closest to ours, use the staggered introduction of disclosure regimes in Europe and
document a reduction in short selling activity and price efficiency. Their evidence points to
a change in investors’ behavior in response to greater transparency. We corroborate their
findings and show that in particular informed short sellers avoid disclosure. Furthermore,
our unique setting allows us not only to identify directly investors’ avoidance of crossing
the publication threshold but also to shed new light on their underlying rationale. Finally,
we establish a direct link between investors’ disclosure avoidance and slower price discovery
for affected stocks.
I. Institutional background and theory
A. Background on the short position disclosure rule
The EU regulation on short selling (No 236/2012) has been in effect since November 1,
2012, requiring investors to report and disclose any short positions of a certain size. The
regulation brought into force a two-tier reporting system: First, a net short position must
be reported to the regulator if the position reaches 0.2% of the issued share capital of the
6
company shorted and for each 0.1% threshold above that. Second, a net short position
must be disclosed to the public if the position reaches 0.5% of the issued share capital of
the company shorted and for each 0.1% above that. Also, notifications have to be updated
when values fall below the relevant thresholds.
The notification and disclosure rules apply to all stocks for which the principal trading
venue is in the EU. Short positions are reported separately for each country on the websites
of the national authorities. In Germany, the national authority for reporting short positions
is the Federal Financial Supervisory Authority (Bundesanstalt fur Finanzdienstleistungsauf-
sicht, or BaFin), and short positions are published on the Internet platform of the Federal
Gazette (Bundesanzeiger).4 Short positions have to be reported or disclosed by 3:30 p.m.
(local time) on the next trading day after they arise. Disclosures contain the name of the
investor, the date of the short position, the International Securities Identification Number
(ISIN), and the name of the shorted stock, as well as the magnitude of the position reported
as a percentage of the issued share capital.
The following example illustrates the disclosure rule. In autumn 2015, the hedge fund
company Marshall Wace LLP shorted the stock of Deutsche Lufthansa AG considerably,
presumably in light of the airline’s restructuring plan and labor disputes with its pilots
and cabin crews. Marshall Wace’s net short position in Lufthansa stock exceeded the
publication threshold of 0.5% on October 29, 2015, with a reported value of 0.59%. On
November 2, 2015, the position exceeded the next reporting threshold of 0.6%, with a
reported value of 0.61%, and on November 5, 2015, the next threshold of 0.7% was crossed,
with a reported value of 0.71%. Thus, short positions are publicly disclosed when each
new reporting threshold is crossed, until the position falls below the threshold of 0.5%.
The regulator receives confidential short position notifications in the same manner, but
here the threshold is 0.2%.
4For recent examples of published net short positions in Germany, see: https://www.bundesanzeiger.de/nlp.
7
After the reporting day, the exact value of a short position between the two disclosure
thresholds is unknown until a new threshold is crossed. That is, the reported short position
of 0.61% on November 2, 2015, could, on the following day, have ranged from 0.60% to
0.69%, since no further information was available about the crossing of another threshold.
Therefore, we sort the positions into short position bins of 10 basis points (bps) each:
0.20–0.29%, 0.30–0.39%, 0.40–0.49%, and so forth. For brevity, we refer to these reporting
intervals as the 0.2, 0.3, 0.4, ... reporting bin or interval.
Several features of the regulation and its scope need to be highlighted. First, the
disclosure rule applies to all investors, irrespective of whether they are domiciled in the
EU or abroad. In fact, a large proportion of the reporting position holders are hedge
funds domiciled outside the EU. Second, the regulation applies not only to short positions
but also to derivative positions, which must be accounted for on a delta-adjusted basis.
Thus, reporting requirements cannot be circumvented by substituting short positions
with positions in derivatives. Third, the reported net short positions are calculated
by netting all the long, short, and delta-adjusted derivative positions in the respective
stock. Consequently, a short position established to hedge an exposure originating from a
derivative position in the same underlying stock does not have to be reported or disclosed
as long as the overall shorting exposure is below the respective threshold. Lastly, market
making activities are exempted from the EU short-selling regulation for the purpose of
ensuring liquidity provision.5
B. Short positions: To publish or not to publish
A priori, it is unclear whether investors prefer to publicize their short positions or to
keep them secret. Lamont (2012) states that, depending on the situation, short sellers
5Exemptions include market making in a specific stock, but also market making in the related derivativesof that stock. To meet the conditions for this exemption, institutional investors are required to file adetailed statement on their market-making activities in specific securities, which are monitored by thenational authorities (ESMA, 2013). According to the list of market makers published by ESMA, it ismainly banks that are using this exemption (ESMA, 2016).
8
might either publicize their positions or try to remain undetected. Publicizing a short
position could be helpful if the investor is seeking to overcome conventional limits to
arbitrage and attempting to convince other investors that a certain stock is overpriced (e.g.,
Kovbasyuk and Pagano, 2015; Ljungqvist and Qian, 2016). If other investors agree and
follow suit, prices will converge faster to fundamentals, thus reducing borrowing costs for
the stock and potential noise trader risk (De Long, Shleifer, Summers and Waldmann, 1990;
Shleifer and Vishny, 1997). For instance, Ljungqvist and Qian (2016) provide evidence
that such short-sale campaigns help in the correction of mispricing, especially in cases of
questionable governance and accounting practices.6 Alternatively, position disclosure may
facilitate the coordination of predatory short-selling strategies to potentially manipulate
stock prices (e.g., Goldstein and Guembel, 2008; Brunnermeier and Oehmke, 2014). While
the two types of short-selling strategies differ in terms of the underlying motives, in both
cases we expect investors to prefer a short position above the publication threshold to one
below it.
At the same time, the literature offers a number of reasons why short sellers may want to
keep their short positions secret and may prefer to avoid crossing the publication threshold.
If other investors follow suit in shorting the stock, existing stock loans may be recalled
by the lender, or borrowing fees may rise. Investors may also be concerned about their
intellectual property, such as information they have gathered about the shorted company
or a proprietary trading strategy which they do not want to reveal to their competitors
(Frank et al., 2004). The publication of a short position may adversely affect an investor’s
ability to access the corporate management of the target, an important source for compiling
fundamental information. The Alternative Investment Management Association (AIMA)
and the Managed Funds Association (MFA) report anecdotal evidence that some hedge
funds “have been excluded from corporate access events as a consequence of their short
6Although such short-sale campaigns have become more popular in recent times, they are still compar-atively rare. In particular, Ljungqvist and Qian (2016) collect 124 short-sale campaigns in the UnitedStates used by 31 individuals or small boutique hedge funds during 2006–2011.
9
positions, notwithstanding the fact that the same managers may have had long positions
in the past, or elsewhere in their firms or in other of their funds” (AIMA/MFA, 2013).
Similarly, Lamont (2012) even documents cases where short sellers were subject to legal
action or harassment when their name and short positions became public.
Given that there are reasons both for and against publicizing a short position, it is
ultimately an empirical question as to whether greater transparency for short sellers and
their positions will result in eagerness to disclose a position or a behavior to avoid doing
so.7
II. Data and descriptive statistics
We obtain public and confidential short position disclosures from the German Federal
Financial Supervisory Authority (BaFin) for the period from November 1, 2012, through
March 31, 2015. We merge the short position notifications with daily stock data from
Thomson Reuters Datastream, equity lending data from Markit, and institutional investor
data from FactSet Ownership, formerly known as LionShares. For our analysis, we restrict
the sample to common equity traded on the German regulated market. To ensure the
quality of the data from Datastream, we apply several standard data filters (see, Ince and
Porter, 2006; Griffin, Kelly and Nardari, 2010; Karolyi, Lee and Van Dijk, 2012). We
start our analysis on November 5, 2012, to account for some delay in the notification of
short positions, due to a statutory holiday in some federal states. The Appendix provides
further details on the sample construction, and Table A.1 contains a description and the
sources of all variables used in our analyses.
Table 1 provides summary statistics on various stock characteristics for the entire
population of stocks in the regulated market and for stocks that have at least one short
7In this paper we focus on short sellers’ behavior around the disclosure threshold of 0.5%. It is possiblethat the two-tier reporting system impacts the behavior at 0.2%, where positions are reported to theregulator but not to the public.
10
position notification. Of all stocks, 19.9% have at least one short position notification.
In particular, 8.3% of the sample consists of stocks that have at least one public short
position disclosure and, 11.6% of stocks have at least one confidential but no public
position. In terms of stock characteristics, we observe short position notifications mainly
for stocks with a large market capitalization, a low book-to-market ratio, and a high share
of institutional investors, as well as stocks that are very liquid, measured by both the
Amihud illiquidity ratio and the bid–ask spread. In fact, the vast majority (73.5%) of
stocks with short positions are in the highest quartile of market capitalization, and almost
all stocks with short position notifications (95.8%) appear above the median value of the
market capitalization distribution. Data from the lending market show a similar picture:
Stocks in our sample are easy and cheap to borrow, as indicated by the larger proportion
of actively lendable stocks and lower borrowing fees relative to the overall German stock
universe. In economic terms, there are no apparent differences between stocks with public
and confidential short positions.
The panel dimension of the analyses in Sections III and IV pertains to the investor-stock
level. Table 2 contains summary statistics for the stock and investor characteristics for
which we observe a short position of at least 0.2% of issued share capital. As these details
indicate, hedge funds constitute the largest investor group, accounting for 76% of the
observations. The remaining groups of investors are mutual funds and other investment
advisors. With respect to the location of the investors, European and U.S. institutions
account essentially for the entire sample, contributing in almost equal proportions. For
just under 10% of the observations, we cannot find any public filings in the comprehensive
FactSet ownership database relating to the investor’s (long) holdings. Finally, a similar
but distinct investor group are the private funds, investors who filed Form D with the SEC
and sought private placements under Regulation D. Private placements are not subject to
some of the laws and regulations designed to protect investors, such as the comprehensive
disclosure requirements that apply to registered offerings. Therefore, the data set not only
11
offers a rare glimpse into the short-selling behavior of investors, it also allows us to analyze
investor groups not previously studied, due to their generally secretive behavior.
Figure 1(a) shows the frequency distribution of days with an open short position
notification over the different reporting bins. As indicated previously, short positions are
reported to the regulator when they are greater than or equal to 0.2% of the total shares
outstanding, and they are disclosed to the public when they are greater than or equal to
0.5%. As can be seen from the graph, publicly disclosed short positions represent only the
tip of the iceberg. The majority of short positions are not disclosed: 79% of observations
with an open short position fall below the publication threshold.
III. Do investors avoid crossing the disclosure
threshold?
Looking at the overall distribution in Figure 1(a), it is hard to determine whether
investors avoid crossing the disclosure threshold. To uncover a potential accumulation
of short positions below the disclosure threshold, we split the sample into positions both
above and below their historic high. Specifically, we define the following dummy variable
for the sequence of position notifications of each investor-stock (i, j) pair:
Position record highi,j,t =
1 if bini,j,t = max
s≤tbini,j,s
0 if bini,j,t < maxs≤t
bini,j,s,(1)
where bini,j,t denotes the short position’s reporting bin of investor i in a given stock j on
trading day t. A sample split according to Equation (1) exploits the fact that avoidance
of crossing the threshold should be particularly pronounced for investors who approach
the threshold from below for the first time. If we compare two different situations where
investors are just below the disclosure threshold, we can see the economic logic behind this
12
approach. First, imagine an investor being short in a specific stock with a position value
of 0.4, and this position is at a record high. As this position has not been higher in the
past, it has also never been public. If an investor is avoiding crossing the threshold, we
would expect to find evidence of avoidance especially in this situation. Now, imagine an
investor with a short position in the 0.4 bin in a specific stock, but the position is below
its record high. This position has been higher in the past and consequently has also been
public. In this situation, we would expect either fewer or no signs of avoidance. The idea
is that investors who revealed the particular position in the past are unlikely to be averse
to cross the publication threshold again in the near future.
Figure 1(b) shows the frequency of days with open short positions for the two subsamples.
Positions at their record high, for which we expect there to be strong signs of avoidance,
amass below the disclosure threshold, suggesting that some investors prefer to stay below
the threshold. The relative frequency in the 0.4 bin is 17.0%, almost at the frequency of the
lower bin. Positions that are below their record high, however, decline fairly geometrically
with increasing reporting bins. Their relative frequency in the 0.4 bin is 8.6%, about half
the relative frequency of positions at their record high. Overall, Figure 1(b) provides initial
descriptive evidence that some investors are trying to stay below the radar.
We now take a more rigorous approach to testing the avoidance hypothesis. First,
we study the probability that the next position change is an increase across reporting
intervals. Second, we investigate the duration spent in each reporting interval. Lastly,
we employ different regression models, in which we control for various measures of short-
sale constraints as well as for unobservable stock and investor effects using saturated
fixed-effects specifications.8
8Due to data reporting issues our methods differ from the approach commonly used to identifydiscontinuities in economic outcomes, in which the density of the variable of interest is estimated justbelow and above a certain threshold (e.g., Degeorge, Patel and Zeckhauser, 1999). This method requiresa continuous outcome variable with a smooth density under the null hypothesis, which is not given inour case. Positions are reported at a two-digit precision only on the day at which the position crossesthe threshold of a reporting bin. Afterwards, the position can take any value between the two thresholdswithout additional reporting (see also Section I.A). Such a setting leads to a distorted picture of how thepositions are distributed within an reporting interval. If a position crosses the 0.4 threshold from below, it
13
A. Probability of a short position increase
Our first hypothesis of avoidance relates to the probability of increase. If there is
avoidance, the probability of a position increase should be lower in the bin just below
the threshold than in the neighboring bins. Table 3 shows the estimated probability that
the next position change is an increase, conditional on currently having a short position
in a specific reporting interval. For this test we leave aside the time dimension of the
positions and we infer our results entirely from the decision as to whether the next change
in investor’s position will be an increase or a decrease. Looking first at the overall sample,
we find that the probability of an increase rises with the value of the current reporting
interval. Beyond this general pattern, we observe an unusual value for the 0.4 reporting
interval. The bin just below the publication threshold exhibits the lowest probability of all
the reporting intervals, which is significantly different from all other bins except the 0.2
bin. This unusually low likelihood of increasing a short position just below the threshold
suggests that some investors avoid crossing the disclosure threshold.
To determine whether this effect is really due to an avoidance strategy, we split the
sample into positions at and positions below their record level. As discussed previously,
avoidance of passing the disclosure threshold should be found particularly when positions
are at a record high, but it should be less pronounced when positions are below their
record high. The second panel shows the probability of increasing a short position only
for positions at their historic high. For this subsample, the avoidance effect of the 0.4 bin
is much stronger than it is for the overall sample. The probability of increasing a short
position takes a minimal value of 0.338 for the 0.4 bin, significantly lower than for all
other reporting intervals except for the lowest. To gauge the economic significance of this
is more likely to have a low rather than a high second digit on that day. Changes in the second digit arenot incorporated into the distribution. In contrast, when the 0.4 is crossed threshold from above, theposition is more likely to have a high rather than a low second digit on that day. This reporting rulemechanically creates a u-shaped distribution of positions within a reporting corridor. For a more detaileddiscussion on the specifics of the reporting rule and the resulting distribution of positions, we refer thereader to the Internet appendix.
14
avoidance effect, we compare this probability to that of the neighboring reporting intervals,
just below and just above. In the 0.3 bin, the probability of increasing a short position is
0.423, equating to a difference of -0.085; in the 0.5 bin, it is 0.514, a difference of -0.176.
In relative terms, in the bin just below the disclosure threshold it is 20% and 34% less
likely to increase a short position than in the two neighboring intervals.
Positions that are below their maximum serve as a comparison. If the low probability
of increasing a short position is truly due to an investor wanting to avoid revealing his or
her position, we should observe a smaller effect or no effect at all for these positions. The
right-hand panel of Table 3 supports this notion: For positions below their record high, we
find nothing extraordinary about the 0.4 reporting interval. The probability of increasing
the position is in fact the second largest, not statistically significant when compared to
the class below, but significantly larger than the class above.
B. Duration in reporting bins
The second testable hypothesis for avoidance concerns to the duration spent in the
reporting intervals. If investors avoid crossing the disclosure threshold, they will remain
just below it, thereby spending longer time in the 0.4 reporting bin. We test this hypothesis
in Table 4, which shows the average duration in trading days in each reporting interval.
The durations in the bins are prone to extreme outliers, so we winsorize the upper tail
at 1% before reporting the mean durations in Panel A. As an alternative, we report the
median durations in Panel B.
In general, the duration in each 0.1 interval declines as the position grows in size. For
example, the mean duration for the overall sample in Panel A declines from 18.3 days
(lowest bin) to 12.6 days for all positions greater than or equal to 1.0% (highest bin).
Again, we observe an unusual value for the 0.4 bin: The duration in this bin, just below
the disclosure threshold, is the longest of all the bins, at 20.6 days. The difference in mean
duration is statistically significant when compared with all other reporting intervals. As in
15
our previous analysis, we discover that the pattern of the overall sample is driven entirely
by the positions at their record high, for which the maximum duration of 26.0 days is
reached in the 0.4 bin, significantly higher than for any other bin. For positions below
their record high, we find no unusual value for the bin just below the publication threshold;
instead, the durations decline fairly monotonically with the position value. The same
pattern can be observed for median durations: For positions at their record high, the
median duration is 10 days for the reporting bin just below the publication threshold,
significantly higher than for any other reporting interval. For positions below their record
level, however, durations decline monotonically.
To illustrate the economic magnitude of the duration effect, it is again useful to compare
the duration of the 0.4 bin to that of its neighboring bins. The mean duration in the bin
just below the publication threshold of 26.0 days is 22% higher than for the next lowest
bin (21.3 days), and 55% higher than for the next highest bin (16.7 days). If we look at
the medians, the results are similar: The median duration of 10 days in the 0.4 interval is
25% higher than for the next lowest interval (8 days), and 43% higher than for the next
highest interval (7 days).
In summary, investors spend an abnormally long time in the reporting bin just below
the publication threshold, significantly longer than in any other reporting bin. In this bin,
the likelihood of increasing the short position is also the lowest. The economic magnitude
of these effects is substantial: Compared with the neighboring bins, the duration in this bin
is 22–55% longer, and the probability of a short position increase is 20–34% lower. Overall,
these combined results suggest that a considerable share of investors avoid crossing the
disclosure threshold.9
9It is worth noting that these results describe the aggregate behavior of investors around the disclosurethreshold. The results are not at odds with the possible existence of individual short-sale campaigns, forexample, as documented by Ljungqvist and Qian (2016).
16
C. Accounting for (un)observed stock and investor characteristics
Our findings that there is both an abnormally low probability of increase and an
abnormally long duration in the bin just below the publication threshold indicate that a
large share of investors avoid increasing their positions because of the publication threshold.
However, this lower probability of increase might stem from stock or investor characteristics
that correlate with a position being just below this threshold. To rule out alternative
mechanisms, we conduct panel regressions with a battery of control variables associated
with short-selling constraints as well as progressively saturated fixed-effects specifications
in the spirit of Jimenez, Ongena, Peydro and Saurina (2014).
We start with a standard binary outcome model and construct a dependent variable
equal to 1 on day t if a short position moves up at least one bin on the next trading day,
and equal to zero otherwise: yi,j,t+1 = 1 (bini,j,t+1 > bini,j,t). To identify whether short
sellers avoid crossing the disclosure threshold, we specify the following regression models:
yi,j,t+1 = α0 + β1 Just below thresholdi,j,t +∑k
βk k bini,j,t + γ′ Xi,t + αt + εi,j,t+1, (2a)
yi,j,t+1 = α0 + β1 Just below thresholdi,j,t +∑k
βk k bini,j,t + αi,t + αj,t + εi,j,t+1. (2b)
where Just below threshold is a dummy variable, indicating the reporting bin just below
the publication threshold (bin = 0.4). The indicator k bini,j,t is equal to 1 if investor j has
a short position in stock i in bin k at time t. In our specifications the omitted benchmark
bin is 0.3, i.e. k ∈ {0.2, 0.5, 0.6+}, where 0.6+ indicates position bins greater than or equal
to 0.6.10 Xi,t represents a vector of control variables related to short selling activity and
αt, αi,t, αj,t denote day, stock-day, and investor-day fixed effects, respectively. If investors
are on average avoiding crossing the publication threshold we would expect the coefficient
of the Just below threshold dummy β1 to be significantly negative relative to the omitted
neighboring 0.3 bin and significantly lower than the regression coefficient of the 0.5 bin
10Our results are robust to using a more detailed specification that includes all reporting bins up to 1.0into the regression separately. These results are shown in the Internet appendix.
17
dummy.
Our first benchmark specification includes daily time-fixed effects in addition to the
reporting bin indicators. Column (1) of Panel A in Table 5 reports the estimates of this
specification. The results are in line with our previous findings. The probability of an
increase conditional on being in the bin just below the disclosure threshold is significantly
lower than in the 0.3 bin. In economic terms, when a short seller is in the 0.4 bin, the
position is 0.54 percentage points less likely to be increased than if the short seller were in
the 0.3 bin. This effect is economically significant, given that the baseline unconditional
probability of an increase on the next day is slightly below 2.0%. In Panel B we observe
that the probability of increase in the bin just below the threshold is also lower than in
the neighboring 0.5 public bin.
The second specification includes various control variables that might be endogenously
related to different dimensions of short-sale constraints. Most importantly, the coefficients
of the bins around the disclosure threshold remain virtually unchanged (see Column (2)
of Panel A and B). Although our main interest is still the coefficient of the 0.4 bin, this
subsection also sheds some light on other drivers of short-selling activity. For example, the
bid–ask spread is negatively associated with the probability of position increase, suggesting
that investors are more reluctant to build up a short position in illiquid stocks. We find
similar but statistically weaker evidence for the Amihud (2002) price impact measure.
Consistent with Cohen et al. (2007) and Kaplan et al. (2013), the positive and statistically
significant coefficient on the percentage of lendable stocks suggests that there are more
short position increases if there is larger supply of stocks to borrow. In line with Prado et al.
(2016), we find a negative relation between inventory concentration and the probability of
a short position increase. Although in our sample the fee itself is not associated with the
probability of a short position increase, we observe that the second moment of the securities
lending fee is negatively associated with short position increases. This finding is consistent
with the notion that investors are concerned with high variation in fees when establishing
18
their position (D’Avolio, 2002; Engelberg, Reed and Ringgenberg, 2017). Furthermore,
consistent with the idea that it is difficult to establish the same relative position in larger
stocks, market capitalization relates negatively to the probability of an increase. As
described in Section I.A, the reported net short positions also include equivalent derivative
positions, which are accounted for on a delta-adjusted basis. Thus, if futures or listed
options exist for an underlying stock, it is presumably easier to establish and increase
a net short position. In keeping with this notion, the dummy indicating the existence
of futures or listed options is associated with a positive coefficient. Interestingly, when
we control for the above-mentioned short-selling frictions, we observe that investors tend
to increase their short position more aggressively for stocks with higher return volatility.
Such an effect is consistent with the notion that low-volatility stocks are insufficient to
provide short sellers with the expected level of profit, one which exceeds their transaction
costs. As a consequence, they demonstrate a preference for high-volatility stocks (Angel,
Christophe and Ferri, 2003).
In Column (3), we exploit the three-dimensional structure of our data set by including
stock-day and investor-day fixed effects to suppress the influence of stock and investor
unobservables on the decision to increase a short position. The fixed effects control for
characteristics, such as the overall borrowing demand for the stock, the supply of stocks
to borrow, borrowing fees, or liquidity, but they also control for any time-varying investor
characteristic such as the investor’s size or leverage, or any sort of funding constraint
that may vary at investor level. The identification of investors’ avoidance to cross the
publication threshold stems from comparing the probability of an position increase on the
same day by different short sellers for the same stock, controlling for any time-varying
investor characteristic. The estimated regression results in Column (3) indicate that neither
unobserved stock-specific nor investor-specific characteristics explain the abnormally low
probability in 0.4 bin. That is, the probability of an increase when the investor is in the
bin just below the publication threshold is on average significantly lower than in the next
19
lower bin (Panel A) but also than in any other reporting bin (Panel B).11 Taking the
results in Table 5 together, the avoidance of a position increase in the bin just below the
threshold is robust to any time-varying stock or investor characteristic.
IV. Understanding investors’ avoidance of position
disclosure
In the previous section, we revealed that short sellers avoid passing the publication
threshold to a considerable extent. As discussed in Section I, there are various possible
motives for concealing one’s short position. Investors may fear rising costs of borrowing
securities or stock recalls induced by position publication and trading by copy-cat investors.
Other reasons for staying below the radar may be to protect one’s private information
from competing investors or to preserve a valuable source of information, i.e. retain one’s
access to the corporate management of the target company. The purpose of this section is
to examine these various, mutually non-exclusive explanations for avoiding disclosure.
A. Investor and stock characteristics of secretive positions
To learn why a large share of investors keep their positions secret, we compare positions
that remained just below the disclosure threshold with positions that just exceeded it.
Therefore, we calculate for each position, i.e. each investor-stock pair (i, j), the maximum
reached:
Position maximum i,j = maxt=1,2,...,Ti,j
bini,j,t, (3)
where Ti,j is the total number of trading days of investor-stock pair (i, j). To understand
to what degree avoidance is investor-specific, we calculate the maximum position value
11The inclusion of the different fixed effects reduces the number of observations in some of the estimatedmodels due to missing variation. However, this loss of observations is relatively small in this rich dataset. With both stock×time and investor×time fixed effects, the number of observations with sufficientvariation within stock and investor level, accounts for 73% of the overall sample size.
20
reached by each investor across all of his or her positions:
Investor maximum i = maxj=1,2,...,Si
maxt=1,2,...,Ti,j
bini,j,t, (4)
where Si is the total number of stocks in which investor i holds short positions.
Figure 2 shows the frequency of short positions according to the maximum bin reached
and thereby reveals two important patterns. First, it captures the avoidance effect we
have already shown: Positions that never crossed the disclosure threshold (positions with
a maximum of 0.4) account for the largest group with 22.8% of all open short position
days. Additionally, a second important finding emerges from Figure 2: Looking at the 0.4
reporting bin, we observe that a large share of positions with a position maximum of 0.4
originate from investors whose positions reach a maximum of 0.4. In other words, many
positions that remain just below the disclosure threshold are held by investors who remain
below it for the entire sample period. For positions with other maximum reporting bins,
the corresponding proportion is strikingly lower. This descriptive graph suggests that the
decision to stay below the radar is mainly investor-specific, with investors sticking to their
general decision to keep their short positions secret.
Now, to better understand what determines avoidance, we use the definition provided
by Equation (3) and compare positions of similar magnitude both to the left and the right
of the disclosure threshold. We will refer to positions with a maximum of 0.4 as secretive
positions and those with a maximum of 0.5 as non-secretive positions.12 Panel A of Table 6
depicts the average values of several investor and stock characteristics for both position
groups and the differences between them.
As shown in Figure 2, the first variable of interest is a dummy that indicates whether
the position maximum equals the overall maximum reached by the investor. In line with
12Not all positions with a maximum of 0.4 are driven by investors’ avoidance of disclosure. For somepositions a maximum of 0.4 is simply optimal, irrespective of the disclosure threshold. Importantly, though,this fact works against finding any differences in characteristics between the position groups just belowand just above the disclosure threshold.
21
the graphical representation in Figure 2, almost 40% of positions with a maximum of 0.4
originate from investors with the same overall maximum. The corresponding proportion in
the 0.5 bin is only 11%. Economically speaking, a significant proportion of positions that
remain just below the disclosure threshold originate from investors who remain below it
for the entire sample period. This persistence by investors in their decision not to disclose
lends support for the idea that some investors follow a general policy of non-disclosure of
short positions.
Next, we investigate whether persistent avoidance of disclosing short positions is also
related to investors’ general opaqueness across different dimensions. First, hedge funds
are generally considered to be more opaque than other investors. For example, Agarwal,
Jiang, Tang and Yang (2013) found that hedge funds seek confidentiality in their quarterly
long holdings more often than other investors when filing with the SEC. Our second proxy
associated with opacity is whether the investor is domiciled in a tax haven country as
defined by the OECD (2000). Moreover, some investment firms generally avoid disclosure
of company and security-related information as much as possible. The FactSet ownership
database (formerly Lionshares) collects data from public disclosures worldwide. We exploit
the rich coverage of this database, but posit that institutional investors with no public
ownership record – and consequently no entry in the FactSet database – tend to follow a
strategy of opaqueness. Also, institutional investment managers that are active in U.S.
financial markets are obliged to disclose their quarterly holdings with the SEC. However,
under certain conditions investment managers may use an exemption under Regulation
D to offer and sell securities without having to register the offering with the SEC. In
most cases, those private funds are restricted to accredited investors, i.e. high net worth
individuals or institutional investors.13 To determine whether a fund engages in private
13Regulation D includes three SEC rules — Rules 504, 505 and 506 — that institutional investorscan draw on to sell securities in unregistered offerings. Each rule has specific requirements that thefund must meet. While Rule 504 merely restricts the amount of securities offered to 1 million USD,the other two rules are more stringent and require that the funds’ shareholders are wealthy individualinvestors or trusts directed by specialist managers. A more detailed description of the requirementsare documented in the following investor bulletin: https://www.investor.gov/additional-resources/news-
22
placements or unregistered offerings, we search SEC’s Edgar database for Form D filings
and match them to our sample using investor names.
As evident from Panel A of Table 6, compared to non-secretive positions just above the
publication threshold, the secretive positions originate to a larger degree from hedge funds.
The share of positions from investors registered in a tax haven is also rather low overall,
but there are distinct differences to the left and right of the disclosure threshold. Investors
registered in tax haven countries account for 2.17% of non-secretive positions, but for
6.89% of secretive positions – more than three times as much. There are also economically
sizable differences with respect to investors having no public ownership filings and with
respect to private funds. Investors with no FactSet entry account for 17.61% of all secretive
positions; their share of non-secretive positions is only 2.65%. Similarly, private funds
account for 16.66% of all secretive positions, but their share of non-secretive positions is
only 3.40%. To test whether secretive positions differ according to the size of the investor,
we compare the value of all assets provided in the FactSet database. The natural logarithm
of total assets is slightly smaller for secretive positions than for non-secretive positions,
and the difference between the two is marginally significant at the 10% level. However,
this result should be interpreted with caution because information about portfolio assets is
not available for all investors. In particular, there is less availability for secretive positions
than for non-secretive positions – a direct outcome of the opaqueness of these investors.
Overall, compared to positions just above the publication threshold, positions just
below it are held more by hedge funds, private funds, investors domiciled in tax havens and
investors for whom no public ownership filings are on record. This suggests that secretive
behavior in short selling is part of the investors’ general strategy to conceal company- or
security-related information from the public and to remain largely out of the spotlight.
Besides these investor-specific drivers of avoidance, there may also be stock character-
istics which influence this decision to keep positions below the radar. In particular, an
alerts/alerts-bulletins/investor-bulletin-private-placements-under
23
alternative hypothesis is that investors are concerned about rising borrowing costs or even
that a position disclosure may lead to a stock recall because of action by copy-cat investors.
If this is the case, we would expect investors to avoid disclosure for stocks where the risk
of rising fees or of a recall is likely to be high. We test this borrowing cost hypothesis by
comparing secretive and non-secretive positions on various dimensions of this risk.
Liquidity is crucial for an investor who wants to cover his or her short position at
the right time or who is forced to do so by a recall. If investors avoid disclosure because
they fear a recall induced by publication, we expect avoidance to be stronger for illiquid
than for liquid stocks. Panel A of Table 6, however, shows that secretive positions do
not differ from non-secretive positions in terms of market capitalization, turnover, bid-ask
spread, or the Amihud illiquidity measure. Secretive positions are slightly more volatile
than non-secretive positions, but the magnitude is small and statistically insignificant.
Impediments to short selling arise if there is insufficient supply of lendable stocks.
Moreover, Prado et al. (2016) highlight the importance of ownership structure, in particular
ownership concentration, in the supply of stocks to borrow. In our context, an insufficient
supply of stocks to borrow may result in higher borrowing fees once copy-cat investors
follow. However, comparing positions to the left and right of the disclosure threshold
provides little support for this hypothesis. Namely, secretive positions are similar to
non-secretive positions in terms of institutional ownership – a commonly used proxy for
inventory supply – and the share of lendable stocks reported by Markit. There are also no
significant differences in terms of lender and inventory concentration.
Short squeezes are likely to occur when the percentage of shares on loan relative to loan
supply (utilization) is at a high level. If investors fear short squeezes or rising borrowing
fees after publication, they should avoid disclosure when utilization is high or, by the same
argument, when short-selling fees are already high. As evident from Table 6, secretive
positions are not significantly different from non-secretive positions in terms of the average
24
utilization and lending fee.14 We also do not find any difference in the tail of the fee
distribution as the percentage of stocks trading “on special”, i.e. with a lending fee above
100 bps, is not significantly different between the two types of positions.
D’Avolio (2002) notes that short sellers are not only concerned about the level of
fees, but also require a premium for the volatility of fees, and empirical support for
this hypothesis is provided by Engelberg, Reed and Ringgenberg (2017). Stocks with a
high variance in borrowing fees are associated with higher risk and publicizing a short
position for these stocks might trigger an increase in fees. Following Engelberg, Reed and
Ringgenberg (2017), we calculate different state variables that predict future loan volatility.
The proxies include the standard deviation of loan fees and utilization measured over the
past year and two tail risk measures on loan fees and utilization. The tail risk measures
are defined as the 99th percentile of a normal distribution using the estimated mean and
standard deviation over the past year. Similar to our previous findings, none of the four
short-selling risk proxies differ significantly for secretive and non-secretive positions.15
Taking the results together, we find no evidence that stock liquidity, inventory supply,
concentration of supply, utilization, borrowing fees or short-selling risk are associated with
avoidance by the short sellers of crossing the publication threshold. The avoidance effect,
however, is related to investor characteristics, and particularly to investors’ general policy
of opaqueness.
14We employ the expected borrowing fee for a hedge fund on a given day as defined by Markit. Theyderive a rate using their proprietary analytics and data set. The calculation uses both securities borrowingcosts between agent lenders and prime brokers as well as rates from hedge funds to produce an indicationof the current market rate.
15It is important to note that the size of a short position is also determined endogenously by the costsassociated with borrowing and short selling the stock as shown in Section III.C. In particular, smallpositions may be small (i.e., below the radar) because the stock is difficult to short (e.g., low supply, highfees, high short-selling risk) irrespective of the publication threshold. Consequently, we even expect thatpositions just below the radar will be to some extent associated with higher (but not abnormally higher)frictions than positions just above the publication threshold. However, we do not account for this patternin our tests given that even without adjusting our null hypothesis to a positive value, we do not findeconomically and statistically significant differences between secretive and non-secretive positions in termsof the borrowing costs hypothesis.
25
B. Are secretive investors better informed?
One possible explanation for investors concealing their positions is that they wish
to protect their private information or their proprietary investment strategies. We now
explore this hypothesis by comparing the performance of positions to the left and the right
of the disclosure threshold. If non-disclosure is associated with superior information, we
expect that stocks with secretive short positions to have lower risk-adjusted returns than
stocks with similar but disclosed short positions.
To compare the performance of the two position groups, we form a portfolio of stocks
for each group, weighted by the number of positions with a maximum of 0.4 and 0.5,
respectively. A stock is included in the portfolio if the investor concerned established
a short position greater than 0.20% the day before, and a stock is excluded from the
portfolio if the investor’s position fell below 0.20% the day before. This timing convention
is conservative in that it assumes investors to be trading at the end of each day. We
estimate risk-adjusted returns (alphas) of the two portfolios employing different factor
models including the Capital Asset Pricing Model (CAPM) (Sharpe, 1964; Lintner, 1965),
the Fama and French (1993) three-factor model, and the Carhart (1997) four-factor model.
To estimate standard errors, we follow Newey and West (1987), with the lag length being
selected according to the optimal lag-selection algorithm proposed by Newey and West
(1994).
The results of the performance comparison are depicted in Table 6, Panel B. As is
evident from the alpha estimates, short positions with a maximum of 0.4 have on average
a negative alpha of -4.12 to -4.74 bps per day, which is statistically significant across
all three factor models. In contrast, short positions with a maximum value of 0.5 have
alphas that are not significantly different from zero at any conventional level. Accordingly,
secretive positions on average outperform non-secretive positions by around 5 bps per
day. This difference in mean returns is robust across all factor models and statistically
significant at the 1% level. Since a considerable share of positions remaining just below
26
the disclosure threshold originate from investors who never disclose their positions, our
second step is to narrow down the portfolio to the positions of this subgroup. If secrecy is
associated with having better information about a stock’s future value, we expect these
very secretive positions, as defined by the position and investor maximum, to yield even
better performance. Indeed stocks shorted by these investors show an even stronger price
decline with alphas ranging between -4.79 and -5.58 bps per day, although these results
are statistically weaker due to a substantially reduced sample of positions. Overall, these
findings suggest that secretive positions have an informational advantage over non-secretive
positions, providing support for the private information hypothesis.
C. Secretive short positions and fundamental news
Next, we try to understand more about the sources of secretive investors’ informational
advantage. In particular, if the performance of secretive positions stems from superior
information about impending changes in companies’ fundamentals, we expect that (1)
they will tend to short stocks with future negative changes in fundamentals and (2) their
performance can be largely explained by days when news about fundamentals are released
and investors are able to incorporate this information into their decisions (Engelberg, Reed
and Ringgenberg, 2012; Engelberg, McLean and Pontiff, 2017).
To test these hypotheses, we decompose the performance of secretive positions into
two components: days on which firm-specific information releases were made, including
earnings announcements and ad-hoc corporate disclosures, and all remaining days. In their
quarterly earnings announcements, companies inform investors about their cash flows and
this news serves as one of the main sources of fundamental analysis. In addition, publicly
traded companies in Germany have a statutory obligation to disclose immediately to the
public, via ad-hoc news releases, any facts about their company that could potentially
affect the stock price.16 Ad-hoc news includes only firm-specific information relevant to
16The ad-hoc corporate disclosure regulation in Germany is very similar to the U.S. Form 8-K reportingrequirement. However, it is important to mention some differences. First, although the regulator provides
27
the value of the company, and not information relating to the conditions of the overall
market or to certain industries.
To estimate the magnitude of each component, we use the regression-based generalized
calendar-time portfolio approach proposed by Hoechle, Schmid and Zimmermann (2016)
with Driscoll and Kraay (1998) standard errors. The advantage of this approach is that it
allows multiple explanatory variables to be included in a regression, rather than requiring
us to construct portfolios for each variable of interest. Importantly, in this generalized
approach, standard errors are robust to general forms of cross-sectional dependence,
autocorrelation, and heteroscedasticity.
In Column (1) of Table 7, we first replicate the results of the calendar-time portfolio
approach shown in Panel B of Table 6. When the excess return for each position day
is used as the dependent variable and the asset-pricing factors as explanatory variables,
the Hoechle et al. (2016) estimation procedure yields exactly the same point estimates of
alphas and corresponding standard errors as the portfolio approach.17 We then decompose
the returns in those that are generated on days with fundamental news releases and those
generated on days without observed news. In particular, we set the value of the indicators
to one for all position days with a fundamental news release. To ensure that we also
capture the price reaction to fundamental news released after the trading hours, we also
flag the day following an event. The result of the decomposition is depicted in Columns (2)
to (4) of Table 7.
practical guidelines on publishing corporate ad-hoc news, the German Securities Trading Act containsno conclusive list of events that can be defined as influencing the stock price of a company. The SECis more precise in defining the categories of Forms 8-K than the German regulator. Also, in Germany,corporate ad-hoc news is much shorter and less detailed than in the U.S. Lastly, while the SEC made anumber of adjustments to the categories of ad-hoc news, no similar changes have been made recently tothe German regulation. This may be partially explained, however, by the broad definition of ad-hoc newsin the German Securities Trading Act.
17Following Hoechle et al. (2016), we estimate a weighted least squares (WLS) regression with observationweights matching the implicit weighting scheme of the calendar-time portfolio approach. This estimationprocedure not only enables us to make an exact replication of the results in Panel B of Table 6 but alsoallows a robust decomposition of the return when additional explanatory variables are included.
28
First, the analysis shows that secretive investors take short positions in stocks for which
negative earnings announcements and corporate ad-hoc news are made. In particular,
when the CAPM is used, average daily abnormal return of secretive positions amounts
to economically large −30.21 bps on earnings announcement days and even −69.41 bps
on days on which there is corporate ad-hoc news. The other two models give estimates
of a similar magnitude. This finding strengthens the evidence that secretive investors
possess information about company fundamentals that will be released in future corporate
disclosures. The magnitude of returns is sizable, given that the average return on news
days for a given stock is even significantly positive. Second, although those corporate
news days represent only 4% of trading days in a year, around 50% of the performance
of secretive positions is generated on those days, irrespective of the asset pricing model
used. The average negative return of stocks shorted by these investors is 33 (14) times
larger in absolute terms on ad-hoc news (earnings announcement) days than on non-event
days. These numbers indicate that the collection and production of private information on
company fundamentals plays a key role in the success of secretive positions. Note that the
remaining days may include other fundamental news which short sellers may trade upon
but which is not firm-specific, such as macroeconomic or industry news; this type of news
is, however, not captured in our analysis.
Altogether, evidence on the trading behavior of secretive short sellers around firm-
specific news releases suggests that their concealment of positions is associated with the
possession of superior information about firm fundamentals.
D. Why do secretive investors leave money on the table?
Our evidence so far shows that some investors consistently stay below the disclosure
threshold, which is in line with those investors following a general policy to never disclose
short positions. However, at the same time, the average risk-adjusted returns of those
investors is substantial in percentage terms. At first sight, this seems a puzzling combination
29
of findings given that positions just below are not significantly different from position just
above the threshold in terms of common measures of short-sale constraints. It raises the
question as to why investors maintain their opaqueness and leave money on the table by
not increasing what is typically a profitable short position above the publication threshold.
What are the downsides of disclosing a short position? In this section we discuss potential
explanations and how they relate to our empirical findings.
Although investors with credible information may reduce noise trader risk by disclosing
their short position (Kovbasyuk and Pagano, 2015; Ljungqvist and Qian, 2016), outside
investors can free-ride on this information without incurring the costs of research. A short
position higher above the threshold would yield on average higher profits in dollar terms,
but at the same time it would narrow the return differential to competing funds. For
mutual funds, Frank et al. (2004) have found that copy-cat funds earn after-fee returns
that are statistically indistinguishable from, and possibly even higher than, the actively
managed funds which they are copying. In the case of the EU rule on short position
disclosure, the threat of mimicking strategies is likely to be even more severe as the
disclosure lag is much shorter, being only one day rather than the 60 days required for
the mutual fund holdings studied in Frank et al. (2004). With regard to the new EU
disclosure rules, hedge fund associations have indeed expressed concern that “firms [adopt]
‘copy-cat’ trading strategies - where trading decisions are based solely upon shadowing
other participants’ actions - which may suggest that some market participants are using
the published information to replicate their competitors’ strategies” (AIMA/MFA, 2013).
In addition to the threat of diminishing competitive advantage, a public short position
might also hamper the investors’ ability to produce fundamental research. Namely, some
short sellers reported that they have been excluded from corporate events because of
their public short position (AIMA/MFA, 2013). The burgeoning literature on private
access to management provide evidence that such corporate meetings continue to be
important, despite regulation on fair disclosure (e.g., Green, Jame, Markov and Subasi,
30
2014; Bushee, Jung and Miller, 2011, 2017). In this context, selective disclosure of non-
material information during investor conferences or small-group breakout sessions is legal
and can give an advantage to individual investors by providing them with information
that complements or updates their private information.18 For investors whose superior
performance is based mainly on trading on fundamental news, having only limited access
to corporate management presents a severe impediment for their information production.
Together, these two pieces of empirical evidence - namely that secretive investors are
informed and that their superior performance comes largely from trading on fundamental
news - are consistent with both the notion that investors are maintaining their competitive
advantage and that they are protecting their sources of information. A lack of available
data prevent us from disentangling these non-mutually exclusive explanations which would
require detailed data on how investors generate their proprietary information, which they
are keen to keep private. Indeed, our evidence shows that the investors whom we are
interested in are trying to reveal as little as possible information in official filings.
V. Implications for stock prices
In this section we analyze the consequences of investors’ avoidant behavior for price
efficiency. So far, we have observed that the publication requirement for short positions
represents an impediment for some investors in terms of increasing their position, and the
evidence suggests that these secretive investors are informed. According to the literature
on limits to arbitrage, when informed investors are confronted with impediments, this
becomes evident in prices adjusting more slowly in response to private information and
in return predictability due to temporary deviation of prices from fundamental values
(Shleifer and Vishny, 1997; Gromb and Vayanos, 2010).19 In particular, in the rational
18For more details, see the comprehensive survey on this topic by Koch, Lefanowicz and Robinson(2013).
19In general, short-selling risks, such as restrictions or the costs associated with establishing a shortposition, represent one prominent set of factors that works against the elimination of mispricing. Theoretical
31
expectations model developed by Diamond and Verrecchia (1987), in which investors are
defined as either informed or uninformed traders, and as traders with prohibited short
selling and unconstrained traders, prices converge more slowly and the speed of adjustment
to private information decreases as the proportion of prohibited short sellers increases.20
Also, Jones et al. (2016) argue, using the framework of DeMarzo et al. (1998), that when
investors are presented with an impediment to trading - in this instance a disclosure
threshold - this reduces price informativeness.21
If the publication threshold represents a restriction on short selling for informed
investors, we argue that the price discovery process is slower when these secretive investors
are just below the threshold. As a consequence, we expect that these stocks will be
temporarily mis-priced and that return predictability will be particularly pronounced when
investors face this friction.
We rely on the following empirical approach to identify periods in which the disclosure
threshold may prevent secretive investors from further increasing their positions:
Secretive positionjust below threshold i,j,t
=
1 if (Position maximumi,j = 0.4) ∩ (bini,j,t = 0.4)
0 otherwise.
(5)
First, as in the previous section, we use positions that reached the bin just below the
threshold, but never crossed it during our sample period, as a proxy for secretive positions.
Second, the friction is only binding when the position is in the maximum bin of 0.4, just
studies in this area include, for example, Miller (1977), Harrison and Kreps (1978), Diamond and Verrecchia(1987), Duffie, Garleanu and Pedersen (2002), Hong and Stein (2003), and Hong, Scheinkman and Xiong(2006). However, the vast majority of empirical studies use market-wide short-selling restrictions, such asshort-selling bans, or stock-specific characteristics to test for a slower process of price discovery and forreturn predictability.
20Also, Diamond and Verrecchia (1987) argue that in a more general model, in which there are noliquidity short sales or in which there are somewhat informed as well as informed and uninformed traders,the effect of short-selling costs on price efficiency is similar to the effect of short-sale prohibitions.
21DeMarzo et al. (1998) focus on the regulators’ optimal policy in insider trading investigations. Theyargue that the investors’ welfare-maximizing outcome entails investigations of large insider trading volumes,whereas small trades remain unpunished. The result from such a policy is to discourage insiders fromtrading above a certain threshold. Although the authors focus on corporate insiders as their agents ofmain interest, their arguments and findings apply to any group of informed investors.
32
below the threshold. In other words, we flag the investor-stock pair if it has a maximum
reporting bin of 0.4 and this maximum has been reached. Note that the measure is diluted
by noise, because Equation (5) defines only a necessary, but not a sufficient condition
for the friction to be binding. During these position days we expect price adjustments
to negative information to be slower resulting in temporary overpricing and predictable
negative abnormal returns.
A. Publication threshold as a friction
To test the hypothesis of return predictability, we employ a calendar-time portfolio
approach as outlined in Section IV.B. Our first step is to form, on a daily basis, a portfolio
of stocks for which we observe a secretive position that was just below the publication
threshold on the previous day.22 We then measure the average risk-adjusted return of the
portfolio by running time-series regressions of the returns using the risk factor models
introduced in Section IV.B. If the publication threshold represents a friction to secretive
investors, we expect that those particular stocks will have future negative abnormal returns
as a result of slower price adjustment to information.
Column (3) of Panel A, Table 8, shows the average risk-adjusted returns of the portfolio
across the three factor models. Stocks for which investors have secretive positions just
below the publication threshold exhibit predictability of strong and statistically significant,
negative returns. The return predictability is robust across the CAPM, the three-factor
model and the four-factor model, with daily alpha values ranging between -5.46 and -6.13
bps; this translates into a return of around -1% per month.23
22For the majority of these stocks (82%) there is only one secretive investor. For 14% of these stocksthere are two secretive investors, while for the remaining 4%, there are between three and six secretiveinvestors. The portfolio contains on average 32 distinct stocks, with a median of 31, a minimum of 21,and a maximum of 49 stocks.
23In our data, the median time period for which a position is restricted is 16 days and the mean durationis 41 days. To reflect this and to make our results comparable with other studies, we report monthlyabnormal returns.
33
Economically, the predictability effect created by the publication threshold is substantial,
especially considering that these short-position notifications are issued almost always for
highly liquid, large-cap stocks (see Table 1 and Section II). It is important to recall that
the measure for identifying restricted investors is diluted considerably by noise. Thus, the
estimated return effect of around 5 bps per day likely represents a lower bound.
B. Placebo tests
Our finding of abnormal negative returns when short sellers hold positions just below
the disclosure threshold is consistent with the notion of slower price discovery and return
predictability originating from the publication avoidance of investors due to short-sale
impediments. To show that the return predictability effect is unique to positions just
below the disclosure threshold, we employ several placebo tests.
First, we conduct an analysis similar to the one reported in Section V.A and form
calendar-time portfolios, but now we choose hypothetical publication thresholds below and
above the true threshold:
PlaceboAi,j,t =
1 if (Position maximumi,j = p) ∩ (bini,j,t = p)
0 otherwise.
(6)
where p = 0.2, 0.3, 0.5, 0.6. Specifically, we look at positions when they reach their maxi-
mum, but when that maximum is other than 0.4. As before, we include the corresponding
stock into the placebo portfolio whenever a position fulfills the condition in Equation (6).
With this exercise, we want to rule out the possibility that the negative return reported in
the previous section is a finding that applies when short positions generally reach their
maximum. If the publication threshold truly constitutes a trading friction, we should find
the strongest return predictability for stocks with investors present in their 0.4 maximum
interval, but not for the other intervals.
34
Columns (1), (2), (4) and (5) in Panel A of Table 8 report the risk-adjusted average
returns for the other maximum reporting bins, and the same factor models are used
as in the previous analysis. Consistent with our hypothesis that binding short-sale
impediments are imposed by the publication threshold, we find that stocks with investors
in a maximum reporting bin other than 0.4 are not associated with significant negative
future returns. Even the average return for the closest non-public maximum reporting
bin, 0.3, is not significantly negative, and its economic magnitude is only one-fourth of the
return associated with the 0.4 bin. These results suggest that the return effect is unique
to stocks with investors who hold a position just below the disclosure threshold and that
it cannot be generalized to other positions.
Another potential concern is that the result in Column (3) of Panel A may be driven
entirely by informed trading in secretive positions rather than by informed investors being
constrained by the threshold. To determine the constraining effect of the threshold, we
compare the returns of the same investor-stock pair when the disclosure threshold is
binding and non-binding. That is, we look at positions with a maximum bin of 0.4 but
during periods below the maximum:
PlaceboBi,j,t =
1 if (Position maximumi,j = 0.4) ∩ (bini,j,t < 0.4)
0 otherwise.
(7)
If the return just below the disclosure threshold is driven solely by informed trading, we
would expect a return of similar magnitude both inside and outside the maximum bin
phase. As before, we include the stock in the placebo portfolio whenever there is a short
position that fulfills the condition in Equation (7).
The average risk-adjusted return of this portfolio is reported in the shaded column in
Panel B of Table 8. Outside the maximum bin phase the average alpha is negative but not
statistically different from zero. This result suggests that the stock return predictability
that arises when a secretive investor is just below the publication threshold cannot be
35
entirely explained by informed trading in this particular stock. The findings hold across
all factor models and are in line with the hypothesis that there is a slower process of price
discovery because investors are avoiding crossing the publication threshold.
In this context we can also use the other reporting bins as a comparison. Is the
predictability generally stronger inside the maximum bin phase than in the phase outside
the maximum? Interestingly, none of the other maximum reporting bins show a pattern
similar to the interval just below the disclosure threshold. Namely, only for positions with
a maximum of 0.4 can we observe that the return in the maximum bin is negative and
lower than the return of the same positions below the maximum bin. For all other bins,
the average return below the maximum bin is even more negative than the return in the
maximum bin.
Overall, we find negative risk-adjusted returns for secretive positions when the investor
is just below the disclosure threshold. There is no negative return predictability when we
choose various hypothetical publication thresholds below and above the true one. Moreover,
within investor-stock pairs, subsequent negative returns occur mainly when the secretive
position is just below the publication threshold and the constraint originating from it is
binding. Altogether, these results suggest that the evasive behavior of short sellers in
response to the disclosure rule leads to slower price discovery.
VI. Robustness Tests
In this section we briefly summarize the various robustness checks which we carried
out on our analyses. For a detailed description of all additional tests and their results, see
the Internet Appendix.
First, we provide a more detailed description of the distribution pattern of reported short
positions when we include the second digit after the decimal point. We offer an intuition as
to why the frequency of short positions spikes around the reporting thresholds, resulting in
36
a u-shaped pattern within each reporting interval. As previously discussed in Footnote 8,
these mechanically generated spikes posit a challenge to conventional discontinuity tests
as used by Degeorge et al. (1999). Although such a test is not well-suited to our setting, it
still provides evidence that supports the notion of disclosure avoidance.
Second, we conduct different sensitivity analyses to whether investors are actually
avoiding crossing the disclosure threshold. In Section III we document how avoidance is
particularly pronounced when investors approach the publication threshold from below
and for the first time. Specifically, we use a sample split into positions that are at their
record high and positions that are below their record high. As shown in the Internet
Appendix, splitting the sample another way by dividing it simply into previously increased
and previously decreased positions, gives comparable results. This measure is simpler but
noisier given that it is based solely on the investor’s last change of the position instead of
the entire position history.
Additionally, we explore two alternative hypotheses related to the avoidance effect.
First, we investigate whether the effect we found comes from investors being hesitant to
cross the threshold but eventually crossing it after a delay, rather than avoiding doing
so altogether. Second, we examine whether investors jump immediately to abnormally
high positions after crossing the disclosure threshold. The results of our analyses support
neither the hesitance hypothesis nor the large jump hypothesis.
We also conduct a series of sensitivity analyses related to the return predictability
discussed in Section V. We apply several modifications to our calendar-time portfolio
approach. The effect is robust to excluding penny stocks or requiring the duration to be at
least five trading days in the 0.4 bin for the publicaton threshold to be defined as binding.
It is similarly robust when we weight the stocks equally rather than by the number of
positions, and when we use European, rather than German, factor portfolios.
37
VII. Conclusion
Using both public and confidential short-sale notifications, we study how a disclosure
threshold for short positions affects investors’ behavior and security prices. We show
that a considerable share of short sellers seek to avoid crossing the disclosure threshold,
which effectively represents a short-sale impediment for these investors. This publication
avoidance is related to the investors’ general strategy of opaqueness and informed trading
on stock-specific fundamental news. When the short-sale constraint imposed by the
disclosure threshold is potentially binding, stocks subsequently exhibit a negative abnormal
return, consistent with prices adjusting more slowly to private information. These findings
suggest that short sellers’ evasive behavior in response to transparency regulation imposes
a negative externality on the informational efficiency of stock prices.
It is worth noting that the disclosure rule may also have other implications beyond the
effects documented in this paper. Our analysis shows that position disclosure discourages
informed market participants from fully exploiting their information, and that it thereby
reduces price discovery in the market. In the spirit of Grossman and Stiglitz (1980),
the short-sale disclosure rule may also diminish the investors’ incentive to collect and
process information in the first place, resulting in an even greater reduction in stock price
efficiency. Conversely, improvements in liquidity and financial stability as a result of
greater transparency could be argued to have positive effects on the functioning of stock
markets (DeMarzo et al., 1998; Jones et al., 2016). Overall, our insights contribute to
the ongoing debate on the design of position disclosure rules and suggest that greater
transparency should be used with caution.
38
Appendix A. Sample construction
We obtain public and confidential short-sale notifications from the German Federal
Financial Supervisory Authority (BaFin). BaFin’s notification data include the position
holder’s name, address and country, the name and ISIN code of the stock shorted, the net
short position in number of equivalent shares and as a percentage of share issued capital,
and the position and reporting date. To construct a panel of investors’ short positions in
different stocks we first account for the ISIN changes of the stocks. Next, we convert the
original short-sale notifications into reporting intervals of 10 basis points by rounding down
to the first decimal place, as we described in Section I.A. We delete duplicate notifications
(i.e., with the same information in all variables). For a few days, we find multiple short-sale
positions for the same investor-stock pair. For these days, we keep the most recent position
(as identified by the reporting date), which is likely to represent corrections of the previous
values. We omit some stale positions, which occur disproportionately in the first days
after the regulation was put in place. We define a stale position as a position which has
been reported only once, has never changed, and is still open after 600 days.24 From the
notifications, we construct a large daily panel of investors’ short positions in different
stocks. Finally, we identify trading days from the official trading calendar of Frankfurt
Stock Exchange.
We merge the BaFin panel of short positions with stock-level data (static characteristics
and time-series data, such as price, return, and market value data) from Thomson Reuters
Datastream using current ISIN codes. We only consider domestic common equity in
the regulated market. Thus, we keep stocks categorized by Datastream as domiciled in
Germany (variable GEOG = 30), equity (variable TYPE = EQ) and major issuance
(variable MAJOR = Y). We exclude preferred stocks, depositary receipts, real estate
investment trusts, and stocks with other special features by screening the names of the
stocks. We filter out all stock-day observations when a stock is no longer trading (variable
24Other cut-off points, such as 500 or 250 days, lead to similar results.
39
P#T = missing value). We only keep shares admitted to trading on the German regulated
market, using the information from the MiFID database of ESMA.
Noting Ince and Porter’s (2006) concerns about return data from Datastream, we apply
the following filters for daily return data, as proposed by Karolyi et al. (2012) and Griffin
et al. (2010): The return (rt = (RIt/RIt−1) − 1, where RI is the dollar return index)
is set to missing if the current or lagged total return index (RI) is below 0.01. If rt or
rt−1 > 100% and (1 + rt−1)(1 + rt)− 1 < 20%, then both rt and rt−1 are set to missing.
Any return greater than 200% is also set to missing.
To gather additional information on the position holders, we manually research the
corresponding unique investor identification number from FactSet, using the position
holder’s name, address, and country information from the BaFin notification data. For
the identified investors, we then obtain investor characteristics such as investor type.
Securities lending data are obtained from Markit Securities Finance Buyside Analytics
Data Feed (market area equals DE Equity) and merged by using the ISIN.
We obtain the risk-free rate (RF), the market excess return (MKTRF), and the returns
on the factor portfolios small-minus-big (SMB), high-minus-low (HML), and winner-minus-
loser (WML) for the German stock market from Andrea Frazzini’s data library, provided
through AQR’s website.25
The detailed computation and data sources of all variables used in the analysis are
provided in Table A.1.
25See: www.aqr.com/library/data-sets/quality-minus-junk-factors-daily/data
40
Table A.1Definitions of variables
Variable: Description: Source:
Amihud illiquidity |r|/(V C × 1000× P )× 106, where r is the USD return,and V C is the consolidated number of shares tradedacross all German exchanges (in thousands), and P isthe price in USD (Amihud, 2002). We winsorize theAmihud illiquidity ratio at 1% at the upper tail and thenaverage it over the last 250 trading days, requiring atleast 100 valid observations.
Datastream
Bid-ask spread (PA− PB)/P , expressed in percentage terms, where Pis the stock’s price, PA is the ask price, and PB is thebid price, all provided by Datastream. We winsorize thebid-ask spread at 1% at the upper tail and then averageit over the last 250 trading days, requiring at least 100valid observations.
Datastream
Book-to-market ratio Calculated as PTBV −1, where PTBV is the price-to-bookvalue provided by Datastream. The book-to-market ratiois set to missing if it is below 0.
Datastream
Daily turnover V C/NOSH, expressed in percentage terms, where V Cis the consolidated number of shares traded across allGerman exchanges (in thousands), and NOSH are thecompanies’ shares outstanding (in thousands).
Datastream
Dummy: “on special” Dummy variable that equals 1 if the indicative fee (p.a.)to borrow a stock is above 100 bps (Prado et al., 2016).
Markit
Dummy: Bank Dummy variable that equals 1 if the investor is a bank. BaFin, own re-search
Dummy: European holder Dummy variable that equals 1 if the investor is domiciledin a European country.
BaFin
Dummy: Earnings an-nouncement
Event dummy variable that equals 1 if there is an earningsannouncement for that specific stock on that day or theprevious day.
Datastream
Dummy: Ad-hoc news Event dummy variable that equals 1 if there is ad-hocnews for that specific stock on that day or the previousday.
DGAP (part ofthe EQS Group)
Dummy: Futures or listedoptions
Dummy variable that equals 1 if futures or listed optionsexist for the underlying stock.
Datastream
Continued on next page
41
Table A.1 – Continued from previous page
Variable: Description: Source:
Dummy: Hedge fund Dummy variable that equals 1 if the investor is definedby FactSet as a hedge fund or has no entry in Factset(similar to the selection criterion of Brunnermeier andNagel (2004)).
BaFin, FactSet
Dummy: No FactSet entry Dummy variable that equals 1 if the investor is notincluded in the FactSet Ownership database. The owner-ship information in FactSet originates from public filings,so a non-appearance is indicative of no public record.
FactSet
Dummy: Private fund Dummy variable that equals 1 if the investor files FormD with the SEC and seeks private placements underRegulation D. Private placements are not subject to someof the laws and regulations designed to protect investors,such as the comprehensive disclosure requirements thatapply to registered offerings.
EDGAR by SEC
Dummy: Tax haven Dummy variable that equals 1 if the investor is domiciledin a tax haven as defined by OECD (2000)
BaFin, OECD
Dummy: US holder Dummy variable that equals 1 if the investor is domiciledin the United States.
BaFin
Fee tail risk The 99th percentile of a normal distribution using theestimated mean and standard deviation of the indicativefee (see below) over the last 250 trading days (Engelberg,Reed and Ringgenberg, 2017). We requiring at least 100valid observations for mean and standard deviation com-putation: mean of fee + 2.33× standard deviation of fee
Markit
Share of lendable stocks Ratio, expressed in percentage terms, between lendablevalue and the market capitalization of the stock. We useMarkit’s lendable value adjusted for inventory which isnot being actively made available for lending.
Markit
Indicative fee The expected securities borrowing cost, in percentageper year terms, for a hedge fund on a given day, whichis derived by Markit.
Markit
Institutional ownership Percentage of shares held by institutional investors inthe previous quarter.
FactSet
Inventory concentration Herfindahl–Hirschman Index (HHI) scaled between 0 and100 to measure inventory concentration.
Markit
Lender concentration Herfindahl–Hirschman Index (HHI) scaled between 0 and100 to measure lender concentration.
Markit
Market value Market capitalization (in USD million) provided by theDatastream variable MV .
Datastream
Continued on next page
42
Table A.1 – Continued from previous page
Variable: Description: Source:
Return volatility Standard deviation of return computed over the last 250trading days, requiring at least 100 valid observations.Expressed in percentage terms.
Datastream
Short interest Ratio, expressed in percentage terms, between the valueof securities on loan (with dividend trading and financingtrades removed) and the market capitalization of thestock.
Markit
Standard deviation of fee Standard deviation of the indicative fee (see above) overthe last 250 trading days, requiring at least 100 validobservations.
Markit
Standard deviation of uti-lization
Standard deviation of utilization (see below) over thelast 250 trading days, requiring at least 100 valid obser-vations.
Markit
Total assets Market capitalization (in USD million) of the investor’slong holdings as provided by FactSet.
FactSet
Utilization Percentage of actively lendable securities in lending pro-grams which are currently out on loan, calculated as thevalue of assets on loan from lenders divided by the activelendable value
Markit
Utilization tail risk The 99th percentile of a normal distribution usingthe estimated mean and standard deviation of utiliza-tion (see above) over the last 250 trading days (En-gelberg, Reed and Ringgenberg, 2017). We requireat least 100 valid observations for mean and stan-dard deviation computation and we limit utilizationtail risk to 100%: min(mean of utilization + 2.33 ×standard deviation of utilization; 100%)
Markit
43
References
Agarwal, V., Jiang, W., Tang, Y. and Yang, B. (2013), ‘Uncovering hedge fund skill fromthe portfolio holdings they hide’, The Journal of Finance 68(2), 739–783.
AIMA/MFA (2013), ‘Joint response to the call for evidence by the European Securitiesand Markets Authority (ESMA) regarding the evaluation of the Regulation (EU)236/2012’, Submitted online. March 15, 2013.URL: https://www.esma.europa.eu/press-news/consultations/call-evidence-evaluation-regulation-short-selling-and-certain-aspects
Aitken, M. J., Frino, A., McCorry, M. S. and Swan, P. L. (1998), ‘Short sales are almostinstantaneously bad news: Evidence from the Australian Stock Exchange’, The Journalof Finance 53(6), 2205–2223.
Amihud, Y. (2002), ‘Illiquidity and stock returns: Cross-section and time-series effects’,Journal of Financial Markets 5(1), 31–56.
Angel, J. J., Christophe, S. E. and Ferri, M. G. (2003), ‘A Close Look at Short Selling onNasdaq’, Financial Analysts Journal 59(6), 66–74.
Asquith, P., Pathak, P. A. and Ritter, J. R. (2005), ‘Short interest, institutionalownership, and stock returns’, Journal of Financial Economics 78(2), 243–276.
Battalio, R. and Schultz, P. (2011), ‘Regulatory uncertainty and market liquidity: The2008 short sale ban’s impact on equity option markets’, The Journal of Finance66(6), 2013–2053.
Beber, A. and Pagano, M. (2013), ‘Short-selling bans around the world: Evidence fromthe 2007–09 crisis’, The Journal of Finance 68(1), 343–381.
Boehmer, E., Jones, C. M. and Zhang, X. (2008), ‘Which shorts are informed?’, TheJournal of Finance 63(2), 491–527.
Boehmer, E., Jones, C. M. and Zhang, X. (2013), ‘Shackling short sellers: The 2008shorting ban’, Review of Financial Studies 26(6), 1363–1400.
Brunnermeier, M. K. and Nagel, S. (2004), ‘Hedge Funds and the Technology Bubble’,The Journal of Finance 59(5), 2013–2040.
Brunnermeier, M. K. and Oehmke, M. (2014), ‘Predatory short selling’, Review ofFinance 18(6), 2153–2195.
Bushee, B. J., Jung, M. J. and Miller, G. S. (2011), ‘Conference Presentations and theDisclosure Milieu’, Journal of Accounting Research 49(5), 1163–1192.
Bushee, B. J., Jung, M. J. and Miller, G. S. (2017), Do Investors Benefit from SelectiveAccess to Management? Journal of Financial Reporting, In-Press.
44
Carhart, M. M. (1997), ‘On persistence in mutual fund performance’, The Journal ofFinance 52(1), 57–82.
Cohen, L., Diether, K. B. and Malloy, C. J. (2007), ‘Supply and demand shifts in theshorting market’, The Journal of Finance 62(5), 2061–2096.
D’Avolio, G. (2002), ‘The market for borrowing stock’, Journal of Financial Economics66(2), 271–306.
De Long, J. B., Shleifer, A., Summers, L. H. and Waldmann, R. J. (1990), ‘Noise traderrisk in financial markets’, Journal of Political Economy 98(4), 703–738.
Degeorge, F., Patel, J. and Zeckhauser, R. (1999), ‘Earnings management to exceedthresholds’, The Journal of Business 72(1), 1–33.
DeMarzo, P. M., Fishman, M. J. and Hagerty, K. M. (1998), ‘The optimal enforcement ofinsider trading regulations’, Journal of Political Economy 106(3), 602–632.
Diamond, D. W. (1985), ‘Optimal release of information by firms’, The Journal ofFinance 40(4), 1071–1094.
Diamond, D. W. and Verrecchia, R. E. (1987), ‘Constraints on short-selling and asset priceadjustment to private information’, Journal of Financial Economics 18(2), 277–311.
Diether, K. B., Lee, K.-H. and Werner, I. M. (2009), ‘Short-sale strategies and returnpredictability’, Review of Financial Studies 22(2), 575–607.
Driscoll, J. C. and Kraay, A. C. (1998), ‘Consistent Covariance Matrix Estimation withSpatially Dependent Panel Data’, Review of Economics and Statistics 80(4), 549–560.
Duffie, D., Garleanu, N. and Pedersen, L. H. (2002), ‘Securities lending, shorting, andpricing’, Journal of Financial Economics 66(2), 307–339.
Duong, T. X., Huszar, Z. R. and Yamada, T. (2015), ‘The costs and benefits of short saledisclosure’, Journal of Banking & Finance 53, 124–139.
Engelberg, J. E., McLean, R. D. and Pontiff, J. (2017), Anomalies and News. The Journalof Finance, forthcoming.
Engelberg, J. E., Reed, A. V. and Ringgenberg, M. C. (2012), ‘How are shorts informed?:Short sellers, news, and information processing’, Journal of Financial Economics105(2), 260–278.
Engelberg, J. E., Reed, A. V. and Ringgenberg, M. C. (2017), Short selling risk. TheJournal of Finance, forthcoming.
ESMA (2013), Guidelines: Exemption for market making activities and primary marketoperations under Regulation (EU) 236/2012 of the European Parliament and theCouncil on short selling and certain aspects of Credit Default Swaps, Technical report.
45
ESMA (2016), ‘List of market makers and authorised primary dealers who are using theexemption under the Regulation on short selling and credit default swaps’. 25 January2016.
Fama, E. F. and French, K. R. (1993), ‘Common risk factors in the returns on stocks andbonds’, Journal of Financial Economics 33(1), 3–56.
Fang, V. W., Huang, A. H. and Karpoff, J. M. (2016), ‘Short selling and earningsmanagement: A controlled experiment’, The Journal of Finance 71(3), 1251–1294.
Frank, M. M., Poterba, J. M., Shackelford, D. A. and Shoven, J. B. (2004), ‘Copycatfunds: Information disclosure regulation and the returns to active management in themutual fund industry’, The Journal of Law and Economics 47(2), 515–541.
Goldstein, I. and Guembel, A. (2008), ‘Manipulation and the Allocational Role of Prices’,The Review of Economic Studies 75(1), 133–164.
Green, T. C., Jame, R., Markov, S. and Subasi, M. (2014), ‘Access to management and theinformativeness of analyst research’, Journal of Financial Economics 114(2), 239–255.
Griffin, J. M., Kelly, P. J. and Nardari, F. (2010), ‘Do market efficiency measures yieldcorrect inferences? A comparison of developed and emerging markets’, Review ofFinancial Studies 23(8), 3225–3277.
Gromb, D. and Vayanos, D. (2002), ‘Equilibrium and welfare in markets with financiallyconstrained arbitrageurs’, Journal of Financial Economics 66(2), 361–407.
Gromb, D. and Vayanos, D. (2010), ‘Limits of Arbitrage’, Annual Review of FinancialEconomics 2(1), 251–275.
Grossman, S. J. and Stiglitz, J. E. (1980), ‘On the impossibility of informationally efficientmarkets’, American Economic Review 70(3), 393–408.
Grundy, B. D., Lim, B. and Verwijmeren, P. (2012), ‘Do option markets undo restrictionson short sales? evidence from the 2008 short-sale ban’, Journal of Financial Economics106(2), 331 – 348.
Harrison, J. M. and Kreps, D. M. (1978), ‘Speculative investor behavior in a stock marketwith heterogeneous expectations’, Quarterly Journal of Economics 92(2), 323–336.
Hirshleifer, J. (1971), ‘The private and social value of information and the reward toinventive activity’, The American Economic Review 61(4), 561–574.
Hoechle, D., Schmid, M. M. and Zimmermann, H. (2016), Decomposing performance.SSRN Working Paper.
Hong, H., Scheinkman, J. and Xiong, W. (2006), ‘Asset float and speculative bubbles’,The Journal of Finance 61(3), 1073–1117.
46
Hong, H. and Stein, J. C. (2003), ‘Differences of opinion, short-sales constraints, andmarket crashes’, Review of Financial Studies 16(2), 487–525.
Huddart, S., Hughes, J. S. and Levine, C. B. (2001), ‘Public disclosure and dissimulationof insider trades’, Econometrica 69(3), 665–681.
Ince, O. S. and Porter, R. B. (2006), ‘Individual equity return data from ThomsonDatastream: Handle with care!’, Journal of Financial Research 29(4), 463–479.
Jimenez, G., Ongena, S., Peydro, J.-L. and Saurina, J. (2014), ‘Hazardous times formonetary policy: What do twenty-three million bank loans say about the effects ofmonetary policy on credit risk-taking?’, Econometrica 82(2), 463–505.
Jones, C. M. and Lamont, O. A. (2002), ‘Short-sale constraints and stock returns’,Journal of Financial Economics 66(2), 207–239.
Jones, C. M., Reed, A. V. and Waller, W. (2016), ‘Revealing shorts an examination oflarge short position disclosures’, The Review of Financial Studies 29(12), 3278–3320.
Kaplan, S. N., Moskowitz, T. J. and Sensoy, B. A. (2013), ‘The Effects of Stock Lendingon Security Prices: An Experiment’, The Journal of Finance 68(5), 1891–1936.
Karolyi, G. A., Lee, K.-H. and Van Dijk, M. A. (2012), ‘Understanding commonality inliquidity around the world’, Journal of Financial Economics 105(1), 82–112.
Koch, A. S., Lefanowicz, C. E. and Robinson, J. R. (2013), ‘Regulation FD: A Reviewand Synthesis of the Academic Literature’, Accounting Horizons 27(3), 619–646.
Kovbasyuk, S. and Pagano, M. (2015), Advertising arbitrage. Working Paper.
Lamont, O. A. (2012), ‘Go down fighting: Short sellers vs. firms’, Review of Asset PricingStudies 2(1), 1–30.
Leland, H. E. (1992), ‘Insider trading: Should it be prohibited?’, Journal of PoliticalEconomy 100(4), 859–887.
Lintner, J. (1965), ‘The valuation of risk assets and the selection of risky investments instock portfolios and capital budgets’, Review of Economics and Statistics 47(1), 13–37.
Ljungqvist, A. and Qian, W. (2016), ‘How constraining are limits to arbitrage?’, Reviewof Financial Studies 29(8), 1975–2028.
Miller, E. M. (1977), ‘Risk, uncertainty, and divergence of opinion’, The Journal ofFinance 32(4), 1151–1168.
Morris, S. and Shin, H. S. (2002), ‘Social value of public information’, The AmericanEconomic Review 92(5), 1521–1534.
Nagel, S. (2005), ‘Short sales, institutional investors and the cross-section of stockreturns’, Journal of Financial Economics 78(2), 277–309.
47
NASDAQ (2015), ‘Request to require disclosure of short positions in parity with requireddisclosure of long positions’, Rulemaking Petition 4-691 submitted to the SEC.December 7, 2015.URL: http://www.sec.gov/rules/petitions/2015/petn4-691.pdf
Newey, W. K. and West, K. D. (1987), ‘A simple, positive semi-definite, heteroskedasticityand autocorrelation consistent covariance matrix’, Econometrica 55(3), 703–708.
Newey, W. K. and West, K. D. (1994), ‘Automatic lag selection in covariance matrixestimation’, Review of Economic Studies 61(4), 631–653.
NYSE (2015), ‘Request to require the periodic public disclosure of short-sale activities byinstitutional investment managers’, Rulemaking Petition 4-689 submitted to the SEC.October 7, 2015.URL: http://www.sec.gov/rules/petitions/2015/petn4-689.pdf
OECD (2000), Towards global tax cooperation: Progress in identifying and eliminatingharmful tax practices, Technical report, OECD.
Prado, M. P., Saffi, P. A. C. and Sturgess, J. (2016), ‘Ownership structure, limits toarbitrage, and stock returns: Evidence from equity lending markets’, The Review ofFinancial Studies 29(12), 3211–3244.
Reed, A. V. (2013), ‘Short selling’, Annual Review of Financial Economics 5(1), 245–258.
Seneca, J. J. (1967), ‘Short interest: bearish or bullish?’, The Journal of Finance22(1), 67–70.
Sharpe, W. F. (1964), ‘Capital asset prices: A theory of market equilibrium underconditions of risk’, The Journal of Finance 19(3), 425–442.
Shleifer, A. and Vishny, R. W. (1997), ‘The limits of arbitrage’, The Journal of Finance52(1), 35–55.
U.S. SEC (2014), Short sale position and transaction reporting: As required by Section417 of the Dodd-Frank Wall Street Reform and Consumer Protection Act, Technicalreport, U.S. Securities and Exchange Commission.
Verrecchia, R. E. (1982), ‘The use of mathematical models in financial accounting’,Journal of Accounting Research 20, 1–42.
48
(a) Overall sample
Public disclosure threshold
010
2030
4050
Freq
uenc
y (in
%)
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
Reporting bins
(b) Sample split
Public disclosure threshold
010
2030
4050
Freq
uenc
y (in
%)
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
Reporting bins
Positions below their record highPositions at their record high
Figure 1.Distribution of open short positionsThis figure shows the distribution of days with open short positions (investor-stock-day observations)across reporting intervals. Reporting intervals are in 10 bps steps, starting from 0.2%. Positions above0.2% but below 0.5% are reported to the regulator but not disclosed to the public; positions of 0.5% andhigher are disclosed to the public. Figure 1(a) shows the relative frequency of days with an open positionfor each interval for the overall sample. Figure 1(b) reports the relative frequency separately for shortpositions at their record high and for positions below their record high. Reporting intervals greater than1.5% are truncated for readability. The sample contains all German domestic equity in the regulatedmarket from November 5, 2012, to March 31, 2015.
49
Public disclosure threshold
05
1015
2025
Freq
uenc
y (in
%)
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
Position maximum
Position maximum ≠ investor's maximumPosition maximum = investor's maximum
Figure 2.Distribution of open short positions according to the maximum bin reachedThis figure shows the distribution of short position days (investor-stock-day observations) according tothe maximum bin reached by the position. Positions that never cross the disclosure threshold (positionswith a maximum of 0.4) account for the largest group (22.8%) of short position days. The red coloringindicates the share of positions where the position maximum is also the investor’s maximum. Reportingintervals greater than 1.5% are truncated for readability. The sample contains all investors with shortposition notifications in German domestic equity in the regulated market from November 5, 2012, toMarch 31, 2015.
50
Table 1Summary statistics:Stocks with short position notifications/disclosures vs. all stocksThis table shows summary statistics for the various stock characteristics of stocks with at least one shortposition notification and for all stocks in the regulated market. Short positions must be reported to theregulator if the position is greater than or equal to 0.2% of the issued share capital of the company shortedand must be publicly disclosed if the position is greater than or equal to 0.5%. Column (1) contains thesummary statistics for the entire population of stocks in the German regulated market, Column (2) reportsstocks that have at least one public or confidential short position notification, Column (3) includes stockswith at least one public short position, and Column (4) refers to stocks with at least one confidential butno public short position notification. The table reports time-series averages of cross-sectional medians.The sample consists of common equity in the German regulated stock market from November 5, 2012,until March 31, 2015. For details on calculation of the stock characteristics, see Table A.1.
(1) (2) (3) (4)
Stocks withAll stocks short position notifications/disclosures
All Public Confidential
Share of stocks 19.9% 8.3% 11.6%
Median values:
Market value (millions USD) 102.1 2,225.3 1,986.6 2,401.2Book-to-market ratio 0.65 0.53 0.55 0.52Daily turnover (%) 0.12 0.32 0.37 0.29Amihud illiquidity 0.456 0.004 0.004 0.004Bid-ask spread (%) 3.10 0.64 0.65 0.63Return volatility (%) 2.47 1.96 2.16 1.85Institutional ownership (%) 3.86 29.21 29.85 29.00Share of lendable stocks (%) 2.42 13.59 12.87 14.31Indicative fee (p.a. %) 0.66 0.53 0.56 0.51
51
Table
2Sum
mary
stati
stic
sT
his
table
conta
ins
sum
mar
yst
atis
tics
for
the
inve
stor
-sto
ckp
anel
wit
hop
ensh
ort
pos
itio
nnot
ifica
tion
sab
ove
0.2%
ofth
eis
sued
shar
eca
pit
al.
The
sum
mar
yst
atis
tics
incl
ude
the
num
ber
ofob
serv
atio
ns
(N),
mea
n,
stan
dar
ddev
iati
on(S
D),
and
the
10th
,25
th,
50th
,75
than
d90
thp
erce
nti
les.
The
sam
ple
consi
sts
ofco
mm
oneq
uit
yin
the
Ger
man
regu
late
dst
ock
mar
ket
from
Nov
emb
er5,
2012
,unti
lM
arch
31,
2015
.F
ordet
ails
onth
eca
lcula
tion
ofth
eva
riab
les,
see
Tab
leA
.1.
Per
centi
les:
Var
iab
leN
Mea
nSD
10th
25th
50th
75th
90th
Mar
ket
valu
e(m
illion
sU
SD
)278,6
49
5,4
43.3
711,8
65.4
5323.6
2845.1
02,1
13.8
15,3
42.1
310,4
67.8
5B
ook
-to-
mar
ket
rati
o276,1
47
0.8
93.2
10.2
60.3
70.5
50.8
61.3
0D
aily
turn
over
(%)
273,5
36
0.5
00.3
10.1
80.2
60.4
30.7
10.9
2A
mih
ud
illiquid
ity
273,5
36
0.0
71.0
10.0
00.0
00.0
00.0
10.0
3B
id-a
sksp
read
(%)
276,8
60
0.7
30.5
00.3
40.4
20.6
40.8
91.2
0R
eturn
vola
tility
(%)
276,8
60
2.2
50.9
21.4
21.7
12.0
82.6
13.0
6In
stit
uti
onal
own
ersh
ip(%
)278,4
75
34.3
717.5
013.0
821.9
331.5
045.8
161.2
1D
um
my:
Fu
ture
sor
list
edop
tion
s(%
)278,6
49
38.3
9Sh
ort
inte
rest
(%)
278,5
70
4.3
53.6
50.7
81.6
03.3
26.2
29.0
8U
tiliza
tion
(%)
278,6
30
33.3
029.5
83.7
39.0
022.9
650.1
884.1
6In
dic
ativ
efe
e(%
p.a
.)278,5
43
2.0
45.7
10.5
00.5
00.5
01.0
06.0
0D
um
my:
Hed
gefu
nd
(%)
278,6
49
76.2
2D
um
my:
Ban
k(%
)278,6
49
1.8
1D
um
my:
Euro
pea
nhol
der
(%)
278,6
49
47.4
1D
um
my:
US
hol
der
(%)
278,6
49
45.3
5D
um
my:
Tax
hav
en(%
)278,6
49
6.1
4T
otal
asse
ts(m
illion
sU
SD
)206,8
23
78,3
95
232,6
84
72
755
4,3
6730,4
66
134,2
92
Dum
my:
No
Fac
tSet
entr
y(%
)278,6
49
9.7
3D
um
my:
Pri
vate
fund
(%)
217,0
36
8.3
6
52
Table
3P
robabilit
yof
ash
ort
posi
tion
incr
ease
This
table
show
sth
ees
tim
ated
pro
bab
ilit
yof
incr
easi
ng
ash
ort
pos
itio
n,
condit
ional
oncu
rren
tly
hav
ing
ap
osit
ion
ina
spec
ific
rep
orti
ng
bin
.Shor
tp
osit
ions
are
rep
orte
din
bin
sof
10bp
s,st
arti
ng
from
0.2%
ofth
eis
sued
shar
eca
pit
alof
the
com
pan
ysh
orte
d.
Pos
itio
ns
abov
e0.
2%b
ut
bel
ow0.
5%are
rep
ort
edto
the
regula
tor
but
not
dis
close
dto
the
public;
posi
tions
of
0.5
%and
ab
ove
are
dis
close
dto
the
public.
The
0.2
bin
den
ote
sth
ebin
rangi
ng
from
0.20
%to
0.29
%,
the
0.3
bin
from
0.30
%to
0.39
%,
the
0.4
bin
from
0.40
%to
0.49
%,
and
sofo
rth.
Rep
orti
ng
bin
sgr
eate
rth
anor
equal
to1.
0%ar
esu
mm
ariz
edin
one
grou
p.
Inad
dit
ion
toth
epro
bab
ilit
ies
ofin
crea
sing
ash
ort
pos
itio
n,
the
table
also
dis
pla
ys
the
diff
eren
cein
pro
bab
ilit
yof
row
(3)
rela
tive
toro
w(k
),th
e0.4
rep
ort
ing
bin
just
bel
owth
ep
ub
lica
tion
thre
shold
(shad
edin
gra
y),
and
thep-v
alu
efo
rth
ediff
eren
ces
inm
ean
s.T
he
tab
ledis
pla
ys
pro
bab
ilit
ies
for
the
over
all
sam
ple
and
for
two
subsa
mple
s,w
her
ew
esp
lit
the
sam
ple
into
shor
tp
osit
ion
sat
thei
rre
cord
hig
han
dp
osit
ion
sb
elow
thei
rre
cord
hig
h.
*,**
,an
d***
ind
icate
signifi
can
ceat
the
10%
,5%
,and
1%
leve
ls,
resp
ecti
vely
.
Posi
tions
Posi
tions
Ove
rall
at
thei
rre
cord
hig
hb
elow
thei
rre
cord
hig
hP
robab
ilit
yD
iffer
ence
Pro
babilit
yD
iffer
ence
Pro
babilit
yD
iffer
ence
Bin
ofin
crea
sero
w(3
)-
(k)
p-v
alu
eof
incr
ease
row
(3)
-(k
)p-v
alu
eof
incr
ease
row
(3)
-(k
)p-v
alu
e
undis-
closed
(1)
0.2
0.36
4−
0.01
0(0
.469)
0.3
47
−0.
010
(0.5
81)
0.3
80
−0.
001
(0.9
47)
(2)
0.3
0.39
1−
0.03
7**
(0.0
14)
0.4
23
−0.0
85***
(0.0
00)
0.3
60
0.018
(0.4
16)
(3)
0.4
0.35
40.3
38
0.3
79
disclosed
(4)
0.5
0.40
8−
0.05
4***
(0.0
07)
0.5
14
−0.
176***
(0.0
00)
0.3
02
0.076***
(0.0
09)
(5)
0.6
0.47
6−
0.12
2***
(0.0
00)
0.5
62
−0.2
25***
(0.0
00)
0.3
95
−0.
017
(0.6
24)
(6)
0.7
0.44
0−
0.08
6***
(0.0
01)
0.6
04
−0.2
66***
(0.0
00)
0.3
08
0.071**
(0.0
48)
(7)
0.8
0.44
9−
0.09
5***
(0.0
01)
0.6
48
−0.3
10***
(0.0
00)
0.2
66
0.113***
(0.0
06)
(8)
0.9
0.47
7−
0.12
3***
(0.0
00)
0.7
02
−0.3
64***
(0.0
00)
0.2
79
0.100**
(0.0
32)
(9)≥
1.0
0.47
4−
0.12
0***
(0.0
00)
0.6
82
−0.3
45***
(0.0
00)
0.2
75
0.103***
(0.0
00)
53
Table
4D
ura
tion
inre
port
ing
bin
sT
his
table
show
sth
eav
erag
eti
me
spen
tin
each
rep
orti
ng
bin
.Shor
tp
osit
ions
are
rep
orte
din
bin
sof
10bps,
star
ting
from
0.2%
ofis
sued
shar
eca
pit
alof
the
com
pany
short
ed.
Posi
tions
ab
ove
0.2
%but
bel
ow0.5
%are
rep
ort
edto
the
regula
tor
but
not
dis
close
dto
the
public;
posi
tions
of
0.5
%and
hig
her
are
dis
clos
edto
the
public.
The
0.2
bin
den
otes
the
bin
rangi
ng
from
0.20
%to
0.29
%,
the
0.3
bin
from
0.30
%to
0.39
%,
the
0.4
bin
from
0.40
%to
0.49
%,
and
sofo
rth.
Rep
orti
ng
bin
sgr
eate
rth
anor
equal
to1.
0%ar
esu
mm
ariz
edin
one
grou
p.
The
table
rep
orts
the
mea
nan
dm
edia
nnum
ber
oftr
adin
gday
ssp
ent
inea
chre
por
ting
bin
(Pan
els
Aan
dB
,re
spec
tive
ly).
Inaddit
ion,
itd
isp
lays
the
diff
eren
cein
mea
n(m
edia
n)
dura
tion
of
row
(3)
rela
tive
toro
w(k
),th
e0.4
rep
ort
ing
bin
just
bel
owth
ep
ublica
tion
thre
shold
(shad
edin
gra
y),
an
dth
ep-v
alu
efo
rdiff
eren
ces
inm
eans
(med
ian
s).
Eac
hpan
eldis
pla
ys
pro
bab
ilit
ies
for
the
over
all
sam
ple
and
for
two
subsa
mple
s,w
her
ew
esp
lit
the
sam
ple
into
shor
tp
osit
ions
atth
eir
reco
rdhig
han
dp
osit
ion
sb
elow
thei
rre
cord
hig
h.
*,**
,an
d***
ind
icate
signifi
can
ceat
the
10%
,5%
,and
1%
leve
ls,
resp
ecti
vely
.
Panel
A:
Mean
dura
tion
wit
hin
rep
ort
ing
bin
intr
ad
ing
days
Posi
tions
Posi
tions
Ove
rall
at
thei
rre
cord
hig
hb
elow
thei
rre
cord
hig
hM
ean
Diff
eren
ceM
ean
Diff
eren
ceM
ean
Diff
eren
ceB
indura
tion
row
(3)
-(k
)p-v
alu
edu
rati
on
row
(3)
-(k
)p-v
alu
ed
ura
tion
row
(3)
-(k
)p-v
alu
e
undis-
closed
(1)
0.2
18.3
2.2**
(0.0
19)
22.3
3.6**
(0.0
12)
14.0
−2.4
**
(0.0
39)
(2)
0.3
17.0
3.6***
(0.0
00)
21.3
4.6***
(0.0
02)
12.6
−1.0
(0.3
76)
(3)
0.4
20.6
26.0
11.7
disclosed
(4)
0.5
14.3
6.2***
(0.0
00)
16.7
9.3***
(0.0
00)
11.8
−0.1
(0.9
19)
(5)
0.6
13.7
6.8***
(0.0
00)
17.4
8.5***
(0.0
01)
9.9
1.8
(0.2
38)
(6)
0.7
13.3
7.3***
(0.0
00)
15.8
10.
1***
(0.0
01)
11.2
0.5
(0.7
72)
(7)
0.8
13.6
7.0***
(0.0
01)
17.2
8.8***
(0.0
08)
10.0
1.6
(0.3
79)
(8)
0.9
11.9
8.7***
(0.0
00)
14.1
11.
8***
(0.0
02)
9.7
2.0
(0.3
55)
(9)≥
1.0
12.6
8.0***
(0.0
00)
16.6
9.4***
(0.0
00)
8.6
3.1***
(0.0
04)
Continued
onnextpage
54
Tab
le4
–Continued
from
previouspage
Panel
B:
Media
ndura
tion
wit
hin
rep
ort
ing
bin
intr
ad
ing
days
Posi
tions
Posi
tion
sO
vera
llat
thei
rre
cord
hig
hb
elow
thei
rre
cord
hig
hM
edia
nD
iffer
ence
Med
ian
Diff
eren
ceM
edia
nD
iffer
ence
Bin
dura
tion
row
(3)
-(k
)p-v
alu
edura
tion
row
(3)
-(k
)p-v
alu
ed
ura
tion
row
(3)
-(k
)p-v
alu
e
undis-
closed
(1)
0.2
61***
(0.0
04)
73***
(0.0
00)
50
(0.1
82)
(2)
0.3
61***
(0.0
03)
82***
(0.0
07)
50
(0.8
24)
(3)
0.4
710
5
disclosed
(4)
0.5
52***
(0.0
00)
73***
(0.0
04)
41
(0.3
64)
(5)
0.6
52***
(0.0
00)
64***
(0.0
00)
41
(0.3
31)
(6)
0.7
43***
(0.0
00)
46***
(0.0
00)
41
(0.2
78)
(7)
0.8
43***
(0.0
00)
55***
(0.0
00)
32***
(0.0
09)
(8)
0.9
43***
(0.0
00)
55***
(0.0
00)
32*
(0.0
53)
(9)≥
1.0
43***
(0.0
00)
55***
(0.0
00)
32***
(0.0
00)
55
Table 5Probability of increase: Regression approach with fixed effectsThis table shows estimates from the linear probability model
yi,j,t+1 = α0 + β1 Just below thresholdi,j,t +∑k
βk k bini,j,t + γ′ Xi,t + αt + εi,j,t+1,
yi,j,t+1 = α0 + β1 Just below thresholdi,j,t +∑k
βk k bini,j,t + αi,t + αj,t + εi,j,t+1.
with yi,j,t+1 = 1 (bini,j,t > bini,j,t−1). Just below threshold is equal to 1 if investor j has a short position instock i in the bin 0.4. k bini,j,t−1 is equal to 1 if investor j has a short position in stock i in the bin k. Theomitted benchmark bin is 0.3, i.e. k ∈ {0.2, 0.5, 0.6+}, where 0.6+ indicates position bins greater or equalto 0.6. Xi,t represents a vector of control variables related to short selling activity. αt, αi,t, and αj,t denoteday, stock-day, and investor-day fixed effects, respectively. Panel A reports the regression coefficientsPanel B shows the differences in probability of increase between the bin just below the disclosure thresholdand the neighboring bins. The unconditional probability of increasing a short position to enter the nextbin (estimated by the sample average of yi,j,t+1) is 2.0%. The estimated coefficients are scaled to reflectchanges in percentage points. Standard errors are clustered at the investor-stock and time level. Thet-statistics are given in parentheses, and *, **, and *** indicate significance at the 10%, 5%, and 1%levels, respectively. For details on the calculation of the variables, see Table A.1.
(1) (2) (3)
Panel A: Regressions with reporting bin dummies
0.2 bin −0.43∗∗∗ (−3.41) −0.42∗∗∗ (−3.26) −0.41∗∗∗ (−2.76)0.3 bin − − − − − −0.4 bin (Just below threshold) −0.54∗∗∗ (−3.67) −0.59∗∗∗ (−3.88) −1.08∗∗∗ (−5.40)0.5 bin 0.64∗∗∗ (2.74) 0.59∗∗ (2.50) −0.16 (−0.54)≥0.6 bin 1.22∗∗∗ (4.82) 1.13∗∗∗ (4.64) 1.02∗∗∗ (4.18)ln(market value) −0.16∗∗ (−2.07)ln(return volatility) 1.55∗∗∗ (5.83)ln(Amihud illiquidity) −0.37 (−1.25)ln(bid-ask spread) −0.46∗∗ (−2.04)Dummy: stock with listed futures or options 0.50∗∗∗ (3.47)Share of lendable stocks 0.04∗∗∗ (3.92)Inventory concentration −0.02∗∗ (−2.53)Indicative fee 0.01 (0.48)Standard deviation of fee −0.07∗∗∗ (−2.67)
Fixed effects:Time Yes Yes NoStock × time No No YesInvestor × time No No Yes
R2 (%) 0.60 0.78 38.08Within R2 (%) 0.20 0.37 0.14Number of observations 278,003 272,340 202,057
Panel B: Differences in the probability of increase
Just below threshold – 0.2 bin −0.11 (−0.86) −0.17 (−1.23) −0.67∗∗∗ (−3.35)Just below threshold – 0.3 bin −0.54∗∗∗ (−3.67) −0.59∗∗∗ (−3.88) −1.08∗∗∗ (−5.40)Just below threshold – 0.5 bin −1.18∗∗∗ (−5.04) −1.18∗∗∗ (−4.96) −0.92∗∗∗ (−2.89)Just below threshold – (≥ 0.6 bin) −1.76∗∗∗ (−7.31) −1.72∗∗∗ (−7.41) −2.10∗∗∗ (−8.51)
56
Table 6Comparison of positions just below with positions just above the disclosurethresholdFor each investor-stock pair we determine the maximum bin reached in the sample period. We comparepositions which never have been public but have reached the 0.4 bin at least once (position maximum:0.4 bin) with positions that have just exceeded the public disclosure threshold (position maximum: 0.5bin). In Panel A we compare the different positions along several investor and stock characteristics,reporting means, the difference in means ∆ and the respective p-value, clustering standard errors atthe investor-stock level. N denotes the number of observations (investor-stock-days) in each group. InPanel B we compare the performance of the two groups. For details on the calculation of the variables,see Table A.1. Additionally, we investigate the performance of positions that remain below the disclosurethreshold which originate from investors who have always remained just below the disclosure threshold(position and investor maximum: 0.4 bin). To measure performance, we form equal-weighted portfolios ofthe respective positions and regress the portfolio excess returns on market excess return (MKTRF), thesize (SMB) and value (HML) factors, and the momentum factor (WML), depending on the factor model.The table reports alphas (in bps per day) of the time-series regression, omitting the factor loadings forthe sake of brevity. The t-statistics are computed with Newey–West standard errors and are shown inparentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4) (5) (6)
Panel A: Stock and investor characteristics
Position max.: Position max.:0.4 bin 0.5 bin Diff. in means
Variable N Mean N Mean ∆ p-value
Investor characteristics:Dummy: Investor max. = position max. (%) 63,454 39.94 20,801 10.99 28.95∗∗∗(0.000)Dummy: Hedge fund (%) 63,454 81.51 20,801 72.80 8.71∗ (0.092)Dummy: Tax haven (%) 63,454 6.89 20,801 2.17 4.72∗∗ (0.020)Dummy: No FactSet entry (%) 63,454 17.61 20,801 2.65 14.96∗∗∗(0.000)Dummy: Private fund (%) 45,107 16.66 17,897 3.40 13.26∗∗∗(0.004)ln(total assets) 40,535 22.06 16,536 22.78 −0.72∗ (0.095)
Stock characteristics:ln(market value) 63,454 7.60 20,801 7.68 −0.08 (0.626)ln(turnover) 62,705 −5.55 20,345 −5.47 −0.08 (0.332)ln(Amihud illiquidity) 62,705 −5.93 20,345 −6.10 0.17 (0.452)ln(bid-ask spread) 62,990 −5.06 20,801 −5.14 0.08 (0.176)ln(return volatility) 62,990 0.75 20,801 0.70 0.05 (0.155)Institutional ownership (%) 63,454 33.07 20,801 35.61 −2.55 (0.212)Share of lendable stocks (%) 63,451 15.95 20,801 16.77 −0.82 (0.365)Dummy: Futures or listed options (%) 63,454 39.94 20,801 34.28 5.66 (0.276)Inventory concentration (HHI, 0-100) 63,451 15.20 20,801 14.16 1.04 (0.126)Lender concentration (HHI, 0-100) 63,451 21.35 20,801 20.82 0.53 (0.665)Utilization (%) 63,451 31.68 20,801 28.94 2.74 (0.343)Indicative fee (%) 63,439 1.61 20,800 1.30 0.31 (0.139)Dummy: “on special” (%) 63,454 23.02 20,801 20.91 2.11 (0.603)Standard deviation of fee (%) 62,957 0.69 20,793 0.59 0.10 (0.319)Standard deviation of utilization (%) 62,957 12.75 20,793 13.00 −0.25 (0.784)Fee tail risk (%) 62,957 3.11 20,793 2.54 0.57 (0.139)Utilization tail risk (%) 62,957 53.69 20,793 51.46 2.24 (0.487)
Continued on next page
57
Table 6 – Continued from previous page
(1) (2) (3) (4) (5) (6)
Panel B: Performance (in bps per day)
Positions max.: 0.4 bin Positions max.: 0.5 bin Diff. in means
Model α t-stat. α t-stat. ∆ t-stat.
CAPM -4.15* (-1.72) 1.34 (0.57) -5.49*** (-2.86)Fama–French -4.74*** (-2.66) 0.68 (0.39) -5.42*** (-2.89)Carhart -4.12** (-2.36) 0.75 (0.41) -4.87*** (-2.72)
Positions and investormax.: 0.4 bin Positions max.: 0.5 bin Diff. in means
Model α t-stat. α t-stat. ∆ t-stat.
CAPM -4.82 (-1.50) 1.34 (0.57) -6.16** (-2.57)Fama–French -5.58** (-2.39) 0.68 (0.39) -6.27*** (-2.66)Carhart -4.79** (-2.14) 0.75 (0.41) -5.54*** (-2.58)
Table 7Secretive short positions and fundamental newsThis table reports the average risk-adjusted daily return of stocks associated with secretive positions aswell as the decomposition of the return into returns generated on days when firm-specific fundamentalnews are released and days when we do not observe news. To measure performance, we regress theexcess return of each position day on the market excess return (MKTRF), the size (SMB) and value(HML) factors, and the momentum factor (WML), depending on the factor model. We employ theHoechle et al. (2016) regression framework that allows an exact replication of the calendar-time portfolioapproach. The table reports alpha estimates (in bps per day), omitting the factor loadings for the sakeof brevity. The t-statistics are computed with Driscoll and Kraay (1998) standard errors and are shownin parentheses. The estimates in Column (1) correspond to those reported in Panel B of Table 6 whentime-series regressions with Newey–West standard errors are used. Columns (2) to (4) report the averagereturns on earnings announcement days, firm-specific ad-hoc news days, and days on which we do notobserve news, respectively. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively
(1) (2) (3) (4)
Days with
Earnings No observedannounce- Ad-hoc firm-specific
All days ments news news
CAPM −4.15∗ −30.21 −69.41∗∗∗ −2.05(−1.72) (−1.47) (−3.32) (−0.98)
Fama–French −4.74∗∗∗ −27.82 −68.64∗∗∗ −2.76∗
(−2.66) (−1.40) (−3.21) (−1.67)Carhart −4.12∗∗ −27.21 −67.84∗∗∗ −2.15
(−2.36) (−1.35) (−3.31) (−1.32)
N 63,390 2,013 1,169 60,390
58
Table 8Short sellers’ avoidance and predictability of stock returnsThis table shows the average risk-adjusted returns of different portfolios constructed on the basis ofdifferent maximum reporting bins (Column (1) to (5)). For each investor-stock pair, we determine themaximum reporting interval reached during the sample period. In Panel A, for each maximum binwe report the average risk-adjusted return of portfolios that include all stocks for which we observe aposition at its maximum reporting bin. In Panel B, for each maximum bin interval we report the averagerisk-adjusted return of portfolios that include all stocks for which we observe a position that is outsidethe maximum reporting bin. In other words, it is the average return of the stocks not generated in themaximum bin phase. We form an position-weighted portfolio of stocks for each test. We then regressthe portfolio returns on market excess return (MKTRF), the size (SMB) and value (HML) factors, andthe momentum factor (WML), depending on the factor model. The table reports alphas (in bps perday) of the time-series regression, omitting the factor loadings for the sake of brevity. The t-statisticsare computed with Newey–West standard errors and are shown in parentheses. *, **, and *** indicatesignificance at the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4) (5)
Panel A: Performance during the maximum bin phaseMaximum reporting bin reached
0.2 0.3 0.4 0.5 0.6
CAPM 0.30 −0.59 −5.46∗∗ 6.37 0.80(0.14) (−0.34) (−2.00) (1.64) (0.21)
Fama–French −0.36 −1.09 −6.13∗∗∗ 6.30∗ −0.22(−0.21) (−0.59) (−2.83) (1.71) (−0.05)
Carhart −0.09 −1.22 −5.66∗∗ 6.02 −1.37(−0.05) (−0.87) (−2.57) (1.58) (−0.36)
Panel B: Performance outside the maximum bin phaseMaximum reporting bin reached
0.2 0.3 0.4 0.5 0.6
CAPM – −4.69 −1.70 0.69 −4.83(−1.58) (−0.64) (0.28) (−1.50)
Fama–French – −5.23∗∗ −2.02 −0.22 −5.62∗∗
(−2.16) (−1.01) (−0.12) (−2.26)Carhart – −4.67∗ −1.43 −0.18 −5.20∗∗
(−1.86) (−0.69) (−0.09) (−2.09)
59
Internet Appendix accompanying
Flying Under the Radar:
The Effects of Short-Sale Disclosure Rules on
Investor Behavior and Stock Prices
March 9, 2018
This appendix contains supplemental material for the paper “Flying Under the Radar: The
Effects of Short-Sale Disclosure Rules on Investor Behavior and Stock Prices.” We provide
additional descriptive statistics, and results of robustness tests, and analyses corroborating
the findings in the main text.
The roadmap is as follows: In Sections A and B we show additional figures describing
the stock universe underlying our analyses. Sections C, D, and E provide further evidence
of investors’ avoidance to disclose their position publicly using alternative estimation
procedures. In Sections F and G we test two alternative hypotheses related to the
disclosure threshold. Section H contains additional descriptive statistics and robustness
tests related to the documented return predictability using the calendar-time portfolio
approach.
Contents
A Additional sample characteristics ........................................................ 2
B Distribution of short position notifications ......................................... 3
C Alternative sample split: Positions approaching the disclosure thresh-
old from below ...................................................................................... 4
D Testing for discontinuity in the distribution of short positions.......... 5
E Fixed-effects regressions: Alternative model specifications................ 8
F Is there hesitance before eventually crossing the disclosure thresh-
old?........................................................................................................ 11
G Do investors jump to large short positions once they cross the
disclosure threshold? ............................................................................ 13
H Additional analyses to the calender-time approach ............................ 14
References ............................................................................................. 38
1
A. Additional sample characteristics
As described in the main text (see page 11), the vast majority (73.5%) of stocks with
short positions are in the highest market capitalization quartile, and almost all stocks
with short position notifications (95.8%) appear above the median value of the market
capitalization distribution. Figure IA.1(a) provides further details by plotting the frequency
of stocks with an open short position across the four quartiles of market capitalization.
The quartile breakpoints are defined according to the overall sample of stocks in the
German regulated stock market. As is evident from the figure, the distribution is highly
right-skewed irrespective of whether stocks have confidential or public short position
notifications. Similar findings emerge when plotting the distribution across the four
quartiles of institutional ownership, as in Figure IA.1(b). Also, stocks with short position
notifications are highly liquid as evident from Figure IA.1(c) (Amihud illiquidity measure)
and Figure IA.1(d) (bid-ask spread). Finally, stocks with short position notifications are
held by a considerable share of institutional investors indicating a sufficiently large supply
of stocks to borrow (Nagel, 2005).
The descriptive statistics in Table 1 of the main text show that almost 20% of all
German stocks in the regulated market have at least one short position notification over
the sample period. Figure IA.2(a) illustrates this share measured at each point in time. As
is evident from the blue line, the share of stocks with at least one short position notification
increases from 15% at the beginning of the sample to around 25% at the end of the sample
period. We also observe a slight increase of the stocks with at least one public short
position notification from around 7% to 10% of the overall stock universe (red line). As
suggested by Figure IA.2(b), the relative increase of stocks with short position notifications
is driven by both an increase in the absolute number of short position notifications and a
decline of the overall number of stocks in the regulated market.
2
B. Distribution of short position notifications
As discussed in the main text, we sort the short positions into bins of 10 basis points
(bps) each. The reason is that the exact (two-digit precision) value of the short position is
only recorded on the day at which the position crosses the respective thresholds at 0.20,
0.30, 0.40%, ... . This feature of the reporting rule mechanically generates a pattern in
the data, as illustrated in Figure IA.3. In this histogram we report the frequency of the
short position notifications with a bin width of 0.01%, i.e. at two-digit precision. The
histogram shows frequency spikes occurring at regular distance always at low and high
second decimal places.
For a better understanding as to why this pattern emerges, we report the frequency of
decimal places occurring in the short position notifications in Figure IA.4(a). It shows that
the u-shaped pattern within the reporting bins originates from two processes overlaying
each other: Positions that come from below are more likely to enter the reporting bin at
low second decimal places. Positions that come from above are more likely to enter the bin
at high second decimal places. The slightly asymmetric u-shape is due to a lower number
of decreases in the sample. Figure IA.4(b) displays the two conditional distributions of
second decimal places, showing a very symmetrical picture. Conditional on the positions
coming from below (increase), the likelihood of reporting a zero as a second decimal is
35.4%, whereas the corresponding likelihood of a nine as a second decimal place is only
2.3%. The mirror image emerges when conditioning on the positions coming from above
(decrease): The likelihood of reporting a short position with a zero as a second decimal
place is 2.7%, in contrast to a probability of 37.9% for a nine as a second decimal place.
The pattern around high and low second decimal places of short sale notifications, as
shown in Figures IA.3, IA.4(a), and IA.4(b), is the main reason why we collect positions
into reporting bins. Because of these“spikes”, it is unfeasible to compare density forecasts of
the true underlying distribution around the disclosure threshold. In the way the reporting
is set up, the share of positions with a reported value of 0.49 does not represent the true
3
fraction of positions with that value. In other words, investors’ short position is essentially
measured at one-digit precision.
C. Alternative sample split: Positions approaching the dis-
closure threshold from below
To uncover avoidance of investors to cross the disclosure threshold, we rely in our main
analyses on a sample split into positions at and positions below their record high. As
discussed in Section III, the avoidance to pass the disclosure threshold should be particularly
pronounced for positions at their record high relative to positions below their record high.
We repeat our main empirical tests from Figure 1(b), and those in Table 3 and Table 4
of the main text with a slightly different definition for the sample split. Specifically, we
divide all observations into position-days for which the last change was an increase, and
positions previously decreased.
The two measures relate to two extremes in terms of considered data history: The
measure in the main text is determined by the running maximum of the entire history of
the short position. The alternative measure is determined by the shortest possible window
of past position values, namely the maximum of the current and previous position value.
Related to the 0.4 bin, “previously increased” positions include position days, which were
never public in the past but also positions which were public prior to the last increase.
Because of the latter positions, we expect that the alternative sample split results in a
weaker avoidance effect relative to the “at-the-record-high” subsample from the main text.
Figure IA.5 plots the frequency of days with open short positions for the two subsamples.
Similar to the findings in the main text in Figure 1(b), our alternative sample split also
provides descriptive evidence for investors’ avoidance to cross the disclosure threshold.
Only for the “previously-increased” subsample we observe an accumulation of observations
in the 0.4 bin.
4
Table IA.1 re-conducts the analysis of Table 3 by using the alternative sample split.
In line with our main findings, Table IA.1 shows that the low probability of an increase
in the 0.4 bin is entirely driven by positions approaching the threshold from below. In
relative terms, for the subsample of previously increased positions, in the bin just below
the disclosure threshold it is 14% and 26% less likely to increase a short position than in
the two neighboring intervals. Comparing these results with findings of Table 3 with a
reduction between 20% and 34%, we observe that the avoidance for previously increased
positions is only slightly lower than for positions at the record high, corroborating the
robustness of our results.
Table IA.2 re-conducts the duration analysis of Table 4, but uses the alternative sample
split. In line with our main findings, the table shows that the abnormal long duration in
the 0.4 bin is unique to the sample of previously increased positions while avoidance is not
evident for the positions previously decreased (and public). In relative terms, the mean
duration in the bin just below the disclosure threshold is 27% and 55% longer than in
the two neighboring intervals. The corresponding reduction in Table 4 was 22% to 55%.
Results for median durations in Panel B are comparable.
Overall, we confirm our main findings from Sections III.A and Section III.B by using
only the last position change instead of the entire history of the short position evolution
as the sample split variable.
D. Testing for discontinuity in the distribution of short posi-
tions
If short positions accumulate below the disclosure threshold, one would expect to find a
jump or discontinuity in the distribution of short positions at or close to this point. A
common test to detect strategic behavior of investors around a given threshold is suggested
by Degeorge, Patel and Zeckhauser (1999).
5
When using this test in our setting, a critical feature of the data should be kept in
mind. As we pointed out in Section III and in more detail in Section A, although short
positions are reported with two-digit precision in principle, these positions are still measured
imprecisely: once a given position enters a reporting bin, say the 0.4 interval [0.40, 0.49),
it is no longer observed until it leaves this particular bin. This feature may give rise to
relatively few observations in the middle of each reporting bin, while observations near the
interval boundaries (0.20, 0.30 and so on) may be overrepresented. Therefore, the u-shaped
pattern may induce any statistical test to find evidence for a discontinuity at more than
one point in the support of the distribution. Thus, to detect strategic behavior of investors
around the 0.5 publication threshold, we compare the outcomes of the discontinuity test
around 0.5 to the results for the two neighboring boundaries, 0.4 and 0.6. While our
main interest is to test for a jump in the distribution below the disclosure threshold, the
two neighboring boundaries help us to disentangle investors’ strategic behavior from the
mechanical u-pattern within each interval. If possible discontinuities in the distribution
are only driven by the u-shaped reporting pattern described above, a powerful test may
detect discontinuities for any of these thresholds. If investors avoid publication of their
short positions, we expect evidence for a discontinuity to be even more pronounced below
the disclosure threshold.
Given this setup, we briefly review the testing procedure and discuss the outcomes
of the test. Let x be a random variable with probability density function f . For a given
random sample of this random variable, let x0, x1, . . . , xn, xn+1 be a set of n+ 2 equidistant
points. Let p (xi) be the proportion of observations that lie in the interval or bin [xi, xi+1),
which can be viewed as an estimate of f (xi), i = 0, . . . , n. Under the null hypothesis,
the density is smooth at a given point in the support of the distribution, while under the
6
alternative hypothesis the distribution may display a discontinuity at this point. To test
this null hypothesis, Degeorge et al. (1999) suggest the following test statistic:
τi =
∆p (xi)− meanxj∈Ri(r),xj 6=xi
{∆p (xj)}
s.d.xj∈Ri(r),xj 6=xi
{∆p (xj)}, (IA.1)
where ∆p (xi) = p (xi)−p (xi−1) is the first difference of the observed proportions, Ri (r) =
{xi−r, . . . , xi, . . . , xi+r} is a symmetric region around xi with radius r, and mean{·}, s.d.{·}
denote the empirical mean and standard deviation taken over selected points in this region.
To motivate this statistic, note that if f is smooth, the first derivative is continuous at
xi, and there are no jumps in the vicinity of this point. Hence, under the null hypothesis, if
∆p (xi) serves as an estimate of f ′ (xi), we expect ∆p (xi) not to be substantially different
from nearby ∆p (xj). Therefore, we reject the null hypothesis of smoothness for large
values of this statistic. Note that Degeorge et al. (1999) suggest to exclude xi itself when
computing the mean of the ∆p (xj) to increase the power of the test.
Although the test is intuitively appealing and imposes relatively few assumptions on
the data generating process, a disadvantage of the test is that the finite-sample or limiting
distribution of the statistic τ is not derived rigorously by Degeorge et al. (1999). Rather,
they conjecture that the statistic follows a t-distribution if ∆p (xi) is Gaussian in the
region R, although the number of degrees of freedom remain unspecified. Degeorge et al.
(1999) consider the standard ±2 critical values, and we highlight these benchmarks in
the figures below. Furthermore, this non-parametric test requires the user to specifiy the
radius r and the interval width of the bins [xi, xi+1). We consider an interval width of 0.01
percentage points as a natural choice, and provide results for several values of r in the
following application.
The top panel in Figure IA.6 shows the empirical distribution of short positions between
0.35% and 0.65%. These histograms can be viewed as close-ups of the histograms displayed
in Figure IA.3 over the relevant region.
7
We find that the frequency of short position notifications with the magnitude of 0.48
and 0.49 are larger than the frequency of positions just above the threshold of 0.50. This
pattern is unique to the disclosure threshold, given that for the two neighboring thresholds
we observe higher frequencies for positions just above the threshold. The bottom panel
presents the test statistic τ for each point in this range, where a radius of r = 0.06 is chosen.
It turns out that for xi = 0.48, the test provides the strongest evidence for rejecting the
null hypothesis of smoothness. The test statistic has a value of 5.1, which is substantially
larger than the second largest statistic of 3.6 at xi = 0.58, while there is no evidence for a
discontinuity around the 0.40 threshold.
Finally, Figures IA.7 and IA.8 repeat this testing procedure for r = 0.04 and r = 0.08,
yielding similar results. Overall, this exercise suggests that there is a discontinuity closely
below the disclosure threshold of 0.50, lending support for the evidence provided in
Section III.
E. Fixed-effects regressions: Alternative model specifications
Controlling for additional reporting bins separately. The results from the linear
probability model (LPM) from Table 5 document an abnormally low probability of increase
in the bin just below the publication threshold. In the regression specification of the main
text, we collect all position days with a position size of 0.6 and above in one reporting bin
to ensure a sufficient number of observations for each dummy variable. In this section,
we extend the number of reporting bins and include the intervals 0.7, 0.8, 0.9, and 1.0+
in the LPM, where 1.0+ denotes all positions shorting 1% or more of the stock’s shares
outstanding. This extension comes at the expense of a low number of observations for the
bins above 0.6. Despite this shortcoming, our main result of short sellers’ avoidance to
cross the publication threshold remains essentially the same.
8
In Table IA.3 we report regression results in line with the findings in the main text. In
particular, the probability of a position increase when the position is in the bin just below
the publication threshold is significantly lower relative to the corresponding probabilities in
the two neighboring bins. This result holds even when we account for different dimensions
of unobserved effects. For example, in the most saturated specification, controlling for both
time-varying stock and time-varying investor fixed effects, we find that the probability of
a position increase is the lowest among all reporting bins.
Fixed-effects regression with at-record high interactions. Throughout Section III
we document that avoidance to cross the disclosure threshold is particularly strong for
positions at their record high. To account for this difference in our fixed-effects regression
model from Equation (2b), we interact all bin dummies with a dummy variable indicating
positions at their record high. The indicator variable Record high is defined according to
Equation (1) of the main text. Formally, we estimate the following model:
yi,j,t+1 = β0 + β1 Just below thresholdi,j,t +∑k
βk k bin i,j,t
+ β1 Just below thresholdi,j,t × Record high i,j,t
+∑k
βk k bin i,j,t × Record high i,j,t
+ γ Record high i,j,t + ui,j,t+1. (IA.2)
where we control for all reporting bins up to 1.0+ separately, and estimate the model using
different specifications of ui,j,t+1 to account for unobserved stock and investor effects as in
Table IA.3.
Similar to Panel B of Table IA.3, in Table IA.4 we report the probability of increase for
the bin just below the publication threshold relative to all other reporting bins. Consistent
with the findings presented in the main text, we find evidence on avoidance only for
positions at their record high. The probability of position increase in the 0.4 bin is
9
significantly lower than in the neighboring bins for positions at their record high across
all specifications. For positions below their record high, in contrast, this difference is
insignificant or significantly positive.
Comparison: “increase”, “decrease” and “no change”. In our analysis we focus
on the probability of a position increase. Analyzing this event is a natural choice because we
are interested in investors’ potential avoidance to cross the disclosure threshold. Naturally,
at any given day, an investor has three options: increasing, decreasing, and not changing
its short position. Essentially, our analysis in the main text compares the event “increase”
against the events “decrease” and “no change”. In the following we shed more light on the
other two events. To do so, we follow our modeling approach of Equation (2b) from the
main text and define our three dependent variables as follows:
yincreasei,j,t = 1 (bini,j,t > bini,j,t−1) , (IA.3)
yno changei,j,t = 1 (bini,j,t = bini,j,t−1) , and (IA.4)
ydecreasei,j,t = 1 (bini,j,t < bini,j,t−1) . (IA.5)
where bini,j,t denotes the reporting bins (0.2, 0.3, 0.4, 0.5, ...). In case of avoidance, we
expect that the probability of not changing the short position in the bin 0.4 is higher than
in the two neighboring bins 0.3 and 0.5. Moreover, given the longer duration in bin 0.4
(see Table 4 in the main text), we expect that the probability of decrease in the bin just
below the publication threshold is lower than in the two neighboring bins.
The results of this extended analysis, using the most saturated fixed-effects model,
are shown in Table IA.5. As in Panel B of Table IA.3, and Table IA.4, we report the
estimates for the bin just below the publication threshold relative to all other reporting
bins. Column (1) shows the relative probability of increase, which corresponds to our
previous analysis shown in Table IA.3, Panel B, Column 5. Column (2) of Table IA.5
10
displays the results for the relative probability of not changing the bin on the next day.
Coefficients of all bins are significantly positive. This result implies that the probability
of staying in the same bin is significantly higher just below the threshold compared to
all other reporting bins. The difference in the probability is economically sizable when
compared to the neighboring bins, amounting to 1.51 to 2.42 percentage points. Since
investors are avoidant to cross the disclosure threshold, they stay longer in the 0.4 bin.
This result reflects our finding from Section III.B of the main text: The duration in the
0.4 bin is the longest among all reporting bins.
In Column (3) we observe that the probability of a position decrease is lower in the
0.4 bin compared to its neighboring bins. This finding can be explained by the longer
duration in the 0.4 bin. In particular, days with a position decrease decline in relative
terms because investors increase the days they stay in the 0.4 bin.
F. Is there hesitance before eventually crossing the disclosure
threshold?
Our results from the main text present strong evidence for investors’ avoidance of crossing
the disclosure threshold. In this section we strive to better understand whether this finding
is driven by positions that do not cross the publication threshold (“absolute avoidance”)
or by positions that eventually cross the publication threshold some time in the future
(“hesitance”).
The analysis of Table 3 in the main text already shows that there is an absolute
avoidance to cross the disclosure threshold. In particular, in this exercise we eliminate the
duration dimension: Instead of looking at the probability of a position increase on the next
day, we investigate the probability that the next position change in the future is an increase.
The low probability of this event in the 0.4 bin shows that investors absolutely avoid
to cross the threshold and that there is not merely a hesitance followed by an eventual
11
position increase. This result is line with Figure 2 of the main text, which shows that
positions that reach the reporting bin below the publication threshold but never crossing
it subsume the largest group of position-days. Moreover, the results of Table 6 suggest
that this absolute avoidance is strongly investor-specific, with investors sticking to their
decision either to cross the threshold or not to cross the threshold.
Nevertheless, there may still exist hesitance to cross the threshold present among those
investors that eventually cross the threshold. We investigate this alternative hypothesis by
repeating the duration analysis of Table 4 but focusing on the subsample of positions that
eventually increase their position value with the next change. If hesitance is present before
crossing the disclosure threshold, we expect a longer duration in the 0.4 bin compared to
its neighboring bins.
The results of this exercise are shown in Table IA.6 and provide no clear support for the
hypothesis that investors are hesitant before eventually crossing the disclosure threshold.
The mean duration (Panel A) in the 0.4 bin is only slightly larger compared to the duration
in the bin 0.3 and 0.5, respectively, but the difference is insignificant. We find similar
results when conditioning on positions at their record high, for which we previously found
the strongest avoidance effect. The median duration, which is more robust to extreme
observations of the highly right-skewed duration distribution, are shown in Panel B. Here
we find that the median time spent in the 0.4 bin is identical to the duration spent in
the neighboring bins. Taken together, these results support the notion that investors’
avoidance is driven by the “absolute avoidance” to cross the publication threshold rather
than investors being hesitant before eventually crossing the threshold.
12
G. Do investors jump to large short positions once they cross
the disclosure threshold?
The main result of the paper is that a sizable share of investors are avoidant to cross
the disclosure threshold. A question that arises in this context is how investors behave if
they have decided to cross the disclosure threshold. A possible behavior when crossing
the threshold could be that investors immediately try to establish an abnormally large
position. The rationale may be as follows: If the investor’s targeted position is above but
not much larger than the publication threshold, the investor might prefer a position just
below the threshold. The costs of a position publication exceed the potential benefits of
the larger targeted short position. As a consequence, only investors that target a very large
position (associated with higher benefits) are willing to tolerate publication and increase
their position accordingly. If so, we would expect that a position increase is particularly
large when the increase was initiated from the bin just below the publication threshold.
We explore this conjecture by looking at future position changes across all reporting
bins. Specifically, we compute the average position change, given that a position increase
occurs. Then, we compare the average position increase across the various reporting bins
from where they were initiated. Evidence in Table IA.7 does not support this “large jump
hypothesis”. The average position increase initiated from the 0.4 bin is 0.114%, which is
not significantly larger than the increase from any other bin. There is also no unusual
pattern when conditioning on positions at their record high or positions below their record
high. Next, we not only look at the mean increase, but at various percentiles of position
increases. As apparent from Table IA.8, the median increase is 0.1 and identical across all
starting bins including the 0.4 bin. Also, when looking at higher percentiles, we find no
abnormal patterns for the bin just below the disclosure threshold even in the tails of the
distribution. No other insights arise from the sample split of positions at or below their
13
record high. Overall, the results do not support the conjecture that investors immediately
jump to very high positions once they decided to cross the disclosure threshold.
H. Additional analyses to the calender-time approach
In Section V.A of the main text we find that a portfolio consisting of stocks with secretive
positions just below the publication threshold yields on average a significant abnormal
return of -5.46 to -6.13 bps per day. In this section of the Appendix we briefly illustrate
the composition of the portfolio over time.
As indicated on page 33 in the main text, the average number of stocks in this portfolio
is 32 ranging from 21 up to 49 stocks on a given day. In Figure IA.9 we plot the number
of distinct stocks in the portfolio over time. In addition, the figure includes the time series
of positions in this portfolio. The number of positions is slightly higher given that there
can be more than one secretive position that is just below the publication threshold for
the same stock by different investors. Two notable findings emerge from the figure. First,
the number of stocks and positions in the portfolio only slightly varies, with an increasing
trend during our sample period. Second, the majority of positions come from distinct
stocks, suggesting that investors’ publication avoidance and its consequences for price
efficiency documented in this study are prevalent effects in the financial market.
In this section we conduct a number of variations to our sample and variable definitions
to test the robustness of the results from Section V.A. Table IA.9 reports the average
abnormal returns of the benchmark model from Panel A of Table 8 and those of our
sensitivity analyses.
In the main analysis we include a stock into the portfolio if we observe a secretive
position that is in the bin just below the publication threshold. Consequently, the returns
within the portfolio are weighted by the number of positions. The rationale is that we
expect overpricing to be positively related to investors’ avoidance and to the number of
14
positions that are possibly constrained. As a robustness test, we now use equal weighting.
That is, we include a stock in the portfolio if we observe one or more secretive positions
that are in the bin just below the publication threshold. As is evident from Panel B of
Table IA.9, employing the alternative, less powerful, portfolio weighting yields only slightly
lower returns in absolute terms relative to the benchmark result in Panel A.
In our second robustness test, we exclude all short positions for which the stock price on
the day before crossing the reporting threshold of 0.2% is below 1 EUR. Such a restriction
ensures that our main overpricing results are not driven by small-cap and illiquid stocks. As
already noted in Section II, the sample of stocks with open short positions consists almost
entirely of stocks above the median market capitalization of the German stock universe.
As a consequence, the exclusion of penny stocks only marginally affects our sample size.
Using the 1 EUR filter, we drop merely four stocks from our sample. Accordingly, the
alphas in Panel C are virtually the same as in the benchmark model in Panel A.
In Panel D we conduct another change in our sample by defining investors to be avoidant
to cross the publication threshold if they spend at least five days in their maximum bin of
0.4. In comparison, in the main analysis we require only a minimum of two days spent in
the maximum bin of 0.4. With the more conservative definition of being avoidant using
restrictions on the shorting period, we expect to better identify avoidance, short-selling
impediments, and, as a result, more accurate estimates. In line with this expectation, we
indeed find that the standard errors of the abnormal returns decrease and the statistical
significance of the effect increases across all asset pricing models compared to the baseline
model in Panel A, despite a slight decrease in the economic magnitude.
Our main results on overpricing are also robust to changes in the asset pricing factors.
Namely, in Panel E of Table IA.9, we use European factor portfolios, which are also
obtained from Andrea Frazzini’s data library, provided through AQR’s website.1 Using
European factors results in economically comparable, but noisier alpha estimates. This
1See: www.aqr.com/library/data-sets/quality-minus-junk-factors-daily/data
15
result is in line with Griffin (2002) who documents that local factor portfolios are generally
better suited for performance evaluation.
16
(a) Market capitalization
020
4060
80Fr
eque
ncy
(in %
)
1 2 3 4
AllConfidentialPublic
Short-sale notifications:
(b) Institutional ownership
020
4060
80Fr
eque
ncy
(in %
)
1 2 3 4
AllConfidentialPublic
Short-sale notifications:
(c) Amihud illiquidity
020
4060
80Fr
eque
ncy
(in %
)
1 2 3 4
AllConfidentialPublic
Short-sale notifications:
(d) Bid-ask spread
020
4060
80Fr
eque
ncy
(in %
)
1 2 3 4
AllConfidentialPublic
Short-sale notifications:
Figure IA.1:Distribution of stocks with open short-position notifications across size, insti-tutional ownership, and illiquidityThis figure displays the distribution of stock-days with open short-sale notifications across quartiles ofmarket capitalization (Figure IA.1(a)), institutional ownership (Figure IA.1(b)), the Amihud illiquiditymeasure (Figure IA.1(c)), and the bid-ask spread (Figure IA.1(d)). The quartile breakpoints are definedaccording to the overall sample of stocks in the German regulated stock market. All short-sale notifi-cations subsume stocks with either public or confidential short-sale notifications, confidential short-salenotifications consist of stocks with positions above 0.2% but below 0.5% of issued share capital, and publicshort-sale notifications consist of stocks with at least one short position of 0.5% or higher. The overallsample contains all German domestic equity in the regulated market from November 5, 2012, to March 31,2015.
17
(a) Percentages
05
1015
2025
30S
hare
of s
tock
s (in
%)
Jan-2013 July-2013 Jan-2014 July-2014 Jan-2015
Stocks with short position notificationsStocks with public short position notifications
(b) Number of stocks
020
040
060
0N
umbe
r of s
tock
s
Jan-2013 July-2013 Jan-2014 July-2014 Jan-2015
All stocksStocks with short position notificationsStocks with public short position notifications
Figure IA.2:Stocks with short position notifications over timeFigure IA.2(a) displays the percentage of stocks that have at least one short position notification belowthe publication threshold (≥ 0.2%) and the percentage of stocks that have at least one short positionnotification above the publication threshold (≥ 0.5%) over time, respectively. Figure IA.2(b) shows thesame statistics over time in absolute terms as well as the overall number of stocks in our sample. Theoverall sample contains all German domestic equity in the regulated market from November 5, 2012, toMarch 31, 2015.
18
05
1015
Freq
uenc
y (in
%)
0.20 0.30 0.40 0.50 0.60 0.70Reported short position (in % of issued share capital)
Figure IA.3:Detailed histogram of reported short positionsThis figure displays the detailed frequency histogram of reported short positions with a bin width of0.01%. Positions greater than 0.70% are truncated for readability. The sample contains all short positionnotifications in the German regulated equity market from November 5, 2012, to March 31, 2015.
19
(a) Overall distribution
05
1015
2025
Freq
uenc
y (in
%)
0 1 2 3 4 5 6 7 8 9Second decimal
Position coming from belowPosition coming from above
(b) Conditional distribution
010
2030
40Fr
eque
ncy
(in %
)
0 1 2 3 4 5 6 7 8 9Second decimal
Position coming from belowPosition coming from above
Figure IA.4:Second decimal places of the short position notificationFigure IA.4(a) displays the relative frequency of second decimal places of short-position notifications.Figure IA.4(b) shows the frequency distribution conditional on positions coming from above or below. Thesample contains all short position notifications in the German regulated equity market from November 5,2012, to March 31, 2015.
20
Public disclosure threshold
010
2030
4050
Rel
ativ
e fre
quen
cy (%
)
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
Reporting bins
Previously increased positionsPreviously decreased positions
Figure IA.5:Distribution of open short positionsThis figure displays the relative frequency separately for short positions that previously increased andpreviously decreased, respectively. Reporting intervals greater than 1.6% are truncated for readability.The overall sample contains all German domestic equity in the regulated market from November 5, 2012,to March 31, 2015.
21
Reporting Threshold
01
23
4
Freq
uenc
y (in
%)
.35 .36 .37 .38 .39 .4 .41 .42 .43 .44
Reported short position (in % of issued share capital)
-5-4
-3-2
-10
12
34
5
Test
Sta
tistic
.35 .36 .37 .38 .39 .4 .41 .42 .43 .44
Reported short position (in % of issued share capital)
Disclosure Threshold
0.3
7.7
41.
111.
481.
85
Freq
uenc
y (in
%)
.45 .46 .47 .48 .49 .5 .51 .52 .53 .54
Reported short position (in % of issued share capital)
-5-4
-3-2
-10
12
34
5
Test
Sta
tistic
.45 .46 .47 .48 .49 .5 .51 .52 .53 .54
Reported short position (in % of issued share capital)
Reporting Threshold
0.1
68.3
36.5
04.6
72.8
4
Freq
uenc
y (in
%)
.55 .56 .57 .58 .59 .6 .61 .62 .63 .64
Reported short position (in % of issued share capital)
-5-4
-3-2
-10
12
34
5
Test
Sta
tistic
.55 .56 .57 .58 .59 .6 .61 .62 .63 .64
Reported short position (in % of issued share capital)
Figure IA.6:Testing for discontinuity in the distribution of short positionsThe top panel displays the histogram of reported short positions with a bin width of 0.01 percentagepoints for a range of positions between 0.35% and 0.65%. These histograms are close-ups of the relevanthistograms in figure IA.3. The bottom panel shows the test statistic suggested by Degeorge et al. (1999),based on a radius of r = 0.06, see Equation (IA.1). Under the null hypothesis, the density function ofshort positions is smooth. Large values of the statistic provide evidence for the alternative hypothesis andhence a discontinuity in the distribution. Critical values of ± 2 are highlighted as a benchmark.
22
Reporting Threshold
01
23
4
Freq
uenc
y (in
%)
.35 .36 .37 .38 .39 .4 .41 .42 .43 .44
Reported short position (in % of issued share capital)
-5-4
-3-2
-10
12
34
5
Test
Sta
tistic
.35 .36 .37 .38 .39 .4 .41 .42 .43 .44
Reported short position (in % of issued share capital)
Disclosure Threshold
0.3
7.7
41.
111.
481.
85
Freq
uenc
y (in
%)
.45 .46 .47 .48 .49 .5 .51 .52 .53 .54
Reported short position (in % of issued share capital)
-5-4
-3-2
-10
12
34
5
Test
Sta
tistic
.45 .46 .47 .48 .49 .5 .51 .52 .53 .54
Reported short position (in % of issued share capital)
Reporting Threshold
0.1
68.3
36.5
04.6
72.8
4
Freq
uenc
y (in
%)
.55 .56 .57 .58 .59 .6 .61 .62 .63 .64
Reported short position (in % of issued share capital)
-5-4
-3-2
-10
12
34
5
Test
Sta
tistic
.55 .56 .57 .58 .59 .6 .61 .62 .63 .64
Reported short position (in % of issued share capital)
Figure IA.7:Testing for discontinuity in the distribution of short positionsThe top panel displays the histogram of reported short positions with a bin width of 0.01 percentagepoints for a range of positions between 0.35% and 0.65%. These histograms are close-ups of the relevanthistograms in figure IA.3. The bottom panel shows the test statistic suggested by Degeorge et al. (1999),based on a radius of r = 0.04, see Equation (IA.1). Under the null hypothesis, the density function ofshort positions is smooth. Large values of the statistic provide evidence for the alternative hypothesis andhence a discontinuity in the distribution. Critical values of ± 2 are highlighted as a benchmark.
23
Reporting Threshold
01
23
4
Freq
uenc
y (in
%)
.35 .36 .37 .38 .39 .4 .41 .42 .43 .44
Reported short position (in % of issued share capital)
-5-4
-3-2
-10
12
34
5
Test
Sta
tistic
.35 .36 .37 .38 .39 .4 .41 .42 .43 .44
Reported short position (in % of issued share capital)
Disclosure Threshold
0.3
7.7
41.
111.
481.
85
Freq
uenc
y (in
%)
.45 .46 .47 .48 .49 .5 .51 .52 .53 .54
Reported short position (in % of issued share capital)
-5-4
-3-2
-10
12
34
5
Test
Sta
tistic
.45 .46 .47 .48 .49 .5 .51 .52 .53 .54
Reported short position (in % of issued share capital)
Reporting Threshold
0.1
68.3
36.5
04.6
72.8
4
Freq
uenc
y (in
%)
.55 .56 .57 .58 .59 .6 .61 .62 .63 .64
Reported short position (in % of issued share capital)
-5-4
-3-2
-10
12
34
5
Test
Sta
tistic
.55 .56 .57 .58 .59 .6 .61 .62 .63 .64
Reported short position (in % of issued share capital)
Figure IA.8:Testing for discontinuity in the distribution of short positionsThe top panel displays the histogram of reported short positions with a bin width of 0.01 percentagepoints for a range of positions between 0.35% and 0.65%. These histograms are close-ups of the relevanthistograms in figure IA.3. The bottom panel shows the test statistic suggested by Degeorge et al. (1999),based on a radius of r = 0.08, see Equation (IA.1). Under the null hypothesis, the density function ofshort positions is smooth. Large values of the statistic provide evidence for the alternative hypothesis andhence a discontinuity in the distribution. Critical values of ± 2 are highlighted as a benchmark.
24
020
4060
80N
umbe
r of p
ositi
ons/
stoc
ks
Jan-2013 July-2013 Jan-2014 July-2014 Jan-2015Date
Number of positionsNumber of distinct stocks
Secretive positions just below threshold
Number of positions and distinct stocksin calendar-time portfolio:
Figure IA.9:Number of stocks with secretive positions just below the publication thresholdThis figure shows the number of secretive positions that are just below the publication threshold overtime (green line). The red line is the corresponding number of distinct stocks with at least one secretiveposition that is just below the publication threshold. The overall sample contains all German domesticequity in the regulated market from November 5, 2012, to March 31, 2015.
-
25
Table
IA.1
:P
robabilit
yof
short
posi
tion
incr
ease
:A
ltern
ati
ve
sam
ple
split
This
table
rep
eats
the
anal
ysi
sof
Tab
le3
usi
ng
anal
tern
ativ
esa
mple
split
ofsh
ort
pos
itio
ns
pre
vio
usl
yin
crea
sed
and
pos
itio
ns
pre
vio
usl
ydec
reas
ed.
The
table
rep
orts
the
pro
bab
ilit
yof
incr
easi
ng
ash
ort
pos
itio
n(i
.e.,
chan
ging
toa
hig
her
rep
orti
ng
bin
),gi
ven
that
anin
vest
orcu
rren
tly
has
ap
osit
ion
ina
spec
ific
bin
.In
addit
ion,
itd
ispla
ys
the
diff
eren
cein
pro
babilit
yof
row
(3)
rela
tive
toro
w(j
),th
e0.4
rep
ort
ing
bin
just
bel
owth
epu
blica
tion
thre
shol
d(s
had
edin
gray
),an
dth
ep-
valu
efo
rth
ediff
eren
ces
inm
eans.
*,**
,an
d**
*in
dic
ate
sign
ifica
nce
atth
e10
%,
5%,
and
1%le
vels
,re
spec
tive
ly.
Posi
tion
sP
osi
tions
Ove
rall
pre
vio
usl
yin
crea
sed
pre
vio
usl
ydec
rease
d
Pro
bab
ilit
yD
iffer
ence
Pro
babilit
yD
iffer
ence
Pro
babilit
yD
iffer
ence
Bin
ofin
crea
sero
w(3
)-
(j)
p-v
alu
eof
incr
ease
row
(3)
-(j
)p-v
alu
eof
incr
ease
row
(3)
-(j
)p-v
alu
e
undis-
closed
(1)
0.2
0.36
4−
0.01
0(0
.469)
0.3
72−
0.0
10
(0.5
24)
0.3
45−
0.012
(0.6
37)
(2)
0.3
0.39
1−
0.03
7**
(0.0
14)
0.4
22−
0.0
60***
(0.0
01)
0.3
28
0.005
(0.8
46)
(3)
0.4
0.35
40.3
62
0.3
33
disclosed
(4)
0.5
0.40
8−
0.05
4***
(0.0
07)
0.4
87−
0.1
25***
(0.0
00)
0.2
63
0.071**
(0.0
37)
(5)
0.6
0.47
6−
0.12
2***
(0.0
00)
0.5
73−
0.2
11***
(0.0
00)
0.3
13
0.020
(0.6
06)
(6)
0.7
0.44
0−
0.08
6***
(0.0
01)
0.5
61−
0.1
99***
(0.0
00)
0.2
47
0.086**
(0.0
36)
(7)
0.8
0.44
9−
0.09
5***
(0.0
01)
0.6
04−
0.2
42***
(0.0
00)
0.1
75
0.158***
(0.0
01)
(8)
0.9
0.47
7−
0.12
3***
(0.0
00)
0.6
67−
0.3
05***
(0.0
00)
0.1
88
0.146***
(0.0
05)
(9)≥
1.0
0.47
4−
0.12
0***
(0.0
00)
0.6
49−
0.2
87***
(0.0
00)
0.2
14
0.120***
(0.0
00)
26
Table
IA.2
:D
ura
tion
wit
hin
rep
ort
ing
bin
s:A
ltern
ati
ve
sam
ple
split
This
table
rep
eats
the
dura
tion
analy
sis
of
Table
4usi
ng
an
alt
ernati
ve
sam
ple
split
of
short
posi
tions
pre
vio
usl
yin
crea
sed
and
posi
tions
pre
vio
usl
ydec
reas
ed.
The
tab
lere
por
tsth
em
ean
and
med
ian
num
ber
oftr
adin
gd
ays
spen
tin
each
dis
clos
ure
bin
(Pan
els
Aan
dB
,re
spec
tive
ly).
Inad
dit
ion
,it
dis
pla
ys
the
diff
eren
cein
mea
n(m
edia
n)
dura
tion
ofro
w(3
)re
lati
veto
row
(j),
the
0.4
rep
orti
ng
bin
just
bel
owth
epu
blica
tion
thre
shol
d(s
had
edin
gray
),an
dth
ep-v
alu
efo
rdiff
eren
ces
inm
eans
(med
ians)
.*,
**,
an
d***
indic
ate
sign
ifica
nce
at
the
10%
,5%
,an
d1%
leve
ls,
resp
ecti
vely
.
Panel
A:
Mean
dura
tion
wit
hin
rep
ort
ing
bin
intr
ad
ing
days
Posi
tions
Posi
tions
Ove
rall
pre
vio
usl
yin
crea
sed
pre
vio
usl
ydec
rease
d
Mea
nD
iffer
ence
Mea
nD
iffer
ence
Mea
nD
iffer
ence
Bin
du
rati
onro
w(3
)-
(j)
p-v
alu
edu
rati
on
row
(3)
-(j
)p-v
alu
edu
rati
on
row
(3)
-(j
)p-v
alu
e
undis-
closed
(1)
0.2
18.3
2.2**
(0.0
19)
19.4
3.8***
(0.0
01)
15.6
−3.
4**
(0.0
27)
(2)
0.3
17.0
3.6***
(0.0
00)
18.3
4.9***
(0.0
00)
13.3
−1.
1(0
.429)
(3)
0.4
20.6
23.2
12.2
disclosed
(4)
0.5
14.3
6.2***
(0.0
00)
15.0
8.3***
(0.0
00)
12.8
−0.
7(0
.690)
(5)
0.6
13.7
6.8***
(0.0
00)
15.3
7.9***
(0.0
00)
10.7
1.4
(0.4
51)
(6)
0.7
13.3
7.3***
(0.0
00)
14.6
8.6***
(0.0
00)
11.0
1.2
(0.5
70)
(7)
0.8
13.6
7.0***
(0.0
01)
15.1
8.1***
(0.0
03)
10.0
2.1
(0.3
73)
(8)
0.9
11.9
8.7***
(0.0
00)
11.8
11.
5***
(0.0
00)
9.4
2.7
(0.2
96)
(9)≥
1.0
12.6
8.0***
(0.0
00)
14.9
8.3***
(0.0
00)
8.2
3.9***
(0.0
02)
27
Table
IA.2
–C
onti
nued
Panel
B:
Media
ndura
tion
wit
hin
rep
ort
ing
bin
intr
adin
gdays
Posi
tion
sP
osi
tions
Ove
rall
pre
vio
usl
yin
crea
sed
pre
vio
usl
ydec
rease
d
Med
ian
Diff
eren
ceM
edia
nD
iffer
ence
Med
ian
Diff
eren
ceB
indu
rati
onro
w(3
)-
(j)
p-v
alu
edu
rati
on
row
(3)
-(j
)p-v
alu
ed
ura
tion
row
(3)
-(j
)p-v
alu
e
undis-
closed
(1)
0.2
61***
(0.0
04)
72***
(0.0
00)
50*
(0.0
96)
(2)
0.3
61***
(0.0
03)
72***
(0.0
00)
50
(0.9
84)
(3)
0.4
79
5
disclosed
(4)
0.5
52***
(0.0
00)
63***
(0.0
00)
41
(0.9
34)
(5)
0.6
52***
(0.0
00)
54***
(0.0
00)
41
(0.3
93)
(6)
0.7
43***
(0.0
00)
54***
(0.0
00)
32
(0.1
16)
(7)
0.8
43***
(0.0
00)
54***
(0.0
00)
23***
(0.0
07)
(8)
0.9
43***
(0.0
00)
54***
(0.0
00)
32*
(0.0
54)
(9)≥
1.0
43*
**
(0.0
00)
45***
(0.0
00)
32***
(0.0
00)
28
Table IA.3:Probability of increase: Regression approach with fixed effectsThis table shows estimates from the linear probability model
yi,j,t+1 = β0 + β1 Just below thresholdi,j,t +∑k
βk k bini,j,t + ui,j,t+1,
with yi,j,t+1 = 1 (bini,j,t+1 > bini,j,t), and ui,j,t+1 models various fixed effects and an error term. k bini,j,t
is equal to 1 if investor j has a short position in stock i in the bin k at day t, with k ∈ {0.2, 0.5, 0.6, . . . , 1.0+},where 1.0+ indicates position bins greater or equal to 1.0. The unconditional probability of increasing ashort position to enter the next bin (estimated by the sample average of yi,j,t) is 2.0%. Panel A showscoefficient estimates using the 0.3 bin as omitted reference category. Panel B tests for differences inthe probability of increase between the bin just below the threshold and all other bins. The estimatedcoefficients are scaled to reflect changes in percentage points. Standard errors are clustered at the investor-stock and time level. The t-statistics are given in parentheses, and *, **, and *** indicate significance atthe 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4) (5)
Panel A: Regression with reporting bin dummies
0.2 bin -0.43*** -0.16 -0.63*** -0.29** -0.41***(-3.41) (-1.30) (-4.72) (-2.20) (-2.76)
0.3 bin – – – – –
0.4 bin (Just below threshold) -0.54*** -0.92*** -0.68*** -0.61*** -1.08***(-3.66) (-5.72) (-4.23) (-3.55) (-5.40)
0.5 bin 0.64*** -0.20 0.45* 0.27 -0.16(2.74) (-0.87) (1.85) (1.10) (-0.53)
0.6 bin 1.26*** 0.65** 1.37*** 0.99*** 0.68*(3.70) (2.03) (4.05) (2.61) (1.85)
0.7 bin 0.92** 0.61* 0.97*** 0.91** 0.37(2.18) (1.87) (2.73) (2.52) (1.02)
0.8 bin 0.69 0.78* 0.69 1.65*** 1.07**(1.35) (1.79) (1.58) (3.57) (2.22)
0.9 bin 1.36* 0.91* 0.93 1.68*** 0.94*(1.90) (1.92) (1.39) (3.76) (1.87)
≥ 1.0 bin 1.42*** 0.74** 1.44*** 2.40*** 1.61***(3.90) (2.20) (4.59) (6.11) (4.46)
Fixed effects:Time Yes Yes No No NoStock No Yes No No NoInvestor No Yes No No NoStock × time No No Yes No YesInvestor × time No No No Yes Yes
R2 (in %) 0.61 2.56 15.71 19.58 38.09Within R2 (in %) 0.20 0.08 0.22 0.22 0.15Number of observations 278,003 277,999 255,274 228,358 202,057
Continued on next page
29
Table IA.3 – Continued from previous page
(1) (2) (3) (4) (5)
Panel B: Differences in the probability of increase
Just below threshold – 0.2 bin -0.11 -0.76*** -0.05 -0.31** -0.66***(-0.85) (-4.70) (-0.31) (-1.99) (-3.34)
Just below threshold – 0.3 bin -0.54*** -0.92*** -0.68*** -0.61*** -1.08***(-3.66) (-5.72) (-4.23) (-3.55) (-5.40)
Just below threshold – 0.5 bin -1.18*** -0.72*** -1.13*** -0.88*** -0.92***(-5.04) (-2.92) (-4.49) (-3.38) (-2.89)
Just below threshold – 0.6 bin -1.81*** -1.57*** -2.05*** -1.60*** -1.76***(-5.43) (-4.85) (-6.10) (-4.22) (-4.69)
Just below threshold – 0.7 bin -1.47*** -1.53*** -1.65*** -1.52*** -1.45***(-3.52) (-4.60) (-4.77) (-4.25) (-3.98)
Just below threshold – 0.8 bin -1.23** -1.70*** -1.37*** -2.26*** -2.15***(-2.44) (-3.86) (-3.09) (-4.79) (-4.30)
Just below threshold – 0.9 bin -1.91*** -1.83*** -1.61** -2.28*** -2.02***(-2.68) (-3.90) (-2.43) (-5.24) (-4.18)
Just below threshold – (≥ 1.0 bin) -1.96*** -1.66*** -2.12*** -3.01*** -2.68***(-5.53) (-5.03) (-6.75) (-7.64) (-7.40)
30
Table IA.4:Probability of increase: Fixed-effects regression approach with at-record-highinteractionsThis table repeats the analysis of Table IA.3 but interacts all coefficients with a dummy variable indicatingwhether positions are at the record high or not. The estimated coefficients are differences in the probabilityof increase between the bin just below the threshold and all other bins. Standard errors are clustered atthe investor-stock and time level. The t-statistics are given in parentheses, and *, **, and *** indicatesignificance at the 10%, 5%, and 1% levels, respectively.
(1) (2) (3) (4)
Just below threshold – 0.2 bin 0.07 0.96** 0.71** 0.57(0.22) (2.42) (2.29) (1.60)
Just below threshold – 0.3 bin 0.05 0.37 0.57* 0.31(0.17) (0.99) (1.82) (0.88)
Just below threshold – 0.5 bin 0.82** 0.80* 0.56 0.77(2.29) (1.88) (1.34) (1.54)
Just below threshold – 0.6 bin -0.79 -0.62 -1.00 -0.61(-1.53) (-0.99) (-1.59) (-0.89)
Just below threshold – 0.7 bin 0.11 0.67 0.79* 1.02**(0.29) (1.24) (1.69) (2.15)
Just below threshold – 0.8 bin 0.24 0.73 0.14 0.22(0.41) (1.10) (0.21) (0.29)
Just below threshold – 0.9 bin 0.52 0.42 0.22 0.49(0.96) (0.61) (0.35) (0.63)
Just below threshold – (≥ 1.0 bin) 0.36 0.20 -0.47 0.41(0.76) (0.39) (-0.78) (0.68)
Record high × (Just below threshold – 0.2 bin) -0.94*** -1.05** -1.15*** -1.37***(-2.86) (-2.54) (-3.32) (-3.40)
Record high × (Just below threshold – 0.3 bin) -1.21*** -1.11*** -1.41*** -1.68***(-3.67) (-2.73) (-3.92) (-3.96)
Record high × (Just below threshold – 0.5 bin) -2.26*** -2.55*** -1.88*** -2.37***(-4.94) (-4.98) (-3.63) (-3.93)
Record high × (Just below threshold – 0.6 bin) -1.01 -1.77** -0.58 -1.59*(-1.59) (-2.36) (-0.78) (-1.91)
Record high × (Just below threshold – 0.7 bin) -2.65*** -3.35*** -3.59*** -4.20***(-3.73) (-4.17) (-4.97) (-4.89)
Record high × (Just below threshold – 0.8 bin) -2.92*** -2.79*** -3.45*** -3.59***(-3.42) (-3.14) (-3.75) (-3.42)
Record high × (Just below threshold – 0.9 bin) -3.46*** -2.66** -3.50*** -3.36***(-3.84) (-2.21) (-3.64) (-3.38)
Record high × (Just below threshold – (≥ 1.0 bin)) -2.86*** -3.08*** -3.49*** -4.42***(-5.41) (-5.08) (-4.87) (-6.18)
Record high 1.31*** 2.30*** 1.89*** 2.15***(4.51) (6.02) (6.05) (5.86)
Fixed effects:Time Yes Yes Yes NoStock Yes No No NoInvestor Yes No No NoStock × time No Yes No YesInvestor × time No No Yes Yes
R2 (in %) 2.62 15.86 19.67 38.17Within R2 (in %) 0.14 0.39 0.33 0.28Number of observations 277,999 255,274 228,358 202,057
31
Table IA.5:The probability of increasing, decreasing, and not changing a positionFollowing the regression approach of Table 5, this table explores the probability of increasing, decreasing ornot changing a short position in the bin just below the publication threshold relativ to other reporting bins.The results of Column 1 coincide with those of Table IA.3, Panel A, Column 5. The estimated coefficientsare scaled to reflect changes in percentage points. Standard errors are clustered at the investor-stock andtime level. The t-statistics are given in parentheses, and *, **, and *** indicate significance at the 10%,5%, and 1% levels, respectively.
(1) (2) (3)Probability of
increase no change decrease
Just below threshold – 0.2 bin -0.66*** 0.44 0.22(-3.34) (1.33) (1.06)
Just below threshold – 0.3 bin -1.08*** 1.51*** -0.44**(-5.40) (4.85) (-2.15)
Just below threshold – 0.5 bin -0.92*** 2.42*** -1.50***(-2.89) (4.82) (-4.54)
Just below threshold – 0.6 bin -1.76*** 3.21*** -1.45***(-4.69) (5.70) (-3.51)
Just below threshold – 0.7 bin -1.45*** 3.06*** -1.62***(-3.98) (4.80) (-3.65)
Just below threshold – 0.8 bin -2.15*** 4.31*** -2.16***(-4.30) (5.35) (-4.14)
Just below threshold – 0.9 bin -2.02*** 3.29*** -1.27**(-4.18) (4.14) (-2.40)
Just below threshold – (≥ 1.0 bin) -2.68*** 4.38*** -1.70***(-7.40) (6.92) (-3.94)
Fixed effects:Stock × time Yes Yes YesInvestor × time Yes Yes Yes
R2 (in %) 38.09 41.63 40.86Within R2 (in %) 0.15 0.24 0.11Number of observations 202,057 202,057 202,057
32
Table
IA.6
:D
ura
tion
wit
hin
rep
ort
ing
bin
s:Is
there
hesi
tance
befo
recr
oss
ing
the
dis
closu
rebin
?T
his
table
rep
eats
the
dura
tion
anal
ysi
sof
Tab
le4
condit
ionin
gon
pos
itio
ns
that
even
tual
lyin
crea
seon
the
nex
tch
ange
.T
he
table
rep
orts
the
mea
nan
dm
edia
nnum
ber
oftr
adin
gday
ssp
ent
inea
chdis
clos
ure
bin
(Pan
els
Aan
dB
,re
spec
tive
ly).
Inad
dit
ion,
itdis
pla
ys
the
diff
eren
cein
mea
n(m
edia
n)
du
rati
onof
row
(3),
the
0.4
rep
orti
ng
bin
just
bel
owth
epub
lica
tion
thre
shol
d(s
had
edin
gray
),re
lati
veto
row
(j),
and
thep-v
alue
for
diff
eren
ces
inm
ean
s(m
edia
ns)
.*,
**,
and
***
ind
icat
esi
gnifi
cance
at
the
10%
,5%
,and
1%
level
s,re
spec
tivel
y.
Panel
A:
Mean
dura
tion
wit
hin
rep
ort
ing
bin
intr
ad
ing
days
Posi
tion
sP
osi
tions
Ove
rall
at
thei
rre
cord
hig
hb
elow
thei
rre
cord
hig
h
Mea
nD
iffer
ence
Mea
nD
iffer
ence
Mea
nD
iffer
ence
Bin
dura
tion
row
(3)
-(j
)p-v
alu
edu
rati
on
row
(3)
-(j
)p-v
alu
edu
rati
on
row
(3)
-(j
)p-v
alu
e
undis-
closed
(1)
0.2
13.9
−0.
2(0
.883)
16.7
−0.
3(0
.862)
11.4
−1.
2(0
.339)
(2)
0.3
12.5
1.2
(0.2
88)
14.5
1.9
(0.3
00)
10.3
−0.
2(0
.901)
(3)
0.4
13.8
16.4
10.1
disclosed
(4)
0.5
11.8
1.9
(0.2
09)
13.1
3.3
(0.1
52)
9.7
0.5
(0.7
69)
(5)
0.6
10.0
3.8**
(0.0
24)
10.8
5.7**
(0.0
34)
8.9
1.2
(0.4
39)
(6)
0.7
10.9
2.9
(0.1
34)
10.9
5.5*
(0.0
59)
10.9
−0.
7(0
.708)
(7)
0.8
8.8
4.9**
(0.0
26)
9.2
7.2**
(0.0
25)
7.9
2.3
(0.3
54)
(8)
0.9
10.9
2.9
(0.2
42)
9.0
7.5**
(0.0
33)
15.0
−4.
9(0
.121)
(9)≥
1.0
9.6
4.1***
(0.0
01)
10.0
6.4***
(0.0
00)
8.7
1.5
(0.3
24)
33
Table
IA.6
–C
onti
nued
Panel
B:
Media
ndura
tion
wit
hin
rep
ort
ing
bin
intr
adin
gdays
Posi
tion
sP
osi
tions
Ove
rall
at
thei
rre
cord
hig
hb
elow
thei
rre
cord
hig
h
Med
ian
Diff
eren
ceM
edia
nD
iffer
ence
Med
ian
Diff
eren
ceB
indu
rati
onro
w(3
)-
(j)
p-v
alu
edu
rati
on
row
(3)
-(j
)p-v
alu
ed
ura
tion
row
(3)
-(j
)p-v
alu
e
undis-
closed
(1)
0.2
6−
1(0
.467)
7−
1(0
.244)
5−
1(0
.775)
(2)
0.3
50
(0.7
89)
60
(0.9
93)
40
(0.5
29)
(3)
0.4
56
4
disclosed
(4)
0.5
50
(0.8
61)
60
(0.9
21)
40
(0.7
38)
(5)
0.6
41**
(0.0
49)
33***
(0.0
06)
40
(0.9
97)
(6)
0.7
32
(0.1
15)
33*
(0.0
70)
40
(0.5
02)
(7)
0.8
41**
(0.0
12)
42**
(0.0
23)
31*
(0.0
53)
(8)
0.9
41*
(0.0
74)
42**
(0.0
19)
5−
1(0
.912)
(9)≥
1.0
32***
(0.0
01)
42***
(0.0
00)
31
(0.5
30)
34
Table
IA.7
:A
vera
ge
posi
tion
change
giv
en
that
ap
osi
tion
incr
ease
occ
urs
This
table
show
sth
eav
erage
posi
tion
changes
,giventhatapositionincrease
occurred
,acr
oss
diff
eren
tre
port
ing
bin
sfr
om
whic
hth
ep
osi
tion
was
incr
ease
d(s
tart
ing
bin
).In
addit
ion,
itd
ispla
ys
the
diff
eren
ces
inm
eans
ofro
w(3
),th
e0.
4re
por
tin
gb
inju
stb
elow
the
pu
blica
tion
thre
shol
d(s
had
edin
gra
y),
rela
tive
toro
w(j
),and
thep-v
alu
efo
rth
ediff
eren
ces
inm
eans.
The
table
dis
pla
ys
posi
tion
changes
for
the
over
all
sam
ple
and
for
two
subsa
mple
s,in
whic
hw
esp
lit
the
sam
ple
into
shor
tp
osit
ions
atth
eir
reco
rdhig
han
din
top
osit
ions
bel
owth
eir
reco
rdhig
h.
*,**
,an
d**
*in
dic
ate
sign
ifica
nce
atth
e10
%,
5%,
and
1%le
vels
,re
spec
tivel
y.
Posi
tion
sP
osi
tions
Ove
rall
at
thei
rre
cord
hig
hb
elow
thei
rre
cord
hig
h
Mea
nM
ean
posi
tion
Mea
nSta
rtin
gp
osit
ion
Diff
eren
cep
osi
tion
Diff
eren
cep
osi
tion
Diff
eren
cebin
chan
gero
w(3
)-
(j)
p-v
alu
ech
ange
row
(3)
-(j
)p-v
alu
ech
ange
row
(3)
-(j
)p-v
alu
e
undis-
closed
0.2
(1)
0.11
10.0
03
(0.3
94)
0.1
09
0.007
(0.1
47)
0.1
12−
0.001
(0.8
30)
0.3
(2)
0.11
10.0
03
(0.3
78)
0.1
12
0.005
(0.3
31)
0.1
10
0.001
(0.8
82)
0.4
(3)
0.11
40.1
17
0.1
11
disclosed
0.5
(4)
0.12
1−
0.007*
(0.0
88)
0.1
16
0.001
(0.8
94)
0.1
30−
0.019***
(0.0
05)
0.6
(5)
0.12
3−
0.009**
(0.0
37)
0.1
21−
0.004
(0.5
03)
0.1
27−
0.016**
(0.0
11)
0.7
(6)
0.12
2−
0.007
(0.1
33)
0.1
24−
0.008
(0.2
75)
0.1
17−
0.006
(0.3
06)
0.8
(7)
0.12
6−
0.012**
(0.0
18)
0.1
23−
0.006
(0.3
60)
0.1
35−
0.024***
(0.0
04)
0.9
(8)
0.13
1−
0.017***
(0.0
04)
0.1
21−
0.004
(0.5
88)
0.1
56−
0.045***
(0.0
00)
≥1.
0(9
)0.
119−
0.005
(0.3
68)
0.1
13
0.003
(0.6
64)
0.1
31−
0.020**
(0.0
23)
35
Table
IA.8
:A
ddit
ional
sum
mary
stati
stic
sfo
rp
osi
tion
changes
giv
en
that
ap
osi
tion
incr
ease
occ
urs
This
table
show
sri
ght-
hand
per
centi
les
of
posi
tion
changes
,giventhatapositionincrease
occurred
,acr
oss
diff
eren
tre
port
ing
bin
sfr
om
whic
hth
ep
osit
ion
was
incr
ease
d(s
tart
ing
bin
).T
he
table
dis
pla
ys
pos
itio
nch
ange
sac
ross
the
diff
eren
tp
erce
nti
les
for
the
over
all
sam
ple
and
for
two
subsa
mple
s,in
wh
ich
we
split
the
sam
ple
into
shor
tp
osit
ions
at
thei
rre
cord
hig
hand
into
posi
tions
bel
owth
eir
reco
rdh
igh.
Posi
tion
sP
osi
tions
Ove
rall
at
thei
rre
cord
hig
hb
elow
thei
rre
cord
hig
h
Sta
rtin
gP
erce
nti
les
Per
centi
les
Per
centi
les
bin
50th
75th
90th
95th
50th
75th
90th
95th
50th
75th
90th
95th
undis-
closed
0.2
(1)
0.1
0.1
0.1
0.2
0.1
0.1
0.1
0.1
0.1
0.1
0.1
0.2
0.3
(2)
0.1
0.1
0.1
0.2
0.1
0.1
0.1
0.2
0.1
0.1
0.1
0.1
0.4
(3)
0.1
0.1
0.1
0.2
0.1
0.1
0.2
0.2
0.1
0.1
0.1
0.2
disclosed
0.5
(4)
0.1
0.1
0.2
0.2
0.1
0.1
0.2
0.2
0.1
0.1
0.2
0.4
0.6
(5)
0.1
0.1
0.2
0.2
0.1
0.1
0.2
0.2
0.1
0.1
0.2
0.3
0.7
(6)
0.1
0.1
0.2
0.2
0.1
0.1
0.2
0.2
0.1
0.1
0.1
0.3
0.8
(7)
0.1
0.1
0.2
0.2
0.1
0.1
0.2
0.2
0.1
0.1
0.2
0.2
0.9
(8)
0.1
0.1
0.2
0.3
0.1
0.1
0.2
0.2
0.1
0.2
0.3
0.3
≥1.
0(9
)0.
10.1
0.2
0.2
0.1
0.1
0.2
0.2
0.1
0.1
0.2
0.3
36
Table IA.9:Robustness tests: Calendar-time portfolio approachThis table shows the performance of stocks for which there is at least one secretive short seller just belowthe publication threshold. The analysis is the same as in Table 8 with different sample specifications. InPanel A, we report the baseline results of Panel A of Table 8. In Panel B, we weight the stocks equallyrather than by the number positions that are secretive and just below the publication threshold. In PanelC, we report results when we exclude penny stocks with a price below 1 EUR. In Panel D, we exclude astock from the portfolio if a secretive short seller stays in the bin just below the publication threshold forless than five days. In Panel E, we use European asset pricing factors instead of local. To account for riskfactors, we regress the portfolio excess return on the market excess return (MKTRF), the size (SMB) andvalue (HML) factors, and the momentum factor (WML). For brevity, the table reports only the alphas(in bps per day) of the time-series regression. The t-statistics are computed with Newey-West standarderrors and are shown in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels,respectively. Returns are daily and the sample period is November 5, 2012 to March 31, 2015.
(1) (2) (3)
CAPM Fama-French Carhart
Panel A: Baseline result from Table 8
Alpha -5.46** -6.13*** -5.66**(-2.00) (-2.83) (-2.57)
Panel B: Equal-weighted portfoliojust below the threshold
Alpha -4.82** -5.42*** -5.17***(-1.98) (-3.02) (-3.01)
Panel C: Stocks above 1 EUR
Alpha -5.35* -6.01*** -5.55**(-1.96) (-2.77) (-2.52)
Panel D: At least 5 days just below threshold
Alpha -5.32** -5.97*** -5.58***(-2.06) (-2.99) (-2.72)
Panel E: European pricing factors
Alpha -4.76 -5.45* -5.81*(-1.57) (-1.87) (-1.95)
37
References
Degeorge, F., Patel, J. and Zeckhauser, R. (1999), ‘Earnings management to exceedthresholds’, The Journal of Business 72(1), 1–33.
Griffin, J. M. (2002), ‘Are the Fama and French factors global or country specific?’,Review of Financial Studies 15(3), 783–803.
Nagel, S. (2005), ‘Short sales, institutional investors and the cross-section of stockreturns’, Journal of Financial Economics 78(2), 277–309.
38