Informed short selling around SEO announcements

Informed short selling around SEO announcements

    Informed Short Selling around SEO Announcements Sanjay Deshmukh, Keith Jacks Gamble, Keith M. Howe PII: DOI: Reference: S0929-1199(1...

432KB Sizes 0 Downloads 13 Views

    Informed Short Selling around SEO Announcements Sanjay Deshmukh, Keith Jacks Gamble, Keith M. Howe PII: DOI: Reference:

S0929-1199(16)30148-1 doi:10.1016/j.jcorpfin.2017.05.013 CORFIN 1207

To appear in:

Journal of Corporate Finance

Received date: Revised date: Accepted date:

5 October 2016 12 May 2017 24 May 2017

Please cite this article as: Deshmukh, Sanjay, Gamble, Keith Jacks, Howe, Keith M., Informed Short Selling around SEO Announcements, Journal of Corporate Finance (2017), doi:10.1016/j.jcorpfin.2017.05.013

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

T

Informed Short Selling around SEO Announcements

NU

SC

RI P

Sanjay Deshmukh, Keith Jacks Gamble, and Keith M. Howe ∗

Abstract

AC CE

PT

ED

MA

While much of the prior research on short selling around announcements of seasoned equity offerings (SEOs) has focused on manipulation, it is unclear whether there is also informed short selling around these announcements. We test for informed short selling around SEO announcements by examining the relation between i) pre-announcement short selling and the announcement-period return and ii) changes in short interest around the SEO announcement and the long-term operating and stock price performance following the equity issue. We find that firms with large increases in short interest prior to the SEO announcement exhibit lower (i.e., more negative) announcement-period returns, and that firms with large increases in short interest around the SEO announcement experience inferior long-term operating and stock price performance following the equity issue. We also find that the negative relation between large increases in short interest and long-term operating and stock price performance is more pronounced among shelf offers. This result highlights the informational role of short sellers in identifying opportunistic market timers of equity issues among shelf filers. Our overall results indicate the presence of informed short selling around SEO announcements.

*Sanjay Deshmukh and Keith Howe (Emeritus) are from the Department of Finance, Driehaus College of Business, DePaul University, 1 East Jackson Blvd., Chicago, IL 60604. Keith Gamble is from the Department of Economics and Finance, Middle Tennessee State University, Business and Aerospace N329C, MTSU Box 27, Murfreesboro, TN 37132. For correspondence, please contact Sanjay Deshmukh at 312.362.8472 or [email protected] We thank Charles Jones and Wei Xu for providing us with the data on short interest for NASDAQ firms. We also thank Jeffry Netter, the editor, Charles Jones, and an anonymous referee for valuable comments and suggestions.

ACCEPTED MANUSCRIPT

T

Informed Short Selling around SEO Announcements

RI P

1. Introduction

The literature suggests that short selling around announcements of seasoned equity of-

SC

ferings (SEOs) is likely to be motivated by either manipulation or negative information. For instance, an SEO announcement can provide short sellers with an opportunity to proÞt from

NU

share-price manipulation. Since the o er price of an SEO is typically based on the closing stock price before the o er date, relentless pre-o er manipulative short selling can depress

MA

the Þrm’s stock price, forcing the Þrm to issue the new shares at a discount (SaÞeddine and Wilhelm, 1996). These short sellers can then proÞt by covering their short position at

ED

the lower o er price.1 In contrast, informed investors are likely to engage in short selling when they uncover negative information about the company around the SEO announcement.

PT

These informed short sellers then expect to proÞt from the eventual decline in the stock price that should occur when the market learns of the negative information. Therefore, manip-

AC CE

ulative short selling is carried out to a ect prices adversely while informed short selling is carried out in anticipation of a decline in the stock price. In this paper, we test for informed short selling around announcements of SEOs. Most of the prior empirical research on short selling around SEO announcements focuses on the manipulation rationale. This research draws on the model in Gerard and Nanda (1993), who argue that manipulative trading before an SEO can distort prices and worsen the winner’s curse problem, causing the issue discount to rise. To uncover potential manipulative trading, SaÞeddine and Wilhelm (1996) and Singal and Xu (2005) examine the direct relation between short selling and the SEO issue discount. In contrast, Corwin (2003), Kim and Shin (2004), and Autore (2010) adopt an indirect approach by examining the e ect of regulatory 1

The Securities and Exchange Commission (SEC) adopted Rule 10b-21 in August 1988, which prohibited short sellers from covering their short positions using shares purchased in the SEO. In April 1997, the SEC softened Rule 10b-21 by replacing it with Rule 105, which prohibits short sellers from covering their short positions using shares purchased in the SEO if the shares were shorted within Þve days of the o er date. Much of the prior empirical research on the manipulation motive uses data after Rule 10b-21 became e ective.

1

ACCEPTED MANUSCRIPT

changes on the relation between short selling and the issue discount. In general, these studies do not Þnd evidence supportive of manipulative trading. But, they do not explicitly consider

RI P

T

nor test for evidence of informed short selling as an alternative explanation.2 In a recent study, Henry and Koski (2010) explore the manipulation rationale using daily short-selling data over a two-year period. They examine the relation between short-selling

SC

activity and stock returns around the SEO-announcement date, along with the issue discount, and argue that their evidence provides strong support for the manipulation rationale while

NU

ruling out the possibility of informed short selling.3 However, recent evidence on the superior information-processing ability of short sellers suggests that the evidence in Henry and Koski

MA

(2010) could be consistent with informed short selling. If short sellers target Þrms that are overvalued, then the larger issue-price discounts, associated with higher short selling, could

ED

actually reßect worsening Þrm prospects. These worsening Þrm fundamentals, as opposed to the pre-issue short selling, could cause the issue-price discount.

PT

In sum, the overall evidence on informed short selling around SEO announcements is scant and the research that does exist is inconclusive. Nonetheless, it is important to de-

AC CE

termine whether there is informed short selling around an SEO announcement because of its implications for regulation, market e!ciency, and price discovery. For example, if short selling is driven by worsening Þrm fundamentals, then it provides useful information to other market participants in the capital allocation process and would represent a positive externality. Consequently, any regulation that hampers short selling around SEOs will make prices less e!cient with negative implications for equity issues and for broader resource allocation. We adopt a three-pronged empirical approach to detect informed short selling around an SEO announcement. We merge two di erent strands of literature to formulate and test three 2 Kim and Shin (2004) note the possibility of informed short selling based on short-term price changes after the SEO but make no explicit attempt to test for its presence. 3 Henry and Koski (2010) also argue that Rule 105 has not been e ective in curbing manipulative short selling. They Þnd that increased short selling is associated with bigger issue-price discounts, both for short selling between 6 and 10 days prior to the equity issue (not covered by Rule 105) and for short selling between 1 and 5 days prior to the issue (covered by Rule 105). In addition, they Þnd no evidence of increased short selling prior to the SEO announcement or of pre-issue short selling activity predicting returns in the week following the issue.

2

ACCEPTED MANUSCRIPT

hypotheses. The Þrst strand of the literature relates to the e ect of the SEO announcement on the Þrm’s stock price, on the operating performance following the equity issue, and on the

RI P

T

long-run stock performance after the equity issue. The second strand of the literature relates to the superior information-processing ability of short sellers, which suggests the possibility of informed short selling around SEO announcements.

SC

We draw on the two strands of literature noted above to formulate three hypotheses. Our Þrst hypothesis states that the larger the increase in short interest prior to the SEO

NU

announcement, the larger the price decrease at the announcement. Our second hypothesis states that the greater the increase in a Þrm’s short interest around the SEO announcement,

MA

the larger the decline in its subsequent operating performance. The third hypothesis states that the larger the increase in a Þrm’s short interest around the SEO announcement, the

ED

lower its long-term stock return.

We use a standard event-study method to test the Þrst hypothesis. To test the second,

PT

we follow the method in Barber and Lyon (1996) for ex-post event studies that use an accounting-based measure of operating performance. The most important feature of this

AC CE

method is matching event Þrms to control Þrms with similar pre-event operating performance, along with size and industry, to ensure that test statistics are properly speciÞed. To test the third hypothesis, we follow the method in Barber and Lyon (1997) for ex-post event studies that examine cumulative stock returns for up to Þve years after the event. Here, abnormal stock returns for an event Þrm are calculated relative to the return of a control Þrm that is matched to the event Þrm in terms of both size and the book-to-market ratio. Our objective is to test whether short sellers, when trading around SEO announcements, are motivated by negative information about the future prospects of the company. To detect potential informed short selling, we examine the relation between increases in short interest around the SEO announcement date and i) Þrm operating performance after the equity issue and ii) abnormal stock returns after the equity issue. If short selling around SEO announcements is completely manipulative so as to increase the issue discount, then we should not Þnd any relation between increases in short interest and subsequent long-term 3

ACCEPTED MANUSCRIPT

operating and stock price performance. In addition, we examine the relation between the increase in short interest prior to the SEO announcement date and announcement-period

RI P

T

returns. Changes in short interest prior to the SEO announcement are much more likely to be motivated by information, making the manipulation concern irrelevant for this test. Therefore, if we Þnd a negative relation between an increase in short interest and the various

SC

performance measures, then our empirical results would indicate the presence of informed short selling. Note that we do not dismiss nor test for manipulative short selling, which

NU

possibly coexists with informed short selling. However, if short selling is purely manipulative as argued in Henry and Koski (2010), then our tests will produce results that are both

MA

statistically and economically nonsigniÞcant.

We calculate short interest as the ratio of the number of shares sold short to the number of

ED

shares outstanding. In our empirical analysis, we focus on changes in the short-interest level that are in the top quartile (i.e., 75th percentile and above) of changes in short interest for

PT

the sample of SEO-announcing Þrms, and term these top-quartile Þrms as "event Þrms." Our overall results, however, remain qualitatively the same when we consider various percentile

AC CE

cuto s of changes in short interest to identify the event Þrms. We Þnd a negative relation between the change in short interest prior to the SEO announcement and the announcement-period return. SpeciÞcally, Þrms in the top quartile of increases in short interest experience an announcement-period return that is about one percentage point lower (i.e., more negative) than that for the rest of the sample Þrms. This di erence of one percentage point is about 50% of the magnitude of the announcementperiod return of about -2% for the overall sample of SEOs. We also Þnd that Þrms in the top quartile of short-interest increases around the SEO announcement experience a signiÞcant decline in operating performance over the two-year period following the equity o ering. The decline in cumulative operating performance, calculated relative to a matched-control sample, is highly economically signiÞcant and approximately 3.44 percentage points lower than the corresponding decline in relative operating performance for the rest of the sample of SEO-announcing Þrms (i.e., the non-event Þrms). This Þnding is largely robust to alternative 4

ACCEPTED MANUSCRIPT

measures of operating performance. Similarly, over the six-month period following the equity o ering, Þrms in the top quartile of short-interest increases around the SEO announcement

RI P

T

experience a cumulative abnormal stock return that is 4.27 percentage points lower than the cumulative abnormal stock return for the rest of the sample of SEO-announcing Þrms. Next, we investigate the relation between short selling and the three measures of perfor-

SC

mance separately for shelf-registrations of SEOs (shelf o ers) and for traditional Þlings of SEOs (non-shelf o ers). We do so in light of recent research documenting a revival of shelf

NU

registrations since the 1990s, as well as the recent dominance of shelf Þlings over traditional Þlings (Heron and Lie, 2004, and Autore, Kumar, and Shome, 2008). A shelf registration

MA

provides the issuing Þrm with the option to issue equity at any time during the three-year period subsequent to the Þling. Therefore, Þrms Þling for shelf registrations enjoy greater

ED

ßexibility in terms of both timing and deferring the equity issue relative to traditional Þlers of the SEO. The greater ßexibility is valuable to those Þrms that need to Þnance proÞtable

PT

investments and can thus time their issues accordingly. However, this same ßexibility provides a Þrm with an opportunity to issue equity when it is overvalued. Identifying such

AC CE

opportunistic timers is important from a resource-allocation perspective. If short sellers possess a superior ability to identify Þrms that are overvalued, then large increases in short interest can help identify potential opportunistic market timers. With a few exceptions, our results from the three tests for shelf o ers and non-shelf o ers mirror those we Þnd for the overall sample of SEOs. We also Þnd that, among Þrms in the top quartile of short-interest increases (i.e., the event Þrms), shelf o ers experience inferior operating and long-run stock price performance relative to non-shelf o ers. In contrast, among Þrms in the bottom three quartiles of short-interest increases (i.e., the non-event Þrms), the di erence in operating and long-run stock price performance between shelf o ers and non-shelf o ers is not as pronounced. The inferior long-term performance of shelf Þlers, among Þrms in the top quartile of short-interest increases, suggests that the potential for opportunistic timing of equity issues is greater among shelf Þlers. The Þndings also suggest that short sellers are able to identify overvalued Þrms among shelf Þlers and highlight the 5

ACCEPTED MANUSCRIPT

potential informational role of short sellers in identifying opportunistic market timers. Our overall evidence indicates a negative relation between increases in short selling and

RI P

T

i) announcement-period returns and ii) long-run operating and stock-price performance. In general, these Þndings are highly economically signiÞcant and provide strong support for our three hypotheses and for the presence of informed short selling around SEO announcements.

SC

We provide three contributions. First, much of the prior work on short selling around the SEO announcement has focused on the manipulation rationale while the evidence on

NU

informed short selling remains inconclusive. We Þll this gap by performing three separate tests to check for the presence of informed short selling. We provide economically signiÞcant,

MA

comprehensive, and unambiguous evidence of informed short-selling around SEOs. Our results indicate that a meaningful portion of short selling around SEO announcements is

ED

information-driven and is thus not all manipulative. With the SEC currently investigating short selling around SEOs, our evidence provides timely and important guidance.

PT

Second, shelf registrations now represent the dominant form of equity Þnancing in the secondary market. One interesting implication of shelf registration is that it provides Þrms

AC CE

with an opportunity to issue equity when their stock is overvalued. Our results suggest that the extent of short selling around the announcement of shelf registrations can be used to identify potential opportunistic market timers. This novel result should serve as an important input in the resource-allocation decision of investors. Third, there is no explicit evidence in the literature on the portion of manipulative versus informed short selling in explaining the announcement-period return for SEOs. This is understandable given the nature of the data available on short interest. We Þll this gap in the literature and present a novel result by applying the evidence in Henry and Koski (2010) to our data and Þndings. SpeciÞcally, we provide estimates and bounds on the portion of manipulative versus informed short selling in explaining the announcement-period return for SEOs. These bounds also suggest that informed short selling around announcements of SEOs is material. The paper proceeds as follows: In Section 2, we derive testable hypotheses and describe 6

ACCEPTED MANUSCRIPT

our data and the method we use. In Section 3, we provide tests of our hypotheses along with a discussion of the results. We conclude in Section 4 with a summary and an interpretation

SC

2. Hypothesis, Data, and Method

RI P

T

of the central Þnding.

Hypothesis Development: As noted earlier, we draw on two strands of literature to

NU

formulate our testable hypotheses. The Þrst strand relates to research on the SEO announcement and its relation to the (short-term) announcement-period return and to long-term op-

MA

erating and stock-price performance. There is a large body of evidence that documents that announcements of equity issues are associated with a decrease in the stock price (Asquith and Mullins, 1986; Masulis and Korwar, 1986; Mikkelson and Partch, 1985; and Ritter, 2003).

ED

This evidence is consistent with the adverse selection model in Myers and Majluf (1984).4

PT

With respect to long-term e ects associated with the SEO, studies focus on either the operating performance or the stock return subsequent to the o ering. Loughran and Ritter (1997)

AC CE

and Heron and Lie (2004) Þnd that Þrms, selling shares in an SEO, experience a decline in subsequent operating performance. Other studies document a decline in the long-run stock price subsequent to the SEO (Loughran and Ritter, 1995; Spiess and A"eck-Graves, 1995; and Brav, Geczy, and Gompers, 2000). The second strand of literature that we draw on relates to the information-processing ability of short sellers. There is a notable body of empirical research that indicates that some short sellers appear to be informed investors. For example, Christophe, Ferri, and Angel (2004) show that abnormal short selling before earnings announcements is signiÞcantly related to post-announcement stock returns. Boehmer, Jones, and Zhang (2008) Þnd that heavily-shorted stocks underperform lightly-shorted stocks and argue that short sellers appear to be well-informed. Christophe, Ferri, and Hsieh (2010) document abnormal levels of

4

If managers know more about the Þrm’s prospects than do outside investors, then investors interpret equity issues as a signal that the stock is overpriced, causing the price to decline at the SEO announcement.

7

ACCEPTED MANUSCRIPT

short selling prior to public announcements of stock downgrades by analysts. Karpo and Lou (2010) Þnd that short sellers appear to anticipate the revelation and severity of Þnancial

RI P

T

misconduct by Þrms. Boehmer and Wu (2013) show that more active short selling results in more accurate stock prices and that greater shorting ßow accelerates the incorporation of public information into stock prices. Engelberg, Reed, and Ringgenberg (2012) exam-

SC

ine short sales around 39 corporate events and argue that short sellers exhibit a superior ability to process publicly available information and trade proÞtably around news events.

NU

Deshmukh, Gamble, and Howe (2015) examine whether short sellers can identify Þrms with deteriorating fundamentals and document that Þrms with large increases in short interest

MA

experience a subsequent decline in operating performance. Their evidence suggests that short sellers trade, at least partly, based on declining Þrm fundamentals and that changes in

ED

short interest provide useful information to market participants. The above empirical evidence is consistent with the implications of the asymmetric in-

PT

formation model in Diamond and Verrecchia (1987), who develop a rational expectations model of the e ect of short-sale constraints on the distribution of prices and their speed

AC CE

of adjustment to private information. One important prediction of their model is that, in the presence of restrictions on short selling, an unexpected increase in the announced short interest in a stock represents bad news and should result in a price adjustment. This model also provides the theoretical underpinnings for our study. We combine the above two strands of literature and draw on their implications to propose the following testable hypotheses: Hypothesis 1. The larger the unexpected increase in a Þrm’s short interest before the SEO announcement, the greater the price decline at the announcement of an SEO. We also examine the relation between the increase in short interest and both the long-term operating performance and the long-term stock return. A Þrm’s operating performance serves as a direct measure of Þrm fundamentals and a Þrm’s stock price should change in response to changes in Þrm fundamentals. The stock price decline must result from deteriorating

8

ACCEPTED MANUSCRIPT

fundamentals (i.e., declining operating performance) if short selling is information-based and partly reßects worsening Þrm fundamentals. In addition, worsening Þrm fundamentals

RI P

T

over time should be associated with a decline in the Þrm’s stock price. Combining this reasoning with the evidence that the SEO announcement, on average, is followed by a decline in both the long-term operating performance and the long-term stock return, we propose

SC

the following two hypotheses:

NU

Hypothesis 2. The larger the unexpected increase in a Þrm’s short interest around the SEO announcement, the greater the decline in its operating performance following the equity o ering.

MA

Hypothesis 3. The larger the unexpected increase in a Þrm’s short interest around the SEO announcement, the lower the long-run stock return following the equity o ering.

ED

It is important to note that the above three hypotheses apply to informed short selling only. If short selling were completely manipulative, then there should be no relation between

PT

changes in short interest and our three measures of performance. The reason is that manipulative short selling should be carried out to a ect the stock price immediately. In contrast,

AC CE

informed short selling should be carried out in anticipation of a future decline in the stock price when the market learns about the negative information. Our testable hypotheses are based on unexpected changes in short interest. The existing literature does not provide guidance to calculate expected short interest in a stock. We adopt a naïve expectations model to calculate expected short interest where the level of short interest in period t equals the level of short interest in period t-1 (Deshmukh, Gamble, and Howe, 2015). We thus treat any increase in short interest as an unexpected increase. Any mismeasurement of expected short interest, and consequently the unexpected increase, only serves to bias the magnitude of the measured relation (between changes in short interest and subsequent changes in operating performance and stock price) toward zero. However, any mismeasurement of expected short interest should not a ect the direction (i.e., the sign) of the relation between changes in short interest and changes in both operating performance and the stock price (Diamond and Verrecchia, 1987). Therefore, to the extent we measure 9

ACCEPTED MANUSCRIPT

unexpected changes in short interest with error, we introduce noise in this measure thereby undermining our ability to identify a relation between short interest changes and subsequent

RI P

T

operating performance and stock price changes.

To identify a meaningful component of informed short selling, we focus on increases in short interest that are in the top quartile (i.e., the 75th percentile) of increases in short

SC

interest reported by the stock exchanges. We use our judgment and adopt the 75th percentile threshold, given the size and composition of our sample of SEOs. In our view, this threshold

NU

should reasonably reduce the noise in our measure of short selling while ensuring that we have adequate statistical power for our tests. However, for robustness, we perform our tests

MA

using various percentile cuto s of increases in the short-interest level to identify event Þrms. The noise in our measure of changes in short interest is likely to arise because changes

ED

in short interest may also be motivated by factors unrelated to short sellers’ information about deteriorating Þrm fundamentals. For example, market makers may need to engage in

PT

short sales to provide liquidity in the marketplace. Short sales may also arise when traders wish to hedge their positions such as a bondholder seeking to hedge the equity feature of a

AC CE

convertible bond. The presence of this noise in our measure of short selling only introduces an attenuation bias against Þnding support for our hypotheses. In sum, changes in short interest, if motivated by information about Þrm fundamentals, should be related to subsequent changes in operating performance and the long-run stock return. On the other hand, if the changes in short interest were motivated by factors unrelated to Þrm fundamentals such as manipulation, hedging, market-making, or exploitation of pricing anomalies, then there should be no relation between changes in short interest and both the subsequent operating performance changes and the long-run stock return. Our empirical design thus provides a clean test of our hypotheses and should allow us to detect informed short selling if it is present in the data. Data: Our initial sample of Þrms consists of all follow-on equity o erings for US stocks from the SDC database over the period 1988-2011. The data include the Þling date, the issue date, o er price, number of primary shares issued and secondary shares sold, and an 10

ACCEPTED MANUSCRIPT

indicator for shelf o erings. We follow the literature on SEOs and exclude both Þnancials (SIC 6000-6999) and utilities (SIC 4900-4999) given their regulatory status. Our sample

RI P

T

includes primary o erings, primary plus secondary o erings, and some secondary o erings of shares. In addition, we include o erings with no identiÞcation regarding the type of o ering. However, excluding these o erings from our empirical analysis has no material

SC

impact on any of our results.

Next, we match this sample to monthly short-interest data on all stocks on the NYSE,

NU

Amex, and Nasdaq over the same period. The Compustat Short Interest Þle contains data for NYSE and Amex Þrms beginning in 1973 and for NASDAQ Þrms beginning in 2003. We

MA

have short interest data for NASDAQ Þrms over 1988-2002 and we supplement these data with those from Compustat over the period 2003-2011.5 We begin our sample period in 1988

ED

because this is the Þrst year for which we have short interest data for Þrms from all three major U.S. stock exchanges, which allows us to maximize our sample size while including a

Þrms.

PT

broader cross-section of Þrms. Our Þnal sample consists of 7,129 equity o erings by 3,584

AC CE

We calculate the level of short interest in a stock as the ratio of the number of shares sold short to the number of shares outstanding. For our announcement-period return tests, we calculate the change in short interest as the change in the short-interest level over the three-month period that ends immediately before the SEO announcement date. Since we use short-interest data before the SEO announcement, we do not assume that short sellers anticipate the announcement. Instead, we assume that short sellers exhibit an ability to identify overvalued Þrms given their superior information-processing skills and that they use publicly-available information to make that determination. Therefore, consistent with Engelberg, Reed, and Ringgenberg (2012), short sellers are using public information, that is already available, to make the determination of overvaluation. For our operating performance and long-run stock return tests, we calculate the change 5

We thank Charles Jones and Wei Xu for providing us the short-interest data for NASDAQ Þrms over the period 1988—2002.

11

ACCEPTED MANUSCRIPT

in short interest as the most recent change in the short-interest level over a three-month period that immediately precedes the o er date. The median number of days between

T

announcement and o er dates for our sample Þrms is 40. For most Þrms in our sample, we

RI P

end up calculating the change in short interest around the announcement date.6 The ending value that we use to calculate the change in short interest is the level of short interest after

SC

the SEO announcement. Therefore, the change in short interest is based on data after the event is announced and is thus consistent with Engelberg, Reed, and Ringgenberg (2012),

NU

who argue that short sellers trade proÞtably after a news release. Method: For the announcement-period return tests, we use a three-day window around

MA

the SEO-announcement date to measure the cumulative abnormal return (CAR) for our sample Þrms. The three-day window begins the day before the announcement date and

ED

ends the day after the announcement date. The daily abnormal returns over the three-day window are based on a market model estimated using the CRSP value-weighted index. The

PT

estimation period for the parameters of the market model is 255 days and ends 46 trading days before the event date.

AC CE

For the operating performance tests, we follow the method in Barber and Lyon (1996) for ex-post event studies like ours that use measures of operating performance based on accounting data. Barber and Lyon (1996) document that commonly-used research designs that do not follow their method produce test statistics that are misspeciÞed. They recommend matching sample Þrms to control Þrms with similar pre-event performance so as to detect ex-post abnormal operating performance changes in the sample Þrms. This matching procedure also controls for performance changes caused by a mean reversion in the performance measure. We follow their recommendations and match our sample Þrms to control Þrms by industry, Þrm size, and pre-event operating performance as of the Þscal-year end that follows the o ering date. We also require that control Þrms have their Þscal-year end that is within 6

However, for some Þrms (such as those with shelf o ers), the time lag between the announcement date and the o er date is longer than three months. For these Þrms, the calculated change in short interest would be based on data between the announcement and o er dates. For ease of exposition, we use "change in short interest around the announcement date" when discussing the results.

12

ACCEPTED MANUSCRIPT

a one-quarter window around the Þscal-year end of the sample Þrm. We measure Þrm size by the total book value of assets and our industry match is based

RI P

T

on the two-digit SIC code. As in Barber and Lyon (1996), we measure a Þrm’s operating performance as the ratio of operating income before depreciation to total assets and winsorize this measure at the 1st and the 99th percentiles. We begin by matching sample Þrms to

SC

control Þrms within 90% and 110% of pre-event operating performance, within 90% and 110% of size, and within the same industry. When we fail to Þnd a control Þrm using these

NU

criteria, we drop the industry requirement while maintaining the requirements on size and pre-event operating performance. If we are still unable to identify a control Þrm, we relax

MA

our requirements further by widening the size window to within 70% and 130% and the pre-event operating performance window to within 70% and 130%, if required.

ED

We deÞne the change in operating performance for a Þrm i in year t as the di erence in the operating performance in year t and the operating performance in year t-1. We deÞne

PT

abnormal operating performance for a sample Þrm i in year t as the change in the operating performance for the sample Þrm minus the median change in the operating performance

AC CE

among matched control Þrms.

Our measure of abnormal operating performance implicitly controls for many other factors that might a ect a Þrm’s operating performance. We calculate abnormal operating performance as the di erence between changes in operating performance of the sample of SEO-announcing Þrms and the matched control Þrms. First, a change in operating performance for any Þrm sweeps away both Þrm Þxed e ects and industry Þxed e ects. Second, since abnormal operating performance equals the di erence in the change in operating performance between sample Þrms and control Þrms, the measure controls for any factor that a ects all Þrms at a given point in time. Third, the change in operating performance of the matched control Þrm serves as an "expected change" for the sample Þrm and incorporates any change in operating performance that stems from systematic or market factors. Thus, the abnormal operating performance measure is designed to reßect changes in operating performance for the sample Þrms that stem from Þrm-speciÞc factors, which short sellers 13

ACCEPTED MANUSCRIPT

should identify if they are motivated by information about Þrm fundamentals. We use this empirical method, as opposed to a regression-based method, because it allows us to zero

RI P

T

in on changes in operating performance most directly and identify any association between short selling and operating performance following the equity issue.

For the tests based on the long-run stock return, we follow the method in Barber and

SC

Lyon (1997) for ex-post event studies that examine cumulative stock returns for up to Þve years after the event. Barber and Lyon (1997) show that test statistics using a reference

NU

portfolio to measure expected returns are misspeciÞed in long-run stock performance studies. To correct this problem, Barber and Lyon (1997) set the expected return for the sample Þrm

MA

to equal the return for a control Þrm that is matched to the sample Þrm based on both size and the book-to-market ratio. Size equals the Þrm’s market value of equity (price x

ED

shares outstanding) as of the end of June preceding the equity issue. The book-to-market ratio equals the ratio of the book value of equity (as of the Þscal-year end in the preceding

PT

calendar year) to the market value of equity (as of the end of the preceding calendar year). We use this book-to-market ratio to match sample Þrms to control Þrms from July to June of

AC CE

the following calendar year. The six-month lag, when calculating the book-to-market ratio, ensures that all of the information in the ratio is publicly available even if there is a delay in the reporting of Þnancial statements. The Þrm whose book-to-market ratio is closest to that of the sample Þrm is chosen as the control Þrm among those whose market capitalization is between 70% and 130% of that of the sample Þrm. We match each of the sample Þrms to a new control Þrm as of the end of June each year. Given the data requirements and data Þlters, the number of Þrms in the subsamples of equity o erings di ers slightly across the three tests. Table 1 provides key summary statistics for the entire sample of SEOs and for the subsamples used for the three tests.

14

ACCEPTED MANUSCRIPT

3. Empirical Results

T

3.1 All Equity O erings

RI P

Short Selling and Announcement-Period Returns: Hypothesis 1 states that the larger the unexpected increase in a Þrm’s short interest before the SEO announcement, the

SC

greater the price decline at the announcement. We test whether the di erence in the average three-day CAR between the event group and the non-event group is signiÞcantly di erent

NU

from zero by performing a t-test with unequal variances. As indicated earlier, the event group comprises SEO-announcing Þrms in the top quartile (i.e., the 75th percentile and above) of

MA

short-interest increases and the non-event group consists of the rest of the sample of SEOannouncing Þrms (i.e., those in the bottom three quartiles of increases in the short-interest level). For robustness, we perform the non-parametric Wilcoxon rank-sum test to examine

ED

whether the di erence in the median CAR between the event group and the non-event group is signiÞcantly di erent from zero.

PT

The results in Table 2 indicate that, on average, the announcement of an SEO is associ-

AC CE

ated with a stock price decrease. The average CAR over the three-day window is -2.09% and this result, in terms of both sign and magnitude, is consistent with prior evidence (Ritter, 2003). In addition, the event Þrms, with a pre-announcement increase in short interest in the top quartile, experience an average CAR of -2.81% while the non-event Þrms (bottom three quartiles) experience an average CAR of -1.85%. Both of these CARs are statistically signiÞcantly di erent from zero at the 1% level. The di erence of -0.96% between the average CARs of event and non-event Þrms is also statistically signiÞcant at the 1% level. The results from the non-parametric Wilcoxon rank-sum test are consistent with those based on the average CAR. The median event Þrm experiences a CAR of -2.78% while the median non-event Þrm experiences a CAR of -1.83%. The median di erence in the CAR between event and non-event Þrms is signiÞcantly di erent from zero at the 1% level. Thus, the di erence in the stock price response, between Þrms in the top quartile of short-interest increases and those in the bottom three quartiles, is also highly economically signiÞcant. 15

ACCEPTED MANUSCRIPT

The results in Table 2 are consistent with Hypothesis 1. They indicate a negative relation between increases in short interest prior to the SEO announcement and the magnitude of

RI P

T

the stock price response to the announcement. Here, we focus on changes in short interest before the announcement of the SEO. Short sellers, on average, are unlikely to be aware of the impending SEO announcement. Hence, their trading activity is more likely to be

SC

motivated by negative information than by a desire to manipulate the stock price. Therefore, the negative relation that we document between increases in short interest

NU

and the announcement-period return indicates the presence of informed short selling before the SEO announcement. It is possible that short selling before the announcement could be

MA

motivated by factors unrelated to either information or share-price manipulation. However, if that were the case, then there should be no relation between changes in short interest and

ED

the announcement-period return and that is contrary to what we Þnd. We base our analysis on the conjecture that short sellers have a superior ability to analyze

PT

publicly-available information and increase their stake (i.e., the extent of their short selling) in Þrms they believe to be overvalued. Hence, there are some short sellers who make informed

AC CE

trades in these stocks rather than just noise trades. One possible channel for the results in Table 2 is that the SEO announcement prompts investors to examine the announcing Þrms more closely (and likely factor in the pre-announcement short-interest change) before responding to the announcement. Thus, a knowledge of short-selling activity, prior to the SEO announcement, appears to be relevant in explaining the announcement-period returns. Short Selling and Operating Performance: Hypothesis 2 states that the larger the unexpected increase in a Þrm’s short interest around the SEO announcement, the greater the decline in the Þrm’s operating performance following the equity o ering. To test Hypothesis 2, we examine the ex-post operating performance of both event and non-event Þrms relative to that of their respective matched control Þrms. We test whether the di erence in the average cumulative abnormal operating performance between event and non-event Þrms, over the following two-year period, is signiÞcantly di erent from zero by performing a simple t-test with unequal variances. Note that we deÞne abnormal operating performance for a 16

ACCEPTED MANUSCRIPT

sample Þrm i in year t as the change in the operating performance for the sample Þrm minus the median change in the operating performance among matched control Þrms. For

RI P

T

robustness, we perform the non-parametric Wilcoxon rank-sum test to examine whether the di erence in the median cumulative abnormal operating performance, between event and non-event Þrms, is signiÞcantly di erent from zero.

SC

Our results in Panel A of Table 3 indicate that the sample of Þrms, announcing an SEO and issuing equity, experience a decline in their operating performance over the following

NU

two-year period. SpeciÞcally, the average cumulative abnormal operating performance of these Þrms, calculated relative to the control Þrms, equals -2.34% over the following two-year

MA

period. This decline in operating performance is statistically signiÞcant at the 1% level and its magnitude represents about 160% of the mean level of operating performance (of 1.48%)

ED

of the sample Þrms in the Þscal year of the SEO. This Þnding of a decline in operating performance following the equity issue is consistent with both Loughran and Ritter (1997)

PT

and Heron and Lie (2004).

The result of interest is that the di erence in the average cumulative abnormal operating

AC CE

performance between event and non-event Þrms, over the following two-year period, is also signiÞcantly di erent from zero. This di erence of -3.44% between event and non-event Þrms is also economically signiÞcant; it represents about 173% of the mean level of operating performance (of -1.99%) of the event Þrms in the Þscal year of the SEO. In other words, the cumulative change in the operating performance of Þrms in the top quartile of short-interest increases (after netting the performance change of matched control Þrms) is 3.44% lower than the corresponding (net) change in the operating performance of Þrms in the bottom three quartiles of short-interest increases. We report results from the non-parametric Wilcoxon rank-sum test in Panel B of Table 3. The median Þrm, completing an SEO, experiences a decline in operating performance over the following two-year period. The median cumulative abnormal operating performance of these Þrms, calculated relative to the control Þrms, equals -0.28% over the following two-year period. This decline in operating performance is statistically signiÞcantly di erent from zero 17

ACCEPTED MANUSCRIPT

at the 1% level and its magnitude represents about 2.7% of the median level of operating performance (of 10.45%) of the sample Þrms in the Þscal year of the SEO. Over the following

RI P

T

two-year period, the di erence in the median cumulative abnormal operating performance between event and non-event Þrms is also signiÞcantly di erent from zero. This di erence of -1.04% between event and non-event Þrms is also economically signiÞcant; it represents

SC

about 11.6% of the median level of operating performance (of 8.95%) of the event Þrms in the Þscal year of the SEO.

NU

The results in Table 3 are thus consistent with Hypothesis 2. The Þndings on the di erence between event and non-event Þrms indicate that Þrms with a larger increase in short interest

MA

around the SEO announcement experience a larger decline in operating performance after the equity issue. Therefore, increases in short interest around the SEO announcement appear

ED

to be, at least partly, motivated by negative information about Þrm fundamentals. As noted earlier, our empirical method rules out the possibility that the decline in op-

PT

erating performance merely reßects a mean reversion in performance over time. The reason is that our method of choosing matched control Þrms based on pre-event performance is

AC CE

designed explicitly to control for a mean reversion in the measure of operating performance we use (Barber and Lyon, 1996). Therefore, our empirical method is designed to control for a possible mean reversion in performance and is more likely to identify an economic change in performance, if such a change is present in the data. For robustness, we follow Barber and Lyon (1996) and use three other measures of operating performance: ratio of cash ßow to assets; ratio of operating income before depreciation to market value of assets; and ratio of operating income before depreciation to sales.7 As with our primary measure of operating performance, we follow the matching procedure based on Barber and Lyon (1996) and winsorize the data at the 1st and the 99th percentiles, respectively. Even after winsorizing, there were very large outliers for the measure based on 7

Cash ßow equals operating income before depreciation plus the decrease in receivables plus the decrease in inventory plus the increase in accounts payable plus the increase in other current liabilities plus the decrease in other current assets. The market value of assets equals the book value of total assets minus the book value of common equity plus the market value of common equity (i.e., share price x shares outstanding).

18

ACCEPTED MANUSCRIPT

sales (i.e., ratio of operating income before depreciation to sales). To minimize the e ect of outliers on our results, we winsorize the data at the 5th and 95th percentiles for the measure

RI P

T

based on sales.

For these alternative measures, the sample Þrms are closely matched to control Þrms. However, the resulting number of observations for these three measures is much smaller than

SC

that for our primary measure. In addition, we Þnd these measures to be very noisy as they are based on cash ßow, market value, and sales, all of which tend to be more volatile than

NU

either the operating income before depreciation or the book value of assets on which our primary measure is based. Therefore, to increase the signal-to-noise ratio and to reduce the

MA

adverse e ect of the noise in the measures on the relation between changes in short interest and operating performance, we focus on changes in short interest at the 90th percentile, as

ED

opposed to the 75th percentile, to identify our event Þrms. As with our primary measure, we deÞne abnormal operating performance for a sample

PT

Þrm i in year t as the change in the operating performance for the sample Þrm minus the median change in the operating performance among matched control Þrms. We test whether

AC CE

the di erence in the average cumulative abnormal operating performance between event and non-event Þrms, over the following two-year period, is signiÞcantly di erent from zero by performing a simple t-test with unequal variances. We also perform the Wilcoxon rank-sum test to examine whether the di erence in the median cumulative abnormal operating performance between event and non-event Þrms, over the following two-year period, is signiÞcantly di erent from zero. Therefore, we perform a total of six tests - three for the di erence in averages and three for the di erence in medians. Our untabulated results indicate that all of the six values for the di erence in cumulative abnormal operating performance, between event and non-event Þrms, are negative. In addition, Þve out of these six values are statistically signiÞcant (at the 10% level or better) and the sixth one is marginally nonsigniÞcant (p = 0.1010 ).8 These results are thus consistent with Hypothesis 2. 8

Based on Barber and Lyon (1996), we also considered a fourth measure: the ratio of operating income before depreciation to cash-adjusted assets. However, there was a great deal of noise in this measure. The reason is that the denominator equals cash-adjusted assets (i.e., book value of assets — cash holdings).

19

ACCEPTED MANUSCRIPT

Short Selling and Long-Run Cumulative Abnormal Returns: Hypothesis 3 states that the larger the unexpected increase in a Þrm’s short interest around the SEO announce-

RI P

T

ment, the lower the long-run stock return following the o ering. To investigate this hypothesis, we calculate the long-run cumulative abnormal return (LRCAR) as the di erence between the return of a sample Þrm and that of its matched control Þrm.

SC

We test whether the di erence in average LRCAR between event and non-event Þrms, over the following two-year period, is signiÞcantly di erent from zero by performing a simple

NU

t-test with unequal variances. The results in Panel A, Table 4 indicate that the sample of SEO-announcing Þrms experiences a negative average LRCAR over the following two-year

MA

period. This negative return, however, is not signiÞcantly di erent from zero. The LRCAR is positive and signiÞcant at the 1% level over the six-month period following the SEO

ED

and negative and signiÞcant (at the 10% level) over the following eighteen-month period. However, the di erence in average LRCAR between event and non-event Þrms, over the Þrst

PT

six-month period following the equity issue, is negative (-4.27%) and signiÞcant. The results based on median LRCAR are stronger than those based on average LRCAR.

AC CE

The results in Panel B, Table 4 indicate that the median LRCAR is negative and signiÞcant (at the 1% level) over the following eighteen-month and two-year periods, respectively. In addition, the median LRCAR is negative and signiÞcant (at the 10% level) over the following one-year period. Importantly, the di erence in median LRCAR, between event and non-event Þrms, is also negative and signiÞcant at the 5% level in each of the six-month, one-year, and eighteen-month periods, respectively. In general, the results in Table 4 are consistent with Hypothesis 3. The overall tenor of the Þndings suggests that the larger the increase in short interest around the SEO announcement date, the larger the decline in the long-run stock price following the equity o ering. Lyandres, Sun, and Zhang (2008) Þnd that SEO Þrms appear to invest at a greater rate than their matched counterparts, implying that they are likely to spend the cash raised through the o ering quickly. This observation a ects cash-adjusted assets but not total assets. Also, the timing of this spending is likely to vary across Þrms resulting in variation in this measure that is unrelated to changes in operating performance. Consistent with this intuition, we do observe large outliers in this measure generating a great deal of noise and, therefore, do not use this measure for our robustness check.

20

ACCEPTED MANUSCRIPT

We do not assume nor do we require ine!cient markets to explain the Þndings in Table 4. Our results are based on the superior information-processing ability of some short

RI P

T

sellers. If short sellers are superior at analyzing publicly-available information, then Þrms with higher short interest, on average, must be those which experience a greater deterioration in fundamentals over time (Deshmukh, Gamble, and Howe, 2015). In this sense, the

SC

underperformance of the long-run stock return for these Þrms merely reßects the unravelling deterioration in their fundamentals, causing a downward pressure on their stock price over

NU

time. Therefore, the stock prices of these Þrms are not likely to fully incorporate this information at the SEO announcement because of the uncertainty in both the timing and the

MA

magnitude of the worsening fundamentals. As our results on operating performance indicate, Þrms with larger increases in short interest do experience worse operating performance and

ED

that stock prices (of these Þrms) likely adapt to this information giving rise to the results on the long-run stock return. Also, note that there is a delay in the collection and public

PT

dissemination of the aggregate short-interest data that we use to identify the set of event Þrms. Thus, this information is not public knowledge at the time of the SEO announcement.

AC CE

In addition, our inferences do not require that SEOs (in general) underperform in the long run. We acknowledge that the Þnding of the long-run underperformance of stock returns after SEOs (as in Loughran and Ritter, 1995) may be due to the fact that SEO Þrms have systematically di erent Þrm-speciÞc characteristics (such as idiosyncratic volatility, liquidity, return momentum, and capital investment) from those of their matched control Þrms as documented in Bessembinder and Zhang (2013). Further, Fu and Huang (2016) show that the long-run abnormal returns of seasoned equity o erings disappear in the most recent period of 2003-2012, which is a part of our sample period. Instead, we assume that there is a subset among the SEO-announcing Þrms that might be overvalued and we use the extent of short selling around the SEO announcement to identify these Þrms. As noted earlier, the declining stock price of these Þrms merely adapts to deteriorating fundamentals. We also acknowledge that SEO Þrms in general may have some characteristics that di er from their matched control Þrms (i.e., those without SEOs) as suggested by Bessembinder 21

ACCEPTED MANUSCRIPT

and Zhang (2013). However, there is no evidence in the literature to suggest that changes in short interest are systematically related to the various Þrm-speciÞc characteristics from

RI P

T

Bessembinder and Zhang (2013). Consequently, there is no reason to believe that the difference between the abnormal long-run stock returns of our event Þrms (those with large changes in short interest) and non-event Þrms should be negative, as we document, if short

SC

sellers played no role in identifying overvalued Þrms.

Alternative DeÞnitions of the Event Firms: As a robustness check, we perform the

NU

tests in Tables 2, 3, and 4 for various deÞnitions of the event Þrms based on the percentile cuto of changes in short interest. SpeciÞcally, we classify our event Þrms based on changes

MA

in the short-interest level that are at the 65th percentile and above, the 70th percentile and above, the 80th percentile and above, the 85th percentile and above, and the 90th percentile

ED

and above. The overall results, which are not tabulated, remain qualitatively the same. The robustness of our Þndings to the percentile cuto s used to identify the event Þrms

PT

(i.e., those with large increases in short interest) suggests a notable presence of informed

AC CE

short selling around SEO announcements. 3.2 Shelf O ers vs. Non-Shelf O ers We now explore the relation between short selling and the three measures of performance separately for shelf-registration Þlings of the SEO (shelf o ers) and traditional Þlings of the SEO (non-shelf o ers). Heron and Lie (2004) note the rebound in shelf registration in the 1990s. Autore, Kumar, and Shome (2008) also document a revival of shelf registrations in the 1990s as well as the recent dominance of shelf Þlings over traditional Þlings. They argue that Þrms have begun to increasingly value the embedded option to defer the equity issue in a shelf Þling and that the value of this option is increasing in the Þrm’s volatility. A shelf registration provides the issuing Þrm with the ability to time its equity issue. For example, Autore, Kumar, and Shome (2008) Þnd that, over their sample period 1990-2003, the median number of days between Þling and o er dates for traditional (non-shelf) o ers is 29 and for shelf o ers, it is 127. We Þnd similar results for our sample of Þrms: the 22

ACCEPTED MANUSCRIPT

median number of days between Þling and o er dates for traditional (non-shelf) o ers is 29 and for shelf o ers, it is 174. Our sample period of 1988-2011 is longer and also includes

RI P

T

the more recent period during which the dominance of shelf o ers over traditional o ers has strengthened further. Heron and Lie (2004) Þnd that Þrms that Þle for shelf registration face tight Þnancial conditions and argue that these Þrms defer equity issues so as not to transfer

SC

too much wealth from existing shareholders. In sum, Þrms Þling for shelf registrations enjoy greater ßexibility in terms of timing and deferring the equity issue.

NU

It is, however, possible that some of the shelf-registration Þlers exploit the embedded ßexibility and issue equity when it is overvalued. From a resource-allocation perspective,

MA

it is important to identify such opportunistic timers. The question that arises is: How can one identify Þrms that time their equity issues to exploit temporary overvaluation? If short

ED

sellers possess a superior information-processing ability to identify Þrms that are overvalued as suggested by Engelberg, Reed, and Ringgenberg (2012) and Deshmukh, Gamble, and

PT

Howe (2015), then large increases in short selling around the SEO announcement can be used to identify such opportunistic timers.

AC CE

We examine the three hypotheses separately for shelf o ers and non-shelf o ers. To identify opportunistic timing of equity issues, we focus on event Þrms (i.e., those in the top quartile of short-interest increases). We Þrst present results for the three measures of performance. Next, we test for di erences between shelf o ers and non-shelf o ers among event Þrms and discuss the results and their implications. For completeness, we also present results on the di erences between shelf o ers and non-shelf o ers among non-event Þrms. Short Selling and Announcement-Period Returns: As noted earlier, the event group comprises Þrms in the top quartile of short-interest increases and the non-event group consists of sample Þrms in the bottom three quartiles of short-interest increases. For each of the two groups (i.e., shelf o ers and non-shelf o ers), we test whether the di erence in the average CAR between the event group and the non-event group is signiÞcantly di erent from zero by performing a t-test with unequal variances. For robustness, we perform the non-parametric Wilcoxon rank-sum test to examine whether the di erence in the median 23

ACCEPTED MANUSCRIPT

CAR between the event group and the non-event group is signiÞcantly di erent from zero. The results in Table 5 indicate that the announcement of an SEO, on average, is associated

RI P

T

with a stock price decrease for both shelf o ers and non-shelf o ers. The magnitude of the decline for non-shelf o ers, based on both average CAR and median CAR, is about twice as large as it is for shelf o ers and this result is consistent with prior evidence (Heron and

SC

Lie, 2004 and Autore, Kumar, and Shome, 2008). One possible explanation is that shelf registration provides the Þrm with an option to not issue equity and the lower (negative)

NU

announcement-period return likely reßects the uncertainty about the actual issue. With respect to event and non-event Þrms, the results in Panel A for shelf o ers indicate

MA

that the event Þrms experience a three-day average CAR of -1.99% while the non-event Þrms experience a three-day average CAR of -1.20%. Both of these values are statistically

ED

signiÞcant at the 1% level. The di erence of -0.79%, between the CARs of event and nonevent Þrms, is statistically signiÞcant at the 5% level. Likewise, the results in Panel B for

PT

non-shelf o ers, indicate that the event Þrms experience a three-day CAR of -3.62% while the non-event Þrms experience a three-day CAR of -2.64%. Both of these values are statistically

AC CE

signiÞcant at the 1% level and the di erence of -0.98%, between the CARs of event and non-event Þrms, is also statistically signiÞcant at the 1% level. We also report results from the non-parametric Wilcoxon rank-sum test. For shelf o ers (Panel A), the median event Þrm experiences a CAR of -2.28% while the median non-event Þrm experiences a CAR of -1.12%. The median di erence in the CAR between event and non-event Þrms is signiÞcantly di erent from zero at the 1% level. For non-shelf o ers (Panel B), the median event Þrm experiences a CAR of -3.40% and the median non-event Þrm experiences a CAR of -2.64%. The median di erence of -0.76% in the CAR between event and non-event Þrms is also signiÞcantly di erent from zero at the 1% level. The results in Table 5 thus indicate a negative relation between increases in short interest before the SEO announcement and the magnitude of the stock price response to the announcement. This Þnding holds separately for both shelf o ers and non-shelf o ers and is consistent with Hypothesis 1 and suggests the presence of informed short selling, before the 24

ACCEPTED MANUSCRIPT

SEO announcement, for both shelf and non-shelf o ers. Announcement-Period Return and Manipulative versus Informed Short Sell-

RI P

T

ing: The literature on short selling around SEO announcements contains no evidence on the portion of manipulative versus informed short selling in explaining the announcementperiod return. We recognize the di!culty in addressing this issue given the general nature

SC

of the data on short selling. Here, we make an attempt to Þll this gap in the literature. For instance, Henry and Koski (2010) argue that manipulative short selling around SEOs

NU

is conÞned to non-shelf o erings. We use this result, along with the Þndings in Table 5, to provide some bounds on the extent of informed short selling versus manipulative short

MA

selling in explaining the announcement-period returns for SEOs.9 As the results in Table 5 indicate, the average CAR for announcements of shelf o ers

ED

is -1.40%, while the average CAR for announcements of non-shelf o ers is -2.88%. Based on Henry and Koski (2010), the stock-price response to shelf o ers should be driven by

PT

informed short sellers and an amalgam of other investors, whom we label as noise traders for expositional purposes. In contrast, the average stock-price response to announcements

AC CE

of non-shelf o ers should be driven by informed short sellers, manipulative short sellers, and noise traders. In terms of investor groups, we are adding manipulative short sellers as we transition from shelf o ers to non-shelf o ers. Therefore, as a rough estimate, the di erence between the CARs of -2.88% (for nonshelf o ers) and -1.40% (for shelf o ers) of -1.48% could be attributable to manipulative short selling, which represents about 50% of the total stock-price response. This 50% likely represents an upper bound. In attributing the entire di erence of -1.48% to manipulative short selling, we are being conservative and in favor of manipulative short selling. The reason is that some of this di erence could be driven by an additional inßux of informed short sellers immediately after the announcement. Note that the three-day announcement-period return includes the day of and the day after the announcement. To classify the event and non-event 9

We thank the referee for pointing this gap in the literature and for nudging us to think about addressing this issue.

25

ACCEPTED MANUSCRIPT

Þrms, we use the change in short interest before the SEO announcement. To obtain a rough estimate of the extent of informed short selling, we focus on shelf o ers

RI P

T

- here the announcement-period return should be driven by informed short sellers and noise traders only. Our event Þrms and non-event Þrms identify groups with high and low changes in short interest. Assume that the change in short interest for the event group consists of

SC

informed short selling while the change in short interest for the non-event group represents short selling that is not information-driven. If so, then the di erence in the average CAR

NU

between the event group and the non-event group (from Table 5) equals -0.79%, which represents about 40% of the CAR of -1.99% for the event Þrms. This value likely represents

MA

a lower bound that could be attributable to informed short selling. Again, we are being conservative in favor of noise traders by allocating the entire CAR of -1.20% (for the non-

ED

event group) to noise traders. Therefore, given the presence of informed short sellers and noise traders, our rough estimate suggests that at least 40% of the announcement-period

traders.

PT

return is driven by informed short sellers and the rest (at most 60%) is driven by noise

AC CE

Going back to the CAR of -2.88% for non-shelf o ers, our rough estimates suggest that at most half of this CAR is likely driven by manipulative short sellers. The rest of the CAR of about -1.44% is then possibly driven by both informed short sellers and noise traders. Consequently, at least 40% of -1.44%, which equals about -0.58%, would be attributable to informed short selling. To summarize, in a sample of SEO announcements, our rough estimates suggest an upper bound of 50% of the announcement-period return may be attributable to manipulative short selling while a lower bound of 20% may be attributable to informed short selling. The magnitudes of these bounds also indicate that the extent of informed short selling around SEO announcements is material. In sum, these Þndings provide further evidence in support of informed short selling around announcements of SEOs. Short Selling and Operating Performance: Next, we examine Hypothesis 2 separately for both shelf o ers and non-shelf o ers. For each group, we test whether the di erence in the average cumulative abnormal operating performance between event and non-event 26

ACCEPTED MANUSCRIPT

Þrms, over the following two-year period, is signiÞcantly di erent from zero by performing a simple t-test with unequal variances.

RI P

T

Our results for shelf o ers in Panel A of Table 6 indicate that Þrms completing an SEO experience a decline in their operating performance over the following two-year period. SpeciÞcally, the average cumulative abnormal operating performance of these Þrms, calcu-

SC

lated relative to the control Þrms, equals -3.28% over the following two-year period. This decline in operating performance is statistically signiÞcant at the 1% level and its magni-

NU

tude represents about 56% of the magnitude of the mean level of operating performance (of -5.86%) of these Þrms in the Þscal year of the SEO.

MA

Importantly, among shelf o ers, the di erence in the average cumulative abnormal operating performance, between event and non-event Þrms, is also signiÞcantly di erent from zero

ED

over the following two-year period. This di erence of -7.67% between event and non-event Þrms is also economically signiÞcant; it represents about 67% of the mean level of operating

PT

performance (of -11.39%) of the event Þrms in the Þscal year of the SEO. In other words, among shelf o ers, the cumulative change in the operating performance of Þrms in the top

AC CE

quartile of short-interest increases (after netting the performance change of matched control Þrms) is 7.67% lower than the corresponding (net) change in the operating performance of Þrms in the bottom three quartiles of short-interest increases. We present results for non-shelf o ers in Panel B of Table 6. The results indicate that Þrms completing an SEO experience a decline in their operating performance over the following two-year period. The average cumulative abnormal operating performance of these Þrms, calculated relative to the control Þrms, equals -1.68% over the following two-year period. This decline in operating performance is statistically signiÞcant at the 1% level and its magnitude represents about 27% of the magnitude of the mean level of operating performance (of 6.26%) of these Þrms in the Þscal year of the SEO. However, among non-shelf o ers, the di erence in performance between event and non-event Þrms is not statistically signiÞcant even though both event and non-event Þrms experience a statistically-signiÞcant negative cumulative abnormal operating performance over the following two-year period. 27

ACCEPTED MANUSCRIPT

We also perform the non-parametric Wilcoxon rank-sum test to examine whether the di erence in median cumulative abnormal operating performance, between event and non-

RI P

T

event Þrms, is signiÞcantly di erent from zero for both shelf o ers and non-shelf o ers. Our untabulated results indicate that, for shelf o ers, the di erence in median cumulative abnormal operating performance between event and non-event Þrms, over the following two-

SC

year period, is signiÞcantly di erent from zero (at the 1% level). In contrast, for non-shelf o ers, the di erence in median cumulative abnormal operating performance between event

NU

and non-event Þrms, over the following two-year period, is not signiÞcantly di erent from zero.

MA

The results in Table 6 indicate that operating performance declines following the equity issue for both shelf o ers and non-shelf o ers. But, the di erence in operating performance

ED

between event and non-event Þrms is signiÞcant for shelf o ers only - this result is consistent with Hypothesis 2. Thus, the negative relation between large increases in short interest and

PT

operating performance holds for shelf o ers only. Short Selling and Long-Run Cumulative Abnormal Returns: We now investigate

AC CE

Hypothesis 3 separately for both shelf o ers and non-shelf o ers. We report the LRCAR in six-month increments over the two-year period following the completion of the SEO. The results in Panel A, Table 7 indicate that the sample of Þrms, completing a shelf o er, experience a negative LRCAR over the following two-year period. This LRCAR of -4.74% is statistically signiÞcant from zero at the 10% level while the LRCAR in each of the six-month, one-year, and eighteen-month periods following the SEO, is negative and signiÞcant at the 1% level. In addition, the di erence in average LRCAR between event and non-event Þrms is also negative and signiÞcant at the 5% level or better in each of the six-month, one-year, and eighteen-month periods. In contrast, for non-shelf o ers, the results in Panel B, Table 7 indicate that the sample of Þrms, announcing a non-shelf o er, experience a positive and signiÞcant average LRCAR over the following six-month and one-year periods. However, the di erence in average LRCAR between event and non-event Þrms is statistically nonsigniÞcant in each of the subsequent 28

ACCEPTED MANUSCRIPT

six-month periods.10 We perform the non-parametric Wilcoxon rank-sum test to examine whether the di er-

RI P

T

ence in median LRCAR between event and non-event Þrms is signiÞcantly di erent from zero for both shelf o ers and non-shelf o ers. Our untabulated results indicate that the results based on median LRCAR mirror those based on average LRCAR. SpeciÞcally, for shelf of-

SC

fers, the di erence in median LRCAR between event and non-event Þrms is also negative and signiÞcant at the 5% level or better in each of the six-month, one-year, and eighteen-month

NU

periods. For non-shelf o ers, however, the di erence in median LRCAR between event and non-event Þrms is statistically nonsigniÞcant in each of the subsequent six-month periods.

MA

The results in Table 7 indicate that Þrms issuing equity via shelf registration experience negative cumulative abnormal stock returns over the long run. In addition, the di erence

ED

in LRCAR between event and non-event Þrms is signiÞcant for shelf o ers only. Thus, the negative relation between large increases in short interest and the long-run stock return holds

PT

for shelf o ers only. These results mirror those we document in Table 6 where the negative relation between large increases in short interest and subsequent operating performance holds

AC CE

for shelf o ers only.

Di erences between Shelf and Non-Shelf O ers among Event Firms: Our overall results in Tables 5, 6, and 7 indicate a noteworthy di erence between shelf o ers and non-shelf o ers in terms of the magnitudes of announcement-period returns, cumulative abnormal operating performance, and LRCAR. We now focus on event Þrms (i.e., those in the top quartile of short-interest increases) and investigate whether there is a di erence in the measures of performance between shelf and non-shelf o ers. We perform both a t-test and the Wilcoxon rank-sum test to determine whether the di erences in average and median 10 We perform the non-parametric Wilcoxon rank-sum test to examine whether the di erence in median cumulative abnormal operating performance between event and non-event Þrms is signiÞcantly di erent from zero for both shelf o ers and non-shelf o ers. Our untabulated results indicate that the results based on median LRCAR mirror those based on average LRCAR. SpeciÞcally, for shelf o ers, the di erence in median LRCAR between event and non-event Þrms is also negative and signiÞcant at the 5% level or better in each of the six-month, one-year, and eighteen-month periods. For non-shelf o ers, however, the di erence in median LRCAR between event and non-event Þrms is statistically nonsigniÞcant in each of the subsequent six-month periods.

29

ACCEPTED MANUSCRIPT

values of the three measures of performance are signiÞcantly di erent from zero Our results in Table 8 indicate that the di erences in both average and median announcement-

RI P

T

period returns between shelf and non-shelf o ers are statistically signiÞcant at the 1% level. The magnitude of the negative announcement-period return for shelf o ers is lower than that for non-shelf o ers, resulting in a positive di erence.

SC

With respect to operating performance, we test whether there is a di erence in average cumulative abnormal operating performance between shelf and non-shelf o ers. The results

NU

in Table 8 indicate that the di erence in average cumulative abnormal operating performance between shelf and non-shelf o ers over the following two-year period is signiÞcant at the 5%

MA

level. In addition, the magnitude of the decline in the operating performance is larger among shelf o ers. However, the di erence in median cumulative abnormal operating performance

ED

between shelf and non-shelf o ers is not statistically signiÞcant (p = 0.1089 ). Next, we test whether there is a di erence in both average and median LRCAR between

PT

shelf and non-shelf o ers among event Þrms. The results in Table 8 indicate that the di erence in both average and median LRCAR is negative and statistically signiÞcant at the 5%

AC CE

level or better in each of the six-month, one-year, and eighteen-month periods. In addition, the di erence in average LRCAR is statistically signiÞcant at the 10% level for the two-year period following the SEO. In sum, the magnitude of the decline in the long-run stock price is larger among shelf o ers.

Di erences between Shelf and Non-Shelf O ers among Non-Event Firms: For completeness, we focus on non-event Þrms (i.e., those in the bottom three quartiles of shortinterest increases) as well and report the Þndings on the di erences between shelf and nonshelf o ers in Table 8. The results with respect to the announcement-period returns are qualitatively similar to those for event Þrms. Further, the di erence in average cumulative abnormal operating performance between shelf and non-shelf o ers over the following two-tear period is nonsigniÞcant. The di erence in median cumulative abnormal operating performance over the two-year period between shelf and non-shelf o ers, however, is positive and signiÞcant at the 1% level. We also Þnd that the di erence in average LRCAR 30

ACCEPTED MANUSCRIPT

is statistically signiÞcant at the 5% level or better in each of the six-month and one-year periods while the di erence in median LRCAR is statistically signiÞcant at the 1% level in

RI P

T

the six-month period and at the 10% level in each of one-year and eighteen-month periods. Implications: The important takeaway from Table 8 is that the di erences in both the operating performance and LRCAR between shelf and non-shelf o ers are more pronounced

SC

among event Þrms than among non-event Þrms. Taken together, the Þndings on the di erence between shelf and non-shelf o ers suggest that shelf o ers experience inferior operating and

NU

long-run stock performance. The overall worse performance of shelf o ers, relative to nonshelf o ers, also suggests that the potential for opportunistic timing of equity issues is higher

MA

among shelf Þlers. Importantly, the inferior performance of shelf o ers among Þrms with large increases in short interest (i.e., the event Þrms) highlights a potential informational

ED

role played by short sellers in identifying such opportunistic market timers.

PT

4. Summary and Conclusion

Is there informed short selling around announcements of seasoned equity o erings (SEOs)?

AC CE

To answer this question, we adopt a three-pronged empirical approach. Prior research on short selling around SEO announcements has largely focused on the manipulation rationale. Consequently, the research on informed short selling around SEO announcements has remained scant with inconclusive Þndings. We attempt to Þll this important gap in the literature.

We formulate and test three hypotheses. Our Þrst hypothesis states that the larger the increase in short interest before the SEO announcement, the greater the price decrease at the announcement. Our second hypothesis states that the larger the increase in a Þrm’s short interest around the SEO announcement, the greater the decline in its subsequent operating performance. The third hypothesis states that the larger the increase in a Þrm’s short interest around the SEO announcement, the lower its long-term stock price performance. We adopt empirical methods and tests that allow us to detect informed short selling if it is present in

31

ACCEPTED MANUSCRIPT

the data. On the other hand, our tests should produce both economically and statistically nonsigniÞcant Þndings if short selling around SEO announcements is purely manipulative

RI P

T

and/or motivated by ‘non-informational’ considerations such as hedging. Our results indicate a negative relation between the increase in short interest prior to the SEO announcement and the announcement-period return. SpeciÞcally, Þrms with large

SC

increases in short interest (i.e., in the top quartile) experience an announcement-period return that is about one percentage point lower (i.e., more negative) than that for the rest

NU

of the sample Þrms. We also Þnd that Þrms in the top quartile of short-interest increases around the SEO announcement experience an economically signiÞcant decline in operating

MA

performance, relative to the matched control sample, over the following two-year period. This result is largely robust to alternative measures of operating performance. Further, over

ED

the following six-month period, Þrms in the top quartile of short-interest increases around the SEO announcement experience a cumulative abnormal stock return that is 4.27 percentage

PT

points lower than the cumulative abnormal stock return for the rest of the SEO-announcing Þrms. In our tests, we di erentiate between Þrms with large increases in short interest (the

AC CE

top quartile) with those with comparatively smaller increases in short interest. Our overall results remain qualitatively the same when we consider several percentile cuto s to identify Þrms with large short-interest increases. We also examine the relation between short selling and the three measures of performance separately for shelf-registrations of SEOs (shelf o ers) and traditional Þlings of SEOs (nonshelf o ers). Firms Þling for shelf registrations enjoy greater ßexibility in terms of both timing and deferring the equity issue relative to traditional Þlers of the SEO. This ßexibility may provide a Þrm with an opportunity to issue equity when it is overvalued. If short sellers possess a superior ability to identify Þrms that are overvalued, then increases in short interest can help identify potential opportunistic market timers. In general, our results for both shelf o ers and non-shelf o ers mirror those we Þnd for the overall sample of SEOs. However, shelf o ers experience inferior operating and long-run stock price performance relative to non-shelf o ers. In addition, the di erence in 32

ACCEPTED MANUSCRIPT

performance between shelf o ers and non-shelf o ers is more pronounced among Þrms with larger short-interest increases. The inferior long-term performance of shelf Þlers suggests

RI P

T

that the potential for opportunistic timing of equity issues is greater among shelf Þlers. The important result is that the inferior performance of shelf o ers among Þrms with large

identifying such opportunistic market timers.

SC

increases in short interest highlights a potential informational role played by short sellers in

Our overall evidence thus provides strong support for our hypotheses and for informed

NU

short selling around SEO announcements with important implications for regulation, market e!ciency, and price discovery. For instance, if short selling is driven by worsening Þrm

MA

fundamentals, then it provides useful information to other market participants in the capital allocation process and would represent a positive externality. Consequently, any regula-

ED

tion that hampers short selling around SEOs will make prices less e!cient with negative

AC CE

PT

implications for equity issues and for broader resource allocation.

33

ACCEPTED MANUSCRIPT

References

T

Asquith, P. and D. W. Mullins, Jr., 1986, Equity issues and o ering dilution, Journal of Financial Economics 15, 61—89.

RI P

Autore, D., R. Kumar, and D. Shome, 2008, The revival of shelf-registered corporate equity o erings, Journal of Corporate Finance 14, 32—50.

SC

Autore, D., 2010, Does Rule 10b-21 increase SEO discounting? Journal of Financial Intermediation 20, 231-247.

NU

Barber, B. M. and J. D. Lyon, 1996, Detecting abnormal operating performance: The empirical power and speciÞcation of test-statistics, Journal of Financial Economics 41, 359-400. –––, 1997, Detecting long-run abnormal stock returns: The empirical power and speciÞcation of test-statistics, Journal of Financial Economics 43, 341-372.

MA

Bessembinder, H. and F. Zhang, 2013, Firm characteristics and long-run stock returns after corporate events, Journal of Financial Economics 109, 83-102.

ED

Boehmer, E., C. M. Jones, and X. Zhang, 2008, Which Shorts Are Informed, Journal of Finance 63, 491-527. Boehmer, E. and J. Wu, 2013, Short Selling and the Price Discovery Process, Review of Financial Studies 26, 287—322.

PT

Brav, A., C. Geczy, and P. A. Gompers, 2000, Is the abnormal return following equity issuances anomalous, Journal of Financial Economics 56, 209-249.

AC CE

Christophe, S., M. Ferri, and J. Angel, 2004, Short selling prior to earnings announcements, Journal of Finance 59, 1845-1875. Christophe, S., M. Ferri, and J. Hsieh, 2010, Informed trading before analyst downgrades: evidence from short sellers, Journal of Financial Economics 95, 85-106. Corwin, S. A., 2003, The determinants of underpricing for seasoned equity o ers, Journal of Finance 58, 2249—2279. Deshmukh, S., K. J. Gamble, and K. M. Howe, 2015, Short Selling and Firm Operating Performance, Financial Management, April 2015. Diamond, D. and R. Verrecchia, 1987, “Constraints on short-selling and asset price adjustment to private information,” Journal of Financial Economics, 18, 277—311. Engelberg, J. E., A. V. Reed, and M. C. Ringgenberg, 2012, How are shorts informed? Short sellers, news, and information processing, Journal of Financial Economics 105: 260-278. Fu, F. and S. Huang, 2016, The Persistence of Long-Run Abnormal Returns Following Stock Repurchases and O erings, Management Science 62, 964-984. Gerard, B. and V. Nanda, 1993, Trading and manipulation around seasoned equity o erings, Journal of Finance 48, 213—245.

ACCEPTED MANUSCRIPT

Henry, T. R. and J. L. Koski, 2010, Short selling around seasoned equity o erings, Review of Financial Studies 23, 4389-4418.

RI P

T

Heron, R. A. and E. Lie, 2004, A comparison of the motivations for and the information content of di erent types of equity o erings, Journal of Business 77, 605—632. Karpo , J. and X. Lou, 2010, Short sellers and Þnancial misconduct, Journal of Finance 65, 18791913.

SC

Kim, K. and H. Shin, 2004, The puzzling increase in the underpricing of seasoned equity o erings, Financial Review 39, 343—365. Loughran, T. and J. R. Ritter, 1995, The new issues puzzle, Journal of Finance, 50, 23—51.

NU

–––, 1997, The operating performance of Þrms conducting seasoned equity o erings, Journal of Finance, 52, 1823-1850.

MA

Lyandres, E., L. Sun, and L. Zhang, 2008, The new issues puzzle: testing the investment-based explanation, Review of Financial Studies 21, 2825-2855. Masulis, R. W. and A. N. Korwar, 1986, Seasoned equity o erings: An empirical investigation, Journal of Financial Economics 15, 91-118.

ED

Mikkelson, W. H. and M. M. Partch, 1985, Stock price e ects and costs of secondary distributions, Journal of Financial Economics 14, 165-194.

PT

Myers, S. C. and N. S. Majluf, 1984, Corporate Þnancing and investment decisions when Þrms have information that investors do not have, Journal of Financial Economics 13, 187-221.

AC CE

Ritter, J., 2003, Investment banking and securities issuance. In G. Constantanides, M. Harris, and R. Stulz (eds.), Handbook of Economics of Finance, Amsterdam, The Netherlands: North Holland. SaÞeddine, A. and W. J. Wilhelm, Jr., 1996, An empirical investigation of short-selling activity prior to seasoned equity o erings, Journal of Finance 51, 729—749. Singal, V. and L. Xu, 2005, Do short sellers know more? Evidence from a natural experiment, Working Paper, Virginia Tech. Spiess, D. K. and J. A!eck-Graves, 1995, Underperformance in long-run stock returns following seasoned equity o erings, Journal of Financial Economics 38, 243-267.

ACCEPTED MANUSCRIPT Table 1 Summary Statistics

SC

RI P

T

This table presents key summary statistics for the sample of seasoned equity o erings (SEO) over the period 1988-2011. Panel A presents summary statistics for the full sample of SEO-announcing Þrms. Panel B presents summary statistics for subsamples of Þrms that we use to perform the three tests: announcement-period return, operating performance following the equity issue, and long-run cumulative abnormal stock return following the equity issue. The event group comprises Þrms in the top quartile (i.e., the 75th percentile and above) of those experiencing an increase in the short-interest level and the nonevent group comprises SEO-announcing Þrms that are not in the top quartile of increases in the short-interest level. Short Interest equals the ratio of the number of shares sold short to the number of shares outstanding. Operating performance equals the ratio of operating income before depreciation (OIBD) to book value of total assets (Assets). In Panel B, with the exception of Number of SEOs, all values represent averages.

MA

Variable

NU

Panel A: Full Sample of SEO Announcements

0.0106 2090.90 1576.07 0.5949 0.0347 0.0369

0.1025 294.05 325.53 0.3826 0.0156 0.0180 7129

PT

ED

Operating Performance Assets ($ millions) Market Capitalization ($ millions) Book-to-Market Ratio Short Interest Before SEO Announcement Short Interest After SEO Announcement Number of SEOs

Full Sample Mean Median

AC CE

Panel B: Subsamples of SEO Announcements Subsample: Announcement-Period Return Variable Full Event

Number of SEOs Short Interest before SEO Announcement Change in Short Interest before SEO Announcement

Variable

6852 0.0351 0.0030

Subsample: Operating Performance Full

Number of SEOs Short Interest after SEO Announcement Change in Short Interest around SEO Announcement Assets ($ millions) Operating Performance

5019 0.0339 0.0024 1909.20 0.0148

1713 0.0677 0.0260

5139 0.0242 -0.0047

Event

NonEvent

1255 0.0639 0.0238 1146.80 -0.0199

3764 0.0239 -0.0047 2163.39 0.0264

Subsample: Long-Run Cumulative Abnormal Stock Return Variable Full Event Number of SEOs Short Interest after SEO Announcement Change in Short Interest around SEO Announcement Market Capitalization ($ millions) Book-to-Market Ratio

5301 0.0400 0.0017 1593.07 0.5951

NonEvent

1326 0.0731 0.0251 1201.00 0.5549

NonEvent 3975 0.0289 -0.0061 1723.86 0.6086

ACCEPTED MANUSCRIPT Table 2

T

Relation between Short Interest and Announcement-Period Return for SEOs

NU

SC

RI P

This table presents results on the cumulative abnormal return (CAR) around the announcement date of an SEO. The CAR is calculated over a three-day window that begins the day before the announcement date and ends the day after the announcement date. The abnormal returns are based on a market model estimated using the CRSP value-weighted index. The estimation period for the parameters of the market model is 255 days and ends 46 trading days before the event date. The event group comprises SEO-announcing Þrms in the top quartile (i.e., the 75th percentile and above) of short-interest increases and the nonevent group comprises Þrms that are not in the top quartile of increases in the short-interest level. Short Interest equals the ratio of the number of shares sold short to the number of shares outstanding. We use both a t-test (based on unequal variances) and the Wilcoxon rank-sum test to examine whether the average and median CAR, respectively, are signiÞcantly di erent from zero. The t-statistics (and p-values where noted) are in parentheses. *** refers to signiÞcance at the 1% level.

Event Firms

NonEvent Firms

Di erence (Event - NonEvent)

Average CAR

-0.0209*** (-20.44)

-0.0281*** (-16.03)

-0.0185*** (-15.04)

-0.0096*** (-4.48)

Median CAR

-0.0209*** (p 0.0001)

-0.0183*** (p 0.0001)

-0.0095*** (p 0.0001)

AC CE

PT

ED

MA

All Equity O erings

-0.0278*** (p 0.0001)

ACCEPTED MANUSCRIPT Table 3 Relation between Short Interest and Cumulative Abnormal Operating Performance for SEOs

NU

SC

RI P

T

This table presents results on the cumulative abnormal operating performance for the full sample of SEOs and on the di erence in the cumulative abnormal operating performance between the event group and the nonevent group. The event group comprises SEO-announcing Þrms in the top quartile (i.e., the 75th percentile and above) of short-interest increases and the nonevent group comprises Þrms that are not in the top quartile of increases in the short-interest level. Short Interest equals the ratio of the number of shares sold short to the number of shares outstanding. Operating performance equals the ratio of operating income before depreciation (OIBD) to total assets. Abnormal operating performance for event Þrm i in year t equals the change in the annual operating performance for the event Þrm minus the median annual change in the operating performance among matched control Þrms. The sample Þrms are matched to control Þrms based on pre-event operating performance, Þrm size, and industry. We use both a t-test (based on unequal variances) and the Wilcoxon rank-sum test to examine whether the average and median cumulative abnormal operating performance, respectively, are signiÞcantly di erent from zero. The t-statistics (and p-values where noted) are in parentheses. * and *** refer to signiÞcance at the 10% and 1% level, respectively.

MA

Panel A: Average Cumulative Abnormal Operating Performance

2

-0.0234*** (-4.61)

ED

-0.0156*** (-3.03)

-0.0116* (-1.93)

-0.0160 (-1.37)

-0.0492*** (-4.22)

-0.0149*** (-2.69)

-0.0344*** (-2.66)

-0.0276*** (-2.75)

AC CE

1

PT

Year

Average Cumulative Abnormal Operating Performance All Equity Di erence O erings Event Firms NonEvent Firms (Event - NonEvent)

Panel B: Median Cumulative Abnormal Operating Performance

Year

Median Cumulative Abnormal Operating Performance All Equity Di erence O erings Event Firms NonEvent Firms (Event - NonEvent)

1

-0.0016** (p = 0.0392)

-0.0057*** (p = 0.0073)

-0.0006 (p = 0.4147)

-0.0051*** (p = 0.0064)

2

-0.0028*** (p = 0.0094)

-0.0114*** (p = 0.0014)

-0.0010 (p = 0.2522)

-0.0104*** (p = 0.0007)

ACCEPTED MANUSCRIPT Table 4 Relation between Short Interest and Long-Run Cumulative Abnormal Return for SEOs

SC

RI P

T

This table presents results on the long-run cumulative abnormal return (LRCAR) for the full sample of SEOs and on the di erence in LRCAR between the event group and the nonevent group. The event group comprises SEO-announcing Þrms in the top quartile (i.e., the 75th percentile and above) of short-interest increases and the nonevent group comprises Þrms that are not in the top quartile of increases in the short-interest level. Short Interest equals the ratio of the number of shares sold short to the number of shares outstanding. The LRCAR for a sample Þrm is calculated relative to a control Þrm that is matched to the sample Þrm in terms of size and the book-to-market ratio. We use both a t-test (based on unequal variances) and the Wilcoxon rank-sum test to examine whether the average and median LRCAR, respectively, are signiÞcantly di erent from zero. The t-statistics (and p-values where noted) are in parentheses. *, **, and *** refer to signiÞcance at the 10%, 5%, and the 1% level, respectively.

Average Long-Run Cumulative Abnormal Return Di erence Event Firms NonEvent Firms (Event - NonEvent)

MA

Year

All Equity O erings

NU

Panel A: Average Long-Run Cumulative Abnormal Return

0.0211*** (2.69)

-0.0109 (-0.74)

0.0318*** (3.46)

-0.0427*** (2.45)

1

0.0001 (0.01)

-0.0332 (-1.41)

0.0110 (0.76)

-0.0442 (-1.60)

1.5

-0.0273* (-1.70)

-0.0636** (-2.00)

-0.0154 (-0.83)

-0.0483 (-1.31)

2

-0.0210 (-1.01)

-0.0105 (-0.43)

-0.0434 (-0.94)

PT

ED

0.5

AC CE

-0.0539 (-1.38)

Panel B: Median Long-Run Cumulative Abnormal Return

Year

All Equity O erings

Median Long-Run Cumulative Abnormal Return Di erence Event Firms NonEvent Firms (Event - NonEvent)

0.0012 (p = 0.8047)

-0.0074 (p = 0.4258)

0.0046 (p = 0.4465)

-0.0119** (p = 0.0301)

1

-0.0161* (p = 0.0552)

-0.0383* (p = 0.0554)

-0.0091 (p = 0.2729)

-0.0292** (p = 0.0188)

1.5

-0.0427*** (p = 0.0037)

-0.0899*** (p = 0.0009)

-0.0254 (p = 0.1543)

-0.0645** (p = 0.0461)

2

-0.0497*** (p = 0.0005)

-0.0904*** (p = 0.0019)

-0.0323** (p = 0.0453)

-0.0581 (p = 0.1599)

0.5

ACCEPTED MANUSCRIPT Table 5 Relation between Short Interest and Announcement-Period Return: Shelf and Non-Shelf O ers

NU

SC

RI P

T

This table presents results on the cumulative abnormal return (CAR), around the announcement date of an SEO, separately for shelf o ers (Panel A) and non-shelf o ers (Panel B). The CAR is calculated over a three-day window that begins the day before the announcement date and ends the day after the announcement date. The abnormal returns are based on a market model estimated using the CRSP value-weighted index. The estimation period for the parameters of the market model is 255 days and ends 46 trading days before the event date. The event group comprises SEO-announcing Þrms in the top quartile (i.e., the 75th percentile and above) of short-interest increases and the nonevent group comprises Þrms that are not in the top quartile of increases in the short-interest level. Short Interest equals the ratio of the number of shares sold short to the number of shares outstanding. We use both a t-test (based on unequal variances) and the Wilcoxon rank-sum test to examine whether the average and median CAR, respectively, are signiÞcantly di erent from zero. The t-statistics (and p-values where noted) are in parentheses. *** refers to signiÞcance at the 1% level.

Panel A: Shelf O ers

Event Firms

NonEvent Firms

Di erence (Event - NonEvent)

Average CAR

-0.0140*** (-10.28)

-0.0199*** (-7.35)

-0.0120*** (-7.64)

-0.0079** (-2.53)

Median CAR

-0.0131*** (p 0.0001)

-0.0112*** (p 0.0001)

-0.0116*** (p = 0.0013)

ED

MA

All Equity O erings

PT

-0.0228*** (p 0.0001)

AC CE

Panel B: Non-Shelf O ers

All Equity O erings

Event Firms

NonEvent Firms

Di erence (Event - NonEvent)

Average CAR

-0.0288*** (-25.14)

-0.0362*** (-15.77)

-0.0264*** (-19.97)

-0.0098*** (-3.71)

Median CAR

-0.0279*** (p 0.0001)

-0.0340*** (p 0.0001)

-0.0264*** (p 0.0001)

-0.0076*** (p = 0.0001)

ACCEPTED MANUSCRIPT Table 6 Relation between Short Interest and Cumulative Abnormal Operating Performance: Shelf and Non-Shelf O ers

NU

SC

RI P

T

This table presents results on operating performance separately for shelf o ers (Panel A) and non-shelf o ers (Panel B). The results include the cumulative abnormal operating performance for all equity o erings and the di erence in the cumulative abnormal operating performance between the event group and the nonevent group. The event group comprises SEO-announcing Þrms in the top quartile (i.e., the 75th percentile and above) of short-interest increases and the nonevent group comprises Þrms that are not in the top quartile of increases in the short-interest level. Short Interest equals the ratio of the number of shares sold short to the number of shares outstanding. Operating performance equals the ratio of operating income before depreciation (OIBD) to total assets. Abnormal operating performance for event Þrm i in year t equals the change in the annual operating performance for the event Þrm minus the median annual change in the operating performance among matched control Þrms. The sample Þrms are matched to control Þrms based on pre-event operating performance, Þrm size, and industry. We use a t-test (based on unequal variances) to examine whether the average cumulative abnormal operating performance is signiÞcantly di erent from zero. The t-statistics are in parentheses. *, **, and *** refer to signiÞcance at the 10%, 5%, and the 1% level, respectively.

-0.0154 (-1.33)

2

-0.0328*** (-3.00)

-0.0426* (-1.91)

-0.0063 (-0.47)

-0.0363 (-1.39)

-0.0896*** (-3.21)

-0.0129 (-1.18)

-0.0767** (-2.56)

Year 1

2

AC CE

PT

1

Average Cumulative Abnormal Operating Performance Di erence Event Firms NonEvent Firms (Event - NonEvent)

ED

Year

All Equity O erings

MA

Panel A: Shelf O ers

All Equity O erings

Panel B: Non-Shelf O ers

Average Cumulative Abnormal Operating Performance Di erence Event Firms NonEvent Firms (Event - NonEvent)

-0.0132** (-2.52)

-0.0190* (-1.88)

-0.0113* (-1.84)

-0.0077 (-0.65)

-0.0168*** (-3.02)

-0.0253** (-2.36)

-0.0141** (-2.16)

-0.0113 (-0.90)

ACCEPTED MANUSCRIPT Table 7 Relation between Short Interest and Long-Run Cumulative Abnormal Return: Shelf and Non-Shelf O ers

SC

RI P

T

This table presents results on the long-run stock return separately for shelf o ers (Panel A) and non-shelf o ers (Panel B). The results include the long-run cumulative abnormal return (LRCAR) for all equity o erings and the di erence in LRCAR between the event group and the nonevent group. The event group comprises SEO-announcing Þrms in the top quartile (i.e., the 75th percentile and above) of short-interest increases and the nonevent group comprises Þrms that are not in the top quartile of increases in the short-interest level. Short Interest equals the ratio of the number of shares sold short to the number of shares outstanding. The LRCAR for a sample Þrm is calculated relative to a control Þrm that is matched to the sample Þrm in terms of size and the book-to-market ratio. We use a t-test (based on unequal variances) to examine whether the average LRCAR is signiÞcantly di erent from zero. The t-statistics are in parentheses. *, **, and *** refer to signiÞcance at the 10%, 5%, and the 1% level, respectively.

MA

Average Long-Run Cumulative Abnormal Return Di erence Event Firms NonEvent Firms (Event - NonEvent)

-0.0255*** (-2.40)

-0.0946*** (-4.56)

-0.0024 (-0.20)

-0.0922*** (-3.82)

1

-0.0530*** (-3.20)

-0.1515*** (-4.65)

-0.0206 (-1.08)

-0.1309*** (-3.46)

1.5

-0.0688*** (-3.07)

-0.1574*** (-3.55)

-0.0395 (-1.52)

-0.1179** (-2.29)

2

-0.0474* (-1.76)

-0.1183* (-3.32)

-0.0248 (-0.78)

-0.0935 (1.60)

AC CE

Year

PT

0.5

ED

Year

All Equity O erings

NU

Panel A: Shelf O ers

All Equity O erings

Panel B: Non-Shelf O ers

Average Long-Run Cumulative Abnormal Return Di erence Event Firms NonEvent Firms (Event - NonEvent)

0.0585*** (5.05)

0.0685*** (3.17)

0.0551*** (4.04)

0.0133 (0.52)

1

0.0414** (2.22)

0.0417 (1.20)

0.0412* (1.88)

0.0005 (0.01)

1.5

-0.0073 (-0.32)

-0.0014 (-0.03)

-0.0093 (-0.35)

0.0079 (0.15)

2

-0.0097 (-0.32)

0.0126 (0.22)

-0.0173 (-0.47)

0.0299 (0.44)

0.5

ACCEPTED MANUSCRIPT Table 8

T

Di erences in the Measures of Performance between Shelf O ers and NonShelf O ers

MA

NU

SC

RI P

This table presents results on the average and median di erences between shelf and non-shelf o ers for the three measures of performance. We calculate the di erences separately among event Þrms and nonevent Þrms. The event group comprises SEOannouncing Þrms in the top quartile (i.e., the 75th percentile and above) of short-interest increases and the nonevent group comprises Þrms that are not in the top quartile of increases in the short-interest level. Short Interest equals the ratio of the number of shares sold short to the number of shares outstanding. The announcement-period return equals the three-day cumulative abnormal return (CAR) around the announcement date of an SEO. The CAR is calculated over a three-day window that begins the day before the announcement date and ends the day after the announcement date. The abnormal returns are based on a market model estimated using the CRSP value-weighted index. The estimation period for the parameters of the market model is 255 days and ends 46 trading days before the event date. Operating performance equals the ratio of operating income before depreciation (OIBD) to total assets. Abnormal operating performance for event Þrm i in year t equals the change in the annual operating performance for the event Þrm minus the median annual change in the operating performance among matched control Þrms. The sample Þrms are matched to control Þrms based on pre-event operating performance, Þrm size, and industry. The long-run cumulative abnormal return (LRCAR) for a sample Þrm is calculated relative to a control Þrm that is matched to the sample Þrm in terms of size and the book-to-market ratio. We use both a t-test (based on unequal variances) and the Wilcoxon rank-sum test to examine whether the average and median di erences, respectively, are signiÞcantly di erent from zero. The t -statistics (and p-values where noted) are in parentheses. *, **, and *** refer to signiÞcance at the 10%, 5%, and the 1% level, respectively. Di erences Between Shelf O ers and Non-Shelf O ers

Year

NonEvent Firms Di erence Di erence in Average in Median

0.0162*** (4.58)

0.0112*** (p 0.0001)

0.0144*** (7.00)

0.0152*** (p 0.0001)

1

-0.0236 (-0.96)

-0.0005 (p = 0.4123)

0.0050 (0.34)

0.0013 (p = 0.2663)

Cumulative Abnormal Operating Performance

2

-0.0642** (-2.15)

-0.0075 (p = 0.1089)

0.0012 (0.09)

0.0068*** (p = 0.0035)

Long-Run Cumulative Abnormal Return (LRCAR)

0.5

-0.1631*** (-5.45)

-0.1284*** (p 0.0001)

-0.0575*** (-3.03)

-0.0625*** (p 0.0001)

Long-Run Cumulative Abnormal Return (LRCAR)

1

-0.1932*** (-4.05)

-0.1525*** (p 0.0001)

-0.0618** (-2.12)

-0.0202* (p = 0.0687)

Long-Run Cumulative Abnormal Return (LRCAR)

1.5

-0.1560** (-2.44)

-0.1354** (p = 0.0232)

-0.0302 (-0.81)

-0.0755* (p = 0.0627)

Long-Run Cumulative Abnormal Return (LRCAR)

2

-0.1309* (-1.75)

-0.0501 (p = 0.3338)

-0.0075 (-0.16)

-0.0398 (p = 0.4701)

AC CE

Cumulative Abnormal Operating Performance

PT

Announcement-Period Return

Event Firms Di erence Di erence in Average in Median

ED

Performance Measure

ACCEPTED MANUSCRIPT

!"#$!"#%&'

AC CE

PT

ED

MA

NU

SC

RI P

T

!"#$%&'$()%&'*!+$",&'#$"!--().$/'&0)+$123$/))&0)4!*!)#"$ 5!'%&'*$#,'!!$"!6/'/#!$#!"#"$#&$+!#!'*()!$7,!#,!'$()%&'*!+$",&'#$"!--().$!8("#"$ 9!"0-#"$6'&:(+!$"#'&).$"066&'#$%&'$()%&'*!+$",&'#$"!--().$$