0

Short Selling and the Informational Efficiency to Private Information:

A Natural Experiment

Hae mi Choi∗

ABSTRACT

Exploiting a regulatory change in short-sale constraints (Regulation SHO) as a natural

experiment, this paper examines the asymmetric effect of short selling on the informational

efficiency of stock prices in response to private information. I find that short-sellers are informed

traders about forthcoming analyst news and trade on their negative private information. When

short-sale constraints are relaxed for pilot stocks, trading volume increases prior to the analyst

announcement only for bad news and not for good news. Short-sellers’ private information is

incorporated into stock prices prior to the announcement, which increases the stock price

sensitivity to bad news but not for good news. The findings are consistent with the Diamond and

Verrecchia (1987) model that predicts that short-sellers are informed traders who increase the

speed of adjustment of stock prices to private information. In the cross-section, the effect of

short-selling activity is stronger for firms with weak information environments (i.e., small firms)

and firms with more differences in opinion (i.e., high analyst forecast dispersion).

JEL classification codes: G10; G14; M40

Keywords: Short selling; Regulation SHO; Informational Efficiency; Analysts; Stock Price Sensitivity; Private Information

∗ Hae mi Choi, Quinlan School of Business, Loyola University Chicago, email: hchoi2@luc.edu, phone (office): (312) 915-6320.

1

1. Introduction

Prior literature finds mixed evidence on whether short-sellers are informed about

forthcoming firm-specific news announcements. For example, Christophe et al. (2010) show that

short activity increases prior to analyst recommendation downgrades, which can be explained by

analyst tipping. On the other hand, Blau and Wade (2012) find evidence of abnormal short

activity prior to both downgrades and upgrades, which indicates that short selling prior to analyst

recommendations are more speculative than informed. One reason for the differences in prior

findings may be due to the endogenous relation between firm performance, analyst

recommendations, and short-selling activity. This paper re-examines whether short sellers are

informed prior to analyst earnings forecast announcements, by exploiting an exogenous shock to

short-sale constraints (Regulation SHO). It is important to understand the informativeness of

short sellers due to their effect on stock price efficiency, especially during firms’ news

announcement periods (Boehmer et al., 2015). The main objective of the paper is to examine

short selling around both negative and positive earnings forecast revisions and its implications on

the informational efficiency of stock prices, while alleviating endogeneity concerns under a

natural experiment setting. If short-sellers correctly anticipate the nature of forthcoming analyst

forecasts, then I expect the effect of short-selling to be asymmetric between negative and

positive news.

A temporary regulatory experiment on short-selling constraints, namely Regulation SHO

(hereafter Reg SHO), serves as a natural experiment to investigate the causal effect of short-

selling activity on the informational efficiency of stock prices. When the Securities and

Exchange Commission (SEC) adopted Reg SHO from May 2005 to August 2007, this mandated

a temporary suspension of short-sale price tests for a set of randomly selected pilot stocks.

2

Stocks included in the Russell 3000 Index were ranked by average daily trading volume, and

every third stock was suspended from the uptick test for the NYSE and the bid test for the

NASDAQ. This led to an exogenous decrease in short-sale constraints and an increase in short-

selling activity for the pilot stocks (SEC, 2007; Diether, Lee, and Werner, 2009; Angelis et al.,

2013; Grullon et al., 2015). Reg SHO provides a natural experimental setting in which to

compare the effect of short-selling activity before and after the regulation for pilot stocks, since

stocks were chosen randomly (Fang et al., 2015; Ke et al., 2015; Li and Zhang, 2016).

This study takes a difference-in-differences (DiD) estimation approach and compares the

trading volume and the stock price sensitivity of pilot stocks (the treatment group)–stocks that

were randomly chosen by Reg SHO–to that of non-pilot stocks (the control group). If short

sellers are informed about forthcoming forecast revisions, I expect the trading activity of these

pessimistic investors to increase only prior to negative revisions and not prior to positive

revisions, particularly for pilot stocks with weaker short-sale constraints. As a result, short sellers’

negative private information is incorporated prior to the analyst’s announcement, which

increases the speed of adjustment of stock prices to negative private information of pilot stocks

relative to non-pilot stocks (Diamond and Verrecchia, 1987).

When short-sale constraints decrease, I find statistically and economically significant

increases in trading activity and the stock price sensitivity to negative forecast revisions but not

to positive forecast revisions. The DiD estimates show that both the trading volume and the stock

price sensitivity of pilot stocks increase prior to negative revisions (i.e., from day -5 to -1), while

there are no changes prior to positive revisions. These findings suggest that short sellers are

informed traders about forthcoming negative revisions prior to analysts’ forecast announcements.

More informed trading improves the price adjustment to negative private information and

3

increases the informational efficiency of stock prices. In the cross-section, I find that the effect of

Reg SHO is stronger for smaller firms, and for firms with more differences in opinion. This

suggests that firms with weaker information environments are affected the most by increased

short-selling activity from Reg SHO.

I further conduct additional robustness tests. First, I examine the changes in trading

volume and the stock price sensitivity around the permanent removal of the uptick test for all

stocks in 2007 to verify the findings. The non-pilot stocks experienced an exogenous shock to

short-selling activity when the uptick test was removed. In contrast, the pilot stocks experienced

no change at this time since the uptick test had already been removed. I find results from the

permanent removal period to be similar to those from the initial pilot program period, which

supports the main empirical findings. The differential changes in trading volume and the stock

price sensitivity between these two groups reverse: trading volume and the stock price sensitivity

to negative revisions increase for non-pilot stocks relative to pilot stocks. In contrast, there are no

changes prior to positive revision announcements. The similar patterns found from the initial

pilot program and the subsequent permanent removal for all remaining stocks provide strong

evidence of the informativeness of shorting activity and its causal effect on the informational

efficiency of stock prices.

Second, I examine short selling around analyst stock recommendation changes, instead of

earnings forecasts, and once again find results supporting the main analysis. Prior to downgrade

announcements, stock prices decline even more for pilot stocks during Reg SHO. I do not find

differences in the stock price sensitivity around upgrade announcements. Lastly, I repeat the

main analysis around firms’ earnings announcement dates, instead of analysts’ forecast

announcement dates, and find similar results.

4

Prior literature finds that short selling relates inversely with future returns over longer

horizons, indicating that the short sellers appear to be informed (Senchack and Starks, 1993;

Aitken et al., 1998; Desai et al., 2002; Cohen et al., 2007; Boehmer et al., 2008). Another stream

of literature examines the changes in shorting activity prior to firm news announcements, since

short sellers’ private information is difficult to observe (Francis et al., 2005; Desai et al., 2006;

Christophe, Ferri, and Hsieh, 2010; Boehmer, Jones, and Zhang, 2015).1 However, less research

is done about the effect of short sellers’ private information on the informational efficiency of

stock prices to firm news announcements over short horizons. For example, Christophe et al.

(2010) and Boehmer et al. (2015) show that short sellers trade on their private information before

analysts announce their forecast revisions, but do not examine price efficiency to private

information. This study aims to fill in the gap in the literature by showing that increased short-

selling activity from Reg SHO led to negative private information being incorporated into stock

prices more efficiently.

To my knowledge, this is the first study to directly examine the asymmetric effect of

short selling on the speed of price adjustment to negative versus positive private information. I

extend the findings of Christophe, Ferri, Hsieh (2010) that examine short-selling activity before

recommendation downgrades but not upgrades. The current findings differ from Blau and Wade

(2012) in that I find that short-sellers are informed and can anticipate the type of forthcoming

analysts’ earnings forecast news. I find asymmetric changes between bad and good news, while

Blau and Wade (2012) find short sellers increase trading prior to both recommendation 1 In general, these papers show that short sellers have an informational advantage and trade on private information prior to public announcements. Christophe, Ferri, and Hsieh (2010) show that short sellers are well informed about analysts’ private information and make a significant profit by taking abnormal short positions before a downgrade announcement. Boehmer, Jones, and Zhang (2015) also find that short sellers trade more heavily during the week before analyst downgrades, negative forecast revisions, or earnings news. Angel, Christophe, and Ferri (2004) also provide evidence that short-selling activity increases prior to earnings announcements. However, Blau and Wade (2012) find different findings.

5

downgrades and upgrades. Another main difference between the two papers is in the research

methodology, where this paper uses a natural experiment and examines the changes in short-

selling for a given firm, and thereby establishes a causal relationship between short selling and

informational efficiency of stock prices.

Finally, this study provides additional evidence on the effects of short-selling regulation

in financial markets. Recent research examines the effect of Reg SHO on short-selling activity

and market quality (Angel, 1997; Alexander and Peterson, 1999, 2008; Diether et al., 2009),

corporate financing and investment decisions (Grullon et al., 2015), and firm management and

analyst behavior (Angelis et al., 2013; Fang et al., 2015; Ke et al., 2015; Li and Zhang, 2015).

These studies provide mixed results on the effect of Reg SHO on stock prices. Alexander and

Peterson (2008) and Diether et al. (2009) find no changes in market quality or stock prices after

the Reg SHO announcement, while Grullon et al. (2015) show a price decline for pilot stocks

prior to the Reg SHO announcement. 2 The current findings suggest that the Reg SHO

significantly improved price discovery and informational efficiency in financial markets.

The rest of the sections are as follows. Section 2 describes the prior literature and

hypotheses development. Section 3 explains the sample and research design, and Section 4

shows the results from the DiD tests. Section 5 includes additional tests and robustness tests.

Section 6 concludes.

2. Prior Literature and Hypotheses Development

2 Boehmer, Jones, and Zhang (2013) also find that the 2008 shorting ban did not affect stock prices. See also Saffi and Sigurdsson (2010) for international evidence.

6

Prior literature finds mixed evidence on whether short-sellers poses informative private

information. While some studies find that short-sellers anticipate forthcoming bad news, such as

analyst recommendation downgrades (Christophe, Ferri, and Hsieh, 2010), insider sales

(Chakrabarty and Shkilko, 2013), financial misconduct (Karpoff and Lou, 2010), or earnings

announcements (Angel, Christophe, and Ferri, 2004), other studies find no evidence of increased

short-selling prior to bad news events (Daske, Richardson, and Tuna, 2005; Boehmer and Wu,

2013; Engelberg, Reed, and Ringgenberg, 2012; Blau and Pinegar, 2013). The differences can be

contributed to the nature of firm news, as short-sellers can be informed about a particular type of

information but not others. 3

Another reason for the differences in prior findings may be due to the endogenous

relationship between firm performance and short-selling activity. In reviewing the short-selling

literature, Reed (2015) acknowledges that this body of literature faces a significant challenge: the

issue of endogeneity. Prior short-sale constraint proxies can be related to firm characteristics and

stock prices, as well as the information environment of the firm.4 Therefore, it is difficult to

establish direct causality between short-sale constraints and the informational efficiency of stock

prices. This study extends and contributes to this literature in the identifying the exogenous 3 For example, Christophe, Ferri, Hsieh (2010) find evidence of brokerage houses tipping information to short sellers about forthcoming recommendation downgrades. However, short sellers are less likely to acquire private information prior to firm’s earnings announcements, since firm management have less incentives to tip their private information to short sellers. 4 Most empirical studies use either short interest data (see, e.g., Asquith and Meulbroek, 1995; Dechow et al., 2001; Asquith, Pathak, and Ritter, 2005) or shorting flow data (Angel, Christophe, and Ferri, 2004; Boehmer, Jones, and Zhang, 2008; Diether, Lee, and Werner, 2009; Boehmer and Wu, 2013). Seneca (1967), Figlewski (1981), Desai, Thiagarajan, Ramesh, and Balachandran (2002), Senchack and Starks (1993), Aitken, Frino, McCorry and Swan (1998) and others also use short interest as a proxy for short-sale constraints. However, Chen et al. (2002) and D’Avolio (2002) argue that short interest is not an effective measure of short-sale constraints since it might reflect the transaction cost of selling short. Other proxies of short-sale constraints such as institutional holdings (Nagel, 2005) and differences of opinion (Boehmer, Danielsen, and Sorescu, 2006; Deither, Malloy and Scherbina, 2002) can also be correlated with various firm characteristics.

7

changes in short-sale constraints (i.e., Reg SHO), by exploiting a natural experiment.5 I use an

exogenous regulatory shock to short-selling activity brought about by the SEC’s Reg SHO Pilot

Program. On May 2, 2005, the SEC implemented the Regulation SHO program and selected a

pilot group of 986 firms that were suspended from the short-sale price tests. This presents an

exogenously imposed reduction in the cost of short selling and an increase in the prospect of

short selling in these stocks. As a result, short-selling activity increased significantly for pilot

stocks relative to other stocks during the suspension period (Deither et al., 2009; Grullon et al.,

2015; Angelis et al., 2013).

The implementation of Reg SHO relaxes the short-sale constraints of pilot stocks, which

increases short sellers’ trading activities based on their negative private information. Analysts are

important information intermediaries, and their negative news announcements are usually

associated with larger price declines (see, e.g., Womack, 1996; Barber, Lehavy, McNichols, and

Trueman, 2001). Therefore, there is a strong incentive for investors to acquire information about

a forthcoming negative forecast and to profit by short selling the stock prior to the

announcement.6 Moreover, Boehmer, Jones, and Zhang (2015) provide evidence that analysts’

forecast announcement dates and earnings announcement dates are important for individual

short-selling activity. A benefit of using analyst forecast announcements is that I can identify the

type of information (i.e., negative or positive revisions) by comparing the analysts’ forecast with

5 Recent papers use instruments to identify exogenous shocks to short-sale constraints, such as restrictions across countries (Bris, Goetzmann, and Zhu, 2007; Beber and Pagano, 2013), regulatory constraints in Hong Kong (Chang, Cheng, and Yu, 2007), and changes in the supply of lendable shares (Kaplan et al., 2013; Thornock, 2013). However, these studies do not examine the informational efficiency of stock prices, which is the research question of the current study. 6 Short sellers can gather information about a forthcoming analyst announcement based upon fundamental analysis of the targeted firm and analysts concurrently or subsequently analyze a correlated information set. Alternatively, they can receive inside information about a forthcoming negative announcement from brokerage firms. Short sellers may have various information sources; however, this study does not necessarily condition on their information collection channel.

8

their previous forecast. If short sellers are informed about forthcoming analyst forecast, I expect

trading volume to increase only prior to negative revisions and not prior to positive revisions for

pilot stocks during Reg SHO (Hypothesis 1).

One of the main implications of informed shorting activity is the increase in the

informational efficiency of stock prices (see, e.g., Miller 1977; Diamond and Verrecchia, 1987).

The main prediction of Diamond and Verrecchia (1987) is that under short-sale constraints, stock

prices adjust more slowly to negative private information than they do to positive private

information. This is because short-sale constraints lead to a decrease in trading activity, and

some informative short-selling orders are removed from the market. Accordingly, an increase in

short-selling activity increases the speed of adjustment to negative private information.

Prior empirical evidence largely supports the theoretical view that short-sale constraints

hinder price discovery. Reed (2007) tests the DV hypothesis and finds that the stock price

sensitivity to earnings announcements are larger for stocks with high rebate rates. Similarly,

Jennings and Starks (1986) and Skinner (1990) find that the informational efficiency of stocks is

different for stocks with options than for stocks without options. Boehmer and Wu (2013)

provide evidence that the shorting flow improves the incorporation of public information into

prices and reduces the post-earnings announcement drift. 7 In comparison, this study examines

the changes in stock price efficiency to private information, which differs from past studies that

examine price efficiency to public information.

7 Karpoff and Lou (2010) show that short sellers improve informational efficiency by identifying financial misconduct. Ke et al. (2015) provide evidence that short-selling activity reduces analysts’ optimistic bias. Fang et al. (2015) find that increased short selling deters firms’ earnings management. They also show that the post-earnings announcement drift after negative earnings news is smaller for pilot stocks. On the other hand, Lasser, Wang, Zhang (2010) find that short interest is not associated with price efficiency.

9

Building on these prior findings, this study compares the changes in the speed of

adjustment to analysts’ forecast revisions of pilot stocks (relative to non-pilot stocks) during Reg

SHO. Pilot stocks that are easier to short experience greater informed trading during Reg SHO.

This increases the speed at which short sellers’ private information is incorporated into stock

prices, especially regarding negative information. Therefore, I expect the stock price sensitivity

to increase prior to analysts’ negative forecast revisions, for pilot stocks during Reg SHO

(Hypothesis 2). Hypothesis 2 predicts an asymmetric stock price sensitivity to negative private

versus positive private information.

3. Data and Empirical Methodology

3.1. Data Sample

The sample of 986 pilot stocks and 1,966 non-pilot stocks are identified based on the

SEC’s methodology as described in Fang et al. (2015). The list of 986 stocks that would trade

without being subject to any price tests during the term of the pilot program is based on the 2004

Russell 3000 index, excluding stocks not listed on NYSE, AMEX, or Nasdaq NM.8 Stocks are

sorted by their average daily dollar volume computed from June 2003 through May 2004 within

each of the three listing markets, and every third stock within each listing market is designated as

a pilot stock.

As the SEC announced the list of pilot stocks on July 28, 2004 and initiated the pilot

program on May 2, 2005, I exclude quarters between these dates (Ke et al., 2015). I use eight

quarters before the announcement date as the pre-period and eight quarters after the program 8 I thank Fang, Huang, and Karpoff (2015) for sharing the list of pilot and non-pilot stocks.

10

start date as the during-period to test the effect of short selling on the informational efficiency of

stock prices. The pre-period is from 2003:Q3 to 2004:Q2, and the during-period is from 2005:Q3

to 2007:Q2. I also delete firms in the financial services (SIC 6000-6999) and utilities industries

(SIC 4900-4949) (Fang et al., 2015).

I then merge this sample with analyst forecast data from I/B/E/S. I use annual earnings

forecasts that are one-year-ahead forecasts, and use the unadjusted file to mitigate the rounding

problem in I/B/E/S (see, for instance, Diether, Malloy, and Scherbina, 2002). Using the split-

adjustment factors from I/B/E/S, I adjust the unadjusted forecast so that it is on the same per-

share basis as the unadjusted actual earnings and retain only the forecast revision closest to July

(but not after July) in a particular year (see, e.g., Hong and Kubik, 2003).9 I obtain data on stock

returns and trading volume from the Center for Research in Security Prices (CRSP). Firm-level

variables are obtained from Compustat Annual Updates, and institutional holdings data is from

the Thomson Reuters Spectrum database. Spectrum collects quarterly data on stock holdings

from the 13F reports that institutions are required to file if their holdings exceed $100 million.

The holdings are aggregated over all institutions to arrive at the total institutional holdings. In

most tests, I require all firms and analysts to have data to calculate firm or analyst characteristics

across the entire sample period. The resulting sample includes 35,012 analyst forecast

announcements from 1490 firms, of which 497 firms are pilot stocks and 993 firms are non-pilot

stocks.

3.2. Key Variables and Research Design

9 I use the most recent forecasts before the cut-off date of July to use a common time frame to compare the information contained in analysts’ forecasts (see, e.g., Crichfield et al., 1978; Hong and Kubik, 2003). The results are robust to alternative cut-off dates. Following Hong and Kubik (2003), I retain only firms with fiscal year ending in December.

11

I create an indicator variable PILOT to denote firms that are pilot stocks. Specifically,

PILOT equals one if a firm’s stock is designated as a pilot stock under Regulation SHO’s pilot

program, and zero otherwise. Pilot firms constitute the treatment group, and non-pilot firms serve

as the control group. I also construct an indicator variable to indicate time periods during Reg

SHO. DURING equals one if the forecast revision is made within the eight-quarter period after

the adoption date of Reg SHO, and zero otherwise. REV is the analyst forecast revision, which is

the difference between analyst i's forecast for firm j in year k and the analyst's prior forecast for

the same firm-year, scaled by the absolute value of the latter.

This study examines short selling and the informational efficiency of stock prices to

private information by utilizing analysts’ forecast announcements. Reed (2007) assumes firms’

earnings information to be private until firms announce their earnings to the public. Similarly,

this study uses analysts’ forecasts as a setting to identify the date when private information

becomes public. To compare short selling and the stock price sensitivity prior to an analyst’s

forecast revision, I first identify each analyst forecast revision date and then measure the trading

volume and the stock price response during the prior 5-day announcement window ([-5, -1]

days).10 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖 is the 5-day cumulative trading volume prior to analyst i's forecast revision for

firm j in year k, in natural logarithm. The first main regression specification is as follows:

𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖 = 𝑏𝑏0 + 𝑏𝑏1(𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 ∗ 𝐷𝐷𝐷𝐷𝐷𝐷𝑃𝑃𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖) + 𝑏𝑏2𝐷𝐷𝐷𝐷𝐷𝐷𝑃𝑃𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖 + 𝑏𝑏3𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖

+ 𝑏𝑏4𝐷𝐷𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖𝑖𝑖 + 𝑏𝑏5𝐸𝐸𝐸𝐸𝐸𝐸.𝑃𝑃𝑇𝑇𝑇𝑇𝑇𝑇𝑅𝑅𝑖𝑖𝑖𝑖 + 𝑏𝑏6𝑆𝑆𝐸𝐸𝑇𝑇𝑅𝑅𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖 + 𝑏𝑏7𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖 ,𝑖𝑖−1 + 𝑏𝑏8𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑖𝑖−1

+ 𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖 (1)

10 The prior window of 5 days follows from Christophe et al. (2010). The findings are robust to alternate window lengths of 3 days and 10 days. The post announcement window results are also materially similar when I examine a 3-day post window.

12

The main explanatory variable of interest is PILOT*DURING, whose coefficient (𝑏𝑏1) measures

the DiD effect, which is the change in trading volume (relative to non-pilot stocks) of pilot

stocks during the adoption of Reg SHO. If short-seller’s trading increases (decreases) during Reg

SHO, I expect 𝑏𝑏1 to be positive (negative).

ExpTrade is the average trading volume 10 days prior ([-10,-6]) to the revision

announcement to control for the average trading activity of the firm (in natural logarithm), and

Spread is the average bid-ask spread 10 days prior ([-10,-6]) to the revision announcement. Here

𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑖𝑖−1 is a vector of analyst characteristics measured in year k−1 and serving as control

variables; 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑖𝑖−1 is a vector of stock-level control variables measured in year k−1.

Analyst characteristics include experience (number of years since first forecast issuance reported

in I/B/E/S, in natural logarithm), All Star status (an indicator variable that equals one if the

analyst is identified as an All Star by the All American Institutional Investor magazine), horizon

(the number of days from forecast issue date to actual earnings announcement date, in natural

logarithm), coverage (number of firms covered by the analyst, in natural logarithm), and

brokerage size (the number of analysts employed by the brokerage firm, in natural logarithm).

Stock characteristics include firm size, market-to-book ratio, cash flow volatility, and

institutional investor holdings. Heteroskedasticity and autocorrelation consistent (HAC) standard

errors are clustered by analyst-firm.

Next, I estimate the stock price sensitivity prior to forecast revisions. 𝑃𝑃𝐶𝐶𝑃𝑃𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖 is the 5-

day cumulative abnormal return prior to analyst i's forecast revision for firm j in year k.

The second regression specification is as follows:

13

𝑃𝑃𝐶𝐶𝑃𝑃𝐷𝐷𝑖𝑖𝑖𝑖𝑖𝑖 = 𝑏𝑏0 + 𝑏𝑏1(𝐷𝐷𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖𝑖𝑖 ∗ 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 ∗ 𝐷𝐷𝐷𝐷𝐷𝐷𝑃𝑃𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖) + 𝑏𝑏2(𝐷𝐷𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖𝑖𝑖 ∗ 𝐷𝐷𝐷𝐷𝐷𝐷𝑃𝑃𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖) + 𝑏𝑏3(𝐷𝐷𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖𝑖𝑖 ∗ 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖)

+ 𝑏𝑏4(𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 ∗ 𝐷𝐷𝐷𝐷𝐷𝐷𝑃𝑃𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖) + 𝑏𝑏5𝐷𝐷𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖𝑖𝑖 + 𝑏𝑏6𝐷𝐷𝐷𝐷𝐷𝐷𝑃𝑃𝐷𝐷𝐷𝐷𝑖𝑖𝑖𝑖 + 𝑏𝑏7𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 + 𝑏𝑏8𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑖𝑖−1

+ 𝑏𝑏9𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑖𝑖−1 + 𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖 (2)

The explanatory variable of interest is the interaction effect between PILOT*DURING and the

magnitude of the forecast revision, REV. REV*PILOT*DURING is the interaction term whose

coefficient (𝑏𝑏1) measures the DiD effect, which is the change in stock price sensitivity to forecast

revisions (relative to non-pilot stocks) of pilot stocks during the adoption of Reg SHO. If the

speed of adjustment to information increases (decreases) during Reg SHO, I expect 𝑏𝑏1 to be

positive (negative). Other analyst and stock characteristics variables follow that in equation (1).

4. Empirical Findings

4.1. Descriptive Statistics

Table 1, Panel A reports summary statistics on trading volume, stock price sensitivity and

other key variables used throughout the study. The mean (median) PCAR in the sample is 0.224

(0.088), while the mean (median) PVOL is 15.384 (15.401). While the mean forecast revision

(REV) is slightly less than zero (-0.0003), the median revision is positive (0.009). Panel B

describes the differences in the main variables between pilot and non-pilot stocks. The findings

show a significant difference between pilot and non-pilot stocks in their cumulative abnormal

returns (PCAR), trading volume (PVOL), and forecast revisions (REV). PCAR is lower for pilot

stocks than for non-pilot stocks, while trading activity (PVOL) is higher for pilot stocks. The

differences in the variables are highly significant at the 1% level. The average REV for pilot

stocks is negative, while it is positive for non-pilot stocks. This finding is consistent with that of

14

Ke et al. (2015), who find that analysts issue less optimistically biased forecasts for pilot stocks

when there is more short-selling activity. Overall, the univariate results suggest that there are

significant differences in the stock prices and trading activity, between pilot and non-pilot stocks,

around analyst forecast announcements. In the next section, I proceed with a multivariate DiD

regression analysis to further examine the effect of Reg SHO on stock prices.

4.2. Main Multivariate Analysis

4.2.1. Trading Volume DiD Tests

I first examine whether Reg SHO impacts the trading activity of pilot stocks. If short-

selling activity increased after the implementation of Reg SHO (Diether, Lee, Werner, 2009),

then I expect trading volume to increase for pilot stocks relative to non-pilot stocks. Moreover, if

short sellers are informed about the forthcoming type of forecast news, the increase in trading

volume from Reg SHO would be concentrated around negative forecast revisions, since short

sellers trade on their negative private information. In contrast, there would be no changes in

trading volume prior to positive forecast revisions.

To examine the changes in trading volume of pilot stocks during Reg SHO, I estimate the

DiD regression in equation (1). A positive coefficient of PILOT*DURING can be interpreted as

an increase in trading volume for pilot stocks (relative to non-pilot stocks) during Reg SHO.

Table 3 presents the regression results for estimating equation (1), which tests for any significant

changes in the trading activity of pilot stocks during Reg SHO. The dependent variable is PVOL,

which is cumulative trading volume during the 5-day period prior to the analyst forecast

announcement. Columns (1)-(3) include negative revisions and columns (4)-(6) include positive

15

revisions. Columns (1) and (4) present parsimonious specifications including only the forecast

revision, expected trading volume, spread, and forecasting horizon as control variables. In

columns (2) and (5), I include analyst characteristics that may affect the informativeness of the

forecast revision. Columns (3) and (6) include stock characteristics that may also affect trading

activity.

I find that trading volume increases only prior to negative revisions and not prior to

positive revisions. In column (1), the coefficient of PILOT*DURING is positive and significant

at the 5% level. This suggests the trading volume of pilot stocks increased from easier short-

selling during Reg SHO. Since short sellers have informative private information, I only find

higher levels of trading volume prior to negative revision announcements. The coefficient of

PILOT*DURING is even more significant at the 1% level in columns (2)-(3), when I control for

analyst and stock characteristics. In contrast, the coefficient of PILOT*DURING is close to zero

and insignificant in columns (4)-(6), when I examine the changes in trading activity prior to

positive revision announcements. In contrast to the significant changes in trading volume for

negative revisions, there are no differences in trading volume around positive revisions. Overall,

the findings in Table 2 show asymmetric changes in trading activity between negative and

positive revision announcements, which implies that informed short-selling increased for pilot

stocks during Reg SHO.

The magnitude of the coefficient from column (3) indicates that pilot stocks experience

an 2.7% increase in trading volume prior to a negative revision. One might argue that a 2.7%

16

increase in trading volume is not very significant economically.11 However, when I sum the

absolute differences in trading volume between pilot and non-pilot stocks, before and after the

analysts’ announcements, the average difference in trading volume is around 8% (2.7%+4.8%).

This difference is for each analyst announcement, and given the number of analysts issuing

revisions for each firm, the sum of the changes in trading volume can translate into significant

amounts.

4.2.2. Reg SHO and the Stock Price Sensitivity to Analyst Forecast Revisions

The second hypothesis is that informed short selling increases the informational

efficiency of stock prices to negative private information. The decrease in short-sale constraints

from Reg SHO increases short-selling activity, which improves the speed at which stock prices

incorporate negative private information. Consistent with this argument, Reed (2007) finds that

the stock price sensitivity to negative earnings news increases for stocks with short-sale

constraints (i.e., rebate rates). Here, I examine the impact of Reg SHO on the changes in

informational efficiency around analyst forecast revisions, since analysts are important producers

of private information. I estimate the regression model in equation (2) separately for negative

and positive forecast revisions.

Table 3 reports the primary DiD tests of the effect of short selling on the informational

efficiency of stock prices, where I compare the changes in stock price sensitivity to analyst

forecast announcements of pilot stocks during Reg SHO. The dependent variable is PCAR,

which is the cumulative abnormal return five days prior to the announcement. The main

coefficient of interest is REV*PILOT*DURING, as this captures the stock price sensitivity to 11 The magnitude of coefficients in Table 2 is in line with prior findings that examine the changes in trading volume around announcements. For example, Bamber (1986) finds that for a given firm, trading volume increases by 2% during the announcement period relative to a non-announcement period.

17

forecast revisions around the announcement event window. Columns (1)-(3) include negative

revisions and columns (4)-(6) include positive revisions. In column (1), the coefficient of

REV*PILOT*DURING is positive and significant at the 5% level, suggesting that the speed of

adjustment to negative private information increased significantly for pilot stocks after the

implementation of Reg SHO. After I include analyst and firm characteristics as control variables

in columns (2) and (3), the coefficient of REV*PILOT*DURING becomes even larger and is

significant at the 1% level. In comparison, in columns (4)-(6), when observing the stock price

sensitivity to positive revisions, the coefficient of REV*PILOT*DURING is negative and

insignificant, indicating that there are no differences in the stock price sensitivity between pilot

and non-pilot stocks for positive revisions. The results in Table 2 show a clear contrast between

negative and positive revisions, which further strengthens previous findings on informed shorting

activity. In sum, the findings in Table 3 show an asymmetric effect of short selling on the stock

price sensitivity to negative versus positive private information that is consistent with the

changes in trading activity reported in Table 2.

Overall, the results presented in Table 2 are consistent with the main prediction that there

is a causal relationship between short-selling activity and consequent stock price sensitivity prior

to negative information announcements. More generally, these results indicate that short-selling

activity from Reg SHO improves the informational efficiency of stock prices, especially in

response to negative private information. I find no evidence that Reg SHO has an impact on the

stock price sensitivity around positive information announcements.

5. Additional Tests

18

5.1. Permanent Removal of the Uptick Test

On June 13, 2007, the SEC announced the permanent removal of the uptick test for all

stocks, ending the pilot program. This permanent removal became effective starting on July 6,

2007. It removed the short-sale constraint for non-pilot stocks, while the short-sale constraint

remained unchanged for pilot stocks. This is another exogenous shock to short-selling activity

for non-pilot stocks that can be used to test the effect of short selling on the informational

efficiency of stock prices. If short-selling activity increased for non-pilot stocks after the

permanent removal of the uptick test, then I expect the stock price sensitivity to negative private

information to increase for non-pilot stocks and not for pilot stocks after the removal.

To examine the effect of the permanent removal of the uptick test on the informational

efficiency of non-pilot stocks, I reestimate equations (1) and (2), replacing the PILOT indicator

variable with the NONPILOT indicator variable, which equals one if the stock was not selected

as a pilot stock during the initial pilot program. I also replace the DURING indicator variable

with the REMOVAL indicator variable, which equals one if the analyst forecast announcement

was made after the removal of the pilot program (2009:Q1-2009:Q4), and zero for the eight

quarters during the pilot program (2005:Q3-2007:Q2).12

Table 4 reports the DiD test results on trading volume, in which the specifications are

similar to those in Table 2. The coefficient of interest is NONPILOT*REMOVAL, which captures

the changes in trading volume for non-pilot stocks (relative to pilot stocks), after the permanent

removal of the uptick test. As previously noted, hypothesis 1 predicts the coefficient to be

positive for negative private information, and to be insignificant for positive private information.

12 Following Ke et al. (2015), I exclude six quarters from 2007:Q3 to 2008:Q4 to mitigate the potential confounding effects of the financial crisis. The results are materially similar when I include these quarters, as in Ke et al. (2015).

19

The findings support hypothesis 1 and the findings in previous tables. In columns (1)-(3), the

coefficient of NONPILOT*REMOVAL is significantly positive at the 1% level, which indicates

that trading volume prior to negative revisions increased for non-pilot stocks when their short-

sale constraints decreased. In comparison, trading volume decreased prior to positive revisions in

columns (4)-(6), which again supports the notion that short-sellers are informed about

forthcoming analyst revision.

Next, I examine the stock price sensitivity of non-pilot stocks after the permanent

removal of the uptick test. The coefficient of interest is REV*NONPILOT*REMOVAL, which

captures the changes in stock price sensitivity to analyst forecast revisions for non-pilot stocks

(relative to pilot stocks). Table 5 reports the DiD test results in which the specifications are

similar to those in Table 3. Similar to the findings in Table 3, the coefficient of

REV*NONPILOT*REMOVAL is positive and significant in columns (1)-(3), which indicates that

the speed of adjustment to negative private information increased for non-pilot stocks when their

short-sale constraints were relaxed upon the permanent removal of the uptick test. Meanwhile,

the coefficient of REV*NONPILOT*REMOVAL is insignificant in columns (4)-(6). The findings

in Table 5 show significant changes in the informational efficiency of non-pilot stocks before

and after the removal of the uptick test, which reinforces the main findings that short-selling

activity improves the informational efficiency of stock prices, especially in response to negative

private information.

5.2. Cross-sectional Analysis

In this section, I conduct a series of cross-sectional analyses to further identify the effect

of short-selling activity on trading activity and the informativeness of stock prices. I compare

20

firms across different firm size and analyst forecast dispersion, as prior studies relate these two

firm characteristics to their information environments.

Grullon, Michenaud, and Weston (2015) and Li and Zhang (2015) find that the impact of

Reg SHO on short-selling activity is larger for smaller pilot stocks. The informational efficiency

of smaller or younger firms is typically lower than that of larger or older firms, since there is

relatively less information available for smaller or younger firms (Zhang, 2006). As a result,

smaller or younger pilot stocks may experience a larger increase in informational efficiency after

the implementation of Reg SHO. Thus, I expect trading volume and the stock price sensitivity

prior to the revision announcement to increase for smaller pilot stocks.

Prior literature suggests that the effect of short-sale constraints is stronger for firms about

which investors have more differences of opinion. For example, Chang et al. (2007) and

Berkman et al. (2009) find that the effect of short-sale constraints on stock overvaluation is

stronger for stocks about which there are wider differences of opinion. Analyst forecast

dispersion is a frequently used proxy for differences of opinion (Diether et al., 2002), but it can

also measure the informational uncertainty about a firm (Zhang, 2006). Zhang (2006) finds that

firms with high analyst forecast dispersion have less informational efficiency around analysts’

forecast revisions. Accordingly, I also whether the effect of Reg SHO is stronger for pilot stocks

about which there are more differences of opinion or greater information uncertainty (i.e., analyst

forecast dispersion).

Similar to Tables 2 and 3, I estimate the DiD regression using equations (1) and (2),

separately for larger and smaller firms, or firms with high and low dispersion levels. Firms are

sorted into two groups: above or below the median level based on their firm size each year.

21

Similarly, firms are sorted annually into two groups based on the level of forecast dispersion

(standard deviation of analyst forecasts of firm j in year k). The results on trading volume are

presented in Table 6, Panel A, and the results on the stock price sensitivity are presented in Panel

B. In both panels, columns (1)-(2) compare across firm size, and columns (3)-(4) compare across

analyst forecast dispersion.

In Table 6, Panel A, I find that trading volume only increases for small pilot stocks and

stocks with high forecast dispersion. These firms are typically regarded to have high information

uncertainty and more difference in opinions among investors, which makes informed trading

easier. I do not find significant changes in trading volume for large pilot stocks, and trading

volume decreases for low dispersion stocks. Next, I compare the changes in the informational

efficiency across firms. In Panel B, columns (1) and (3), I find that the stock price sensitivity

prior to the revision announcement increases only for small pilot stocks and stocks with high

dispersion. For large pilot stocks, there are no significant differences in the stock price sensitivity

before the announcement. Pilot stocks with less dispersion experience decreased stock price

sensitivity. In sum, the findings in Table 6 show that there are cross-sectional differences among

pilot stocks. Smaller pilot stocks and stocks about which analysts have more differences of

opinion experience a greater increase in informational efficiency, while the same is not true for

the other subset of pilot stocks.

6. Robustness Tests

6.1. Short Selling around Recommendation Downgrades

22

Prior studies provide evidence of an increase in short-selling activity prior to analyst

recommendation downgrades. Christophe, Ferri, and Hsieh (2010) even suggest the likelihood

that short sellers receive tips from brokerage firms about a forthcoming downgrade. Since

analyst downgrades are an alternative output of analysts’ private information, I next compare

trading activity and the informational efficiency of pilot stocks around analysts’ stock

recommendation changes. Individual analyst stock recommendations follow from the coded

I/B/E/S text. 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 is the numeric value of the stock recommendations, where strong buy =1,

buy =2, hold =3, underperform =3, and sell =4. A negative recommendation change, Downgrade,

equals one if the stock recommendation, 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 , is larger than the previous

recommendation, 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖−1, and zero otherwise. Similarly, a positive recommendation change,

Upgrade, equals one if the stock recommendation 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 is smaller than 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖−1, and zero

otherwise. To have a sample consistent with that of the earnings forecast analysis, I include only

stock recommendations issued by analysts who also issue earnings forecasts. I also limit the

stock recommendations to those that are announced closest to July of year k, since

recommendations are even more highly serially correlated than earnings forecasts. I can also

examine the impact of recommendations in a common timeframe (Hong and Kubik, 2003).13

Table 7, Panel A reports the changes in trading activity, and Panel B reports the changes

in the stock price sensitivity prior to analyst stock recommendation announcements. I estimate

the differences in trading volume and stock returns separately for recommendation downgrades

and upgrades. In both panels, columns (1)-(3) include downgrades and columns (4)-(6) include

upgrades. In Panel A, the dependent variable is PVOL and in Panel B, the dependent variable is 13 The number of observations for stock recommendations is smaller than that of earnings forecasts since the frequency of a stock recommendation change is much lower than forecast revisions. For this reason, the sample size of earnings forecast data is more than twice than the size of earnings forecast data in I/B/E/S. I also require my recommendation data to also have earnings forecast characteristics.

23

PCAR. PVOL is the trading volume and PCAR is the cumulative abnormal return around the five

day window [-5, -1] prior to the recommendation announcement. The coefficients of

PILOT*DURING measure the changes in trading volume and stock returns around

recommendation changes of pilot stocks during Reg SHO. In Panel A, the coefficient of

PILOT*DURING is positive in columns (1)-(3), which indicates an increase in trading activity

for pilot stocks prior to downgrade announcements. On the other hand, the coefficient

PILOT*DURING is insignificant in columns (4)-(6), suggesting no changes in trading activity

prior to upgrades. The findings are supportive of the informed short selling hypothesis 1. In

Panel B, columns (1)-(3), I find that the coefficient of PILOT*DURING is negative and

significant at the 1% level, which indicates that the decline in stock prices prior to downgrades

is greater for pilot stocks than for non-pilot stocks during Reg SHO. Increased short-selling

activity results in price declines prior to the announcement. In contrast, the coefficient of

PILOT*DURING is insignificant for upgrades in columns (4)-(6). I do not find any differences

between the stock prices of pilot stocks and non-pilot stocks for upgrades.

6.2. Short Selling around Earnings Announcements

In this section, I reexamine the changes in trading volume and the stock price sensitivity

for pilot stocks around quarterly earnings announcements. Firm earnings announcements are also

important informational events for which the dates can easily be identified. As in the previous

analyses of analysts’ forecast revisions, I consider firms’ earnings to be private information prior

to the announcement (Reed, 2007). The regression model is similar to equations (1) and (2),

except that I replace REV with SURP, which is the earnings surprise variable. SURP is the

difference between actual earnings and the latest mean analyst forecast prior to the earnings

announcement, scaled by the stock price of the previous fiscal year end. The dependent variable

24

is PVOL is Table 8, Panel A, and PCAR in Panel B. 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 is the cumulative trading volume (in

logs), and 𝑃𝑃𝐶𝐶𝑃𝑃𝐷𝐷𝑖𝑖𝑖𝑖 is the 5-day cumulative abnormal return around the 5-day window ([-5, -1])

prior to firm j's earnings announcement in quarter q of year k. Since this is a firm-level analysis, I

exclude analyst characteristics but include stock characteristics. In both panels, columns (1)-(2)

include negative earnings news and columns (3)-(4) include positive earnings news. In Panel A,

the coefficient of interest is the interaction term, PILOT*DURING, which measures the changes

in trading volume of pilot stocks (relative to non-pilot stocks) during Reg SHO. I find that

trading volume of pilot stocks increases for both types of earnings news, which indicates that

short sellers are uninformed about the type of forthcoming earnings news. The findings differ

between analyst announcements and suggest that short sellers have more difficulty in predicting

forthcoming negative earnings news than for analysts’ negative revisions. Blau and Pinegar

(2013) also find that short-sellers are less informed prior to earnings announcements.

Next, I examine the stock price sensitivity in Panel B. The coefficient of interest is the

interaction term, SURP*PILOT*DURING, which measures the changes in stock price sensitivity

of pilot stocks (relative to non-pilot stocks) during Reg SHO. In columns (1)-(2), the coefficient

of SURP*PILOT*DURING is positive and significant at the 1% level, which indicates that the

stock price sensitivity to negative earnings news increased. In contrast, I do not find significant

changes in the stock price sensitivity prior to positive earnings news in columns (3)-(4).

Overall, the findings in Table 8 suggest that short-selling activity increases the

informational efficiency of stock prices to negative earnings news, which is also consistent with

earlier results on the stock price sensitivity around analyst forecast revision announcements.

25

7. Conclusion

This paper examines the effect of short selling on the informational efficiency of stock

prices to private information using a natural experiment–Reg SHO. I find that relaxing the short-

sale constraints increases trading activity and the speed of adjustment of stock prices prior to

negative analyst forecast revisions. Pilot stocks with increased short-selling activity experience

an increase in the stock price sensitivity prior to the negative revision announcement. I do not

find significant differences in trading volume nor the stock price sensitivity prior to positive

revision announcements. These findings are consistent with the Diamond and Verrecchia (1987)

model, which predicts that short-sale constraints reduce the speed at which prices adjust to

private information more for negative private information than for positive information. The

differences in informational efficiency only respect to forthcoming negative revisions supports

the view that short sellers are informed traders who trade on their negative private information.

In the cross-section, I find that the effect of Reg SHO is stronger for pilot stocks that are

smaller and stocks with larger differences of opinion. I also find that the informational efficiency

of pilot stocks increases after the permanent removal of the uptick test and around firm earnings

announcements. The findings are also consistent when I examine the stock price sensitivity

around analyst stock recommendation changes.

This study provides evidence that the short-selling activity brought out by Reg SHO

improved informational efficiency in response to negative private information. The main

departure of the current study from prior literature is that I am able to examine the causal effect

of short sale constraints on price efficiency by exploiting a natural experiment (i.e. Reg SHO). In

addition, the current findings add to the literature on the positive externalities of short-selling

26

activity, and on the effect of regulatory changes in financial markets. It also provides empirical

evidence consistent with the theoretical predictions of Diamond and Verreccia (1987) that short

sellers are important information intermediaries whose trading activity increases the

informational efficiency of stock prices to private information.

27

References

Aitken, M., A. Frino, M. McCorry, and P. Swan, 1998, Short sales are almost instantaneously bad news: Evidence from the Australian Stock Exchange, The Journal of Finance 53, 2205-2223.

Alexander, G., and M. Peterson, 1999, Short selling on the New York Stock Exchange and the effects of the uptick rule, Journal of Financial Intermediation 8, 90-116.

Alexander, G., and M. Peterson, 2008, The effect of price tests on trader behavior and market quality: An analysis of Reg SHO, Journal of Financial Markets 11, 84-111.

Angel, J., 1997, Tick size, share prices, and stock splits, The Journal of Finance 52, 655-681.

Angel, J., S. Christophe, and M. Ferri, 2004, Short-selling prior to earnings announcements, Journal of Finance 59, 1845-1875.

Angelis, D., G. Grullon, and S. Michenaud, 2013, Downside risk and the design of CEO incentives: Evidence from a natural experiment, Working paper, Rice University.

Asquith, P., and L. Meulbroek, 1996, An empirical investigation of short interest, Working paper, Harvard Business School.

Asquith, P., P. Pathak, and J. Ritter, 2005, Short interest, institutional ownership and stock returns, Journal of Financial Economics 78, 243-276.

Bamber, L. S., 1986, The information content of annual earnings releases: A trading volume approach, Journal of Accounting Research 24(1), 40-56. Barber, B., R. Lehavy, M. McNichols, and B. Trueman, 2001, Can investors profit from the prophets? Security analyst recommendations and stock returns, The Journal of Finance 56(2), 531-563.

Beber, A., and M. Pagano, 2013, Short-selling bans around the world: Evidence from the 2007-09 crisis, The Journal of Finance, 68(1), 343-381.

Berkman, H., V. Dimitrov, P. Jain, P. Koch, and S. Tice, 2009, Sell in the news: Differences of opinion, short-sales constraints, and returns around earnings announcements, Journal of Financial Economics 92, 376-399.

Blau, B., and J. M. Pinegar, 2013, Are short-sellers incrementally informed prior to earnings announcements?, Journal of Empirical Finance 21, 142-155.

Blau, B., and C. Wade, 2012, Informed or speculative: Short selling analyst recommendations, Journal of Banking and Finance 36, 14-25.

Boehme, R., B. Danielsen, and S. Sorescu, 2006, Short-sale constraints, differences of opinion, and overvaluation, Journal of Financial and Quantitative Analysis 41, 455-487.

Boehmer, E., C. Jones, and X. Zhang, 2008, Which shorts are informed? The Journal of Finance 63, 491-527.

28

Boehmer, E., C. Jones, and X. Zhang, 2013, Shackling short sellers: The 2008 shorting ban, Review of Financial Studies 26, 1363-1400.

Boehmer, E., C. Jones, and X. Zhang, 2015, What do short sellers know? Available at SSRN 2192958.

Boehmer, E., and J. Wu, 2013, Short selling and the price discovery process, Review of Financial Studies 26, 287-322.

Bris, A., W. Goetzmann, and N. Zhu, 2007, Efficiency and the bear: Short sales and markets around the world, Journal of Finance 62, 1029-1079.

Chang, E., J. Cheng, and Y. Yu, 2007, Short-sales constraints and price discovery: Evidence from the Hong Kong market, Journal of Finance 62, 2097-2121.

Chakrabarty, B., and A. Shkilko, 2013, Information transfers in financial markets: Evidence from short selling around insider sales, Journal of Banking and Finance 37, 1560-1572.

Chen, J., H. Hong, and J. Stein, 2002, Breadth of ownership and stock returns, Journal of Financial Economics 66, 171-205.

Christophe, S., M. Ferri, and J. Hsieh, 2010, Informed trading before analyst downgrades: Evidence from short sellers, Journal of Financial Economics 95, 85-106.

Cohen, L., K. Diether, and C. Malloy, 2007, Supply and demand shifts in the shorting market, Journal of Finance 62, 2061-2096.

Conrad, J., 1989, The price effect of option introduction, Journal of Finance 44, 487-498.

Crichfield, T., T. Dyckman, and J. Lakonishok, 1978, An evaluation of security analysts' forecasts, Accounting Review 651-668. Daske, H., S. Richardson, and I. Tuna, 2005, Do short sale transactions precede bad news events?, Working paper SSRN 722242.

D’Avolio, G., 2002, The market for borrowing stock, Journal of Financial Economics 66, 271-306.

Dechow P., A. Hutton, L. Meulbroek, R. Sloan, 2001, Short-sellers, fundamental analysis, and stock returns, Journal of Financial Economics 61, 77-106.

Desai, H., S. Krishnamurthy, and K. Venkataraman, 2006, Do short sellers target firms with poor earnings quality? Evidence from earnings restatements, Review of Accounting Studies 11, 71-90. Desai, H., K. Ramesh, S. Thiagarajan, and B. Balachandran, 2002, An investigation of the informational role of short interest in the Nasdaq market, The Journal of Finance 57, 2263-2287.

Diamond, D., and R. Verrecchia, 1987, Constraints on short-selling and asset price adjustment to private information, Journal of Financial Economics 18, 277-311.

29

Diether, K., K. Lee, and I. Werner, 2009, It’s SHO time! Short-sale price tests and market quality, Journal of Finance 64, 37-73.

Diether, K., C. Malloy, and A. Scherbina, 2002, Differences of opinion and the cross-section of stock returns, Journal of Finance 57, 2113-2141.

Engelberg, J., A. Reed, and M. Ringgenberg, 2012, How are shorts informed? Short sellers, news, and information processing, Journal of Financial Economics 105, 260-278.

Fang, V., A. Huang, and J. Karpoff, 2015, Short selling and earnings management: A controlled experiment, Journal of Finance 71, 1251-1294.

Figlewski, S., 1981, The informational effects of restrictions on short sales: Some empirical evidence, Journal of Financial and Quantitative Analysis 4, 463-476.

Francis, J., R. LaFond, P. Olsson, and K. Schipper, 2005, The market pricing of accruals quality, Journal of Accounting and Economics 39, 295-327.

Grullon, G., S. Michenaud, and J. Weston, 2015, The real effects of short-selling constraints, Review of Financial Studies 28, 1737-1767. Hong, H., and J. Kubik, 2003, Analyzing the analysts: Career concerns and biased earnings forecasts, Journal of Finance 58, 313-351.

Jennings, R., and L. Starks, 1986, Earnings announcements, stock price adjustment, and the existence of option markets, Journal of Finance 41, 107-125.

Kaplan, S. N., T. J. Moskowitz, and B. A. Sensoy, 2013, The effects of stock lending on security prices: An experiment. The Journal of Finance 68, 1891-1936. Karpoff, J. M., and X. Lou, 2010, Short sellers and financial misconduct, The Journal of Finance 65, 1879-1913.

Ke, Y., L. Kin, J. Sheng, and J. Zhang, 2015, Does short selling mitigate optimism in financial analyst forecast? Evidence from a randomized experiment, Working paper, University of British Columbia.

Lasser, D., X. Wang, and Y. Zhang, 2010, The effect of short selling on market reactions to earnings announcements, Contemporary Accounting Research 27, 609-638.

Li, Y., and L. Zhang, 2015, Short selling pressure, stock price behavior, and management forecast precision: Evidence from a natural experiment, Journal of Accounting Research 53, 79-117.

Miller, E., 1977, Risk, uncertainty and divergence of opinion, Journal of Finance 32, 1151-1168.

Nagel, S., 2005, Short sales, institutional investors and the cross-section of stock returns, Journal of Financial Economics 78, 277-309.

30

Reed, A., 2007, Costly short selling and stock price adjustment to earnings announcements, Working paper, University of North Carolina.

Reed, A., 2015, Connecting supply, short-sellers and stock returns: Research challenges, Journal of Accounting and Economics 60, 97-103.

Saffi, P., and K. Sigurdsson, 2010, Price efficiency and short selling, Review of Financial Studies 24, 821-852.

Securities and Exchange Commission (SEC), 2007, Economic analysis of the short sale price restrictions under the regulation SHO pilot, Office of Economic Analysis.

Senchack, A., and L. Starks, 1993, Short-sale restrictions and market reaction to short-interest announcements, Journal of Financial and Quantitative Analysis 28, 177-194.

Seneca, J., 1967, Short interest: Bearish or bullish? The Journal of Finance 22, 67-70.

Skinner, D., 1990, Options markets and the information content of accounting earnings releases, Journal of Accounting and Economics 13, 191-211.

Thornock, J., 2013, The effects of dividend taxation on short selling and market quality, The Accounting Review 88, 1833-1856.

Womack, K., 1996, Do brokerage analysts' recommendations have investment value? The Journal of Finance 51, 137-167.

Zhang, F., 2006, Information uncertainty and stock returns, The Journal of Finance 61, 105-137.

31

TABLE 1 Descriptive Statistics Panel A: Summary Statistics of Key Variables

Panel A reports the descriptive statistics for the main variables. The sample period is from 2003:Q2 to 2007:Q2. PCAR is the 5-day cumulative abnormal return [-5, -1] prior to analyst i's forecast revision for firm j in year k. CAR is the 2-day cumulative abnormal return [0, 1] after analyst i's forecast revision for firm j in year k. PVOL is the 5-day cumulative trading volume [-5, -1] prior to analyst i's forecast revision for firm j in year k (in logs). VOL is the 2-day cumulative trading volume [0, 1] after analyst i's forecast revision for firm j in year k (in logs). Analyst forecast revision, Rev, is the difference between analyst i's forecast for firm j in year k and the analyst's prior forecast for the same firm-year, scaled by the absolute value of the latter. Horizon is the number of days from the forecast to the actual announcement date (in logs). Experience is the natural logarithm of the number of years analyst i issues a forecast for a firm, averaged across the firms the analyst covers in year k. All Star is an indicator variable that equals one if the analyst is included in the All Star analyst list by the Institutional Investors magazine in year k. Coverage is the number of firms covered by analyst i in year k. Brokerage Size is measured by the log of the number of analysts in a given brokerage firm in year k. Size is the natural logarithm of the market value of equity for firm j in year k. Market/Book is the market-to-book ratio of firm j in year k calculated as the market value of the firm’s equity at the end of year k plus the difference between the book value of the firm’s assets and the book value of the firm’s equity in year k, divided by the book value of firm j’s assets in year k. Cashflow volatility is the standard deviation of cash flow from operations in the past 5 years (with a minimum of 3 years) of firm j in year k, where cash flow from operating activity is earnings before extraordinary items minus total accruals, scaled by average total assets. Institutional Holdings is the percentage of institutional investor holdings in firm j at year k.

Mean Stddev Median

Stock Price Reaction Variables

PCAR 0.224 3.707 0.088 PVOL 15.384 1.507 15.401 REV -0.0003 .097 0.009 Analyst Characteristics Horizon 5.438 0.137 5.488 Experience 1.932 0.568 1.945 All Star 0.182 0.386 0 Coverage 2.704 0.529 2.772 Brokerage Size 3.700 1.069 3.828

Firm Characteristics

Size 8.041 1.610 7.911 Market/Book

2.077 1.378 1.608

Cashflow Volatility

0.099 0.126 0.065 Institutional Holdings

0.750 0.208 0.775

32

Panel B: Test for Differences between Pilot and Non-Pilot Stocks

Panel B describes the differences in stock price reaction variables between pilot and non-pilot stocks. Variables are defined in Panel A.

Pilot Non-pilot Test Statistic P-value

(1) (2) (1)-(2) (1)-(2)

Stock Price Reaction Variables

PCAR 0.091 0.247 2.899 0.004 PVOL 15.579 15.350 10.489 <0.001 REV -0.004 0.0001 2.736 0.006

33

TABLE 2 Short Selling and Trading Activity around Analyst Forecast Revisions

Table 2 compares trading activity around analysts' forecast revisions for pilot stocks during Reg SHO. The dependent variable is PVOL, which is the 5-day cumulative trading volume [-5, -1] prior to analyst i's forecast revision for firm j in year k (in logs). Analyst forecast revision, Rev, is the difference between analyst i's forecast for firm j in year k and the analyst's prior forecast for the same firm-year, scaled by the absolute value of the latter. Exp Trade is the average trading volume of firm j 5 days [-10, -6] prior to the announcement. Spread is the bid-ask spread of firm j in the month prior to the announcement. PILOT is an indicator variable that equals one if the stock was selected into the pilot program by the SEC, and zero otherwise. DURING is an indicator variable that equals one for the Reg SHO periods, and zero for the pre-Reg SHO periods. All other variables follow from the previous tables. HAC standard errors are clustered by firm-analyst and reported in parentheses. ***, **, and * denote significance levels of 1%, 5%, and 10%, respectively.

34

Dependent Variable PVOL

Negative Revisions Positive Revisions

(1) (2) (3) (4) (5) (6)

PILOT*DURING 0.023** 0.026*** 0.027*** 0.003 -0.001 0.004

(0.012) (0.010) (0.009) (0.006) (0.004) (0.003)

REV 0.001 0.006 -0.051*** 0.100*** 0.100*** 0.190***

(0.009) (0.008) (0.011) (0.008) (0.009) (0.012)

PILOT -0.010 -0.006 -0.007 -0.010*** -0.010*** -0.008***

(0.008) (0.009) (0.007) (0.003) (0.003) (0.002)

DURING -0.012 -0.008 -0.026** 0.039*** 0.042*** 0.035*** (0.010) (0.009) (0.012) (0.010) (0.010) (0.009) Exp Trade 0.949*** 0.950*** 0.873*** 0.950*** 0.950*** 0.888***

(0.003) (0.003) (0.013) (0.003) (0.003) (0.003)

Spread -0.113*** -0.106*** -0.212*** -0.167*** -0.164*** -0.253***

(0.027) (0.028) (0.024) (0.038) (0.038) (0.035)

Horizon 0.093** 0.090** 0.072** 0.037 0.038 -0.025

(0.044) (0.040) (0.033) (0.027) (0.034) (0.036)

Experience 0.002 -0.006* 0.001 -0.006

(0.004) (0.003) (0.005) (0.004)

All Star -0.004 -0.015*** -0.017*** -0.027***

(0.005) (0.001) (0.004) (0.004)

Coverage -0.022** -0.015 -0.025*** -0.020*** (0.011) (0.012) (0.007) (0.007) Brokerage Size

-0.000*** -0.000*** -0.000*** -0.000***

(0.000) (0.000) (0.000) (0.000) Size

0.092*** 0.072***

(0.013) (0.004) Market/Book

0.021*** 0.008***

(0.007) (0.003) Cashflow Volatility

0.300*** 0.309***

(0.018) (0.031) Institutional Holdings

0.210*** 0.099***

(0.010) (0.013) Constant 1.949*** 2.018*** 2.214*** 2.231*** 2.294*** 2.785*** (0.218) (0.212) (0.194) (0.113) (0.155) (0.180) N 16356 15282 15282 21213 19699 19699 R-sq 0.908 0.907 0.911 0.920 0.921 0.923

35

TABLE 3 Short Selling and the Informational Efficiency of Stock Prices Table 3 reports the results of the multivariate difference-in-differences (DiD) tests that examine the effect of Reg SHO on the informational efficiency of pilot stocks. This table reports the stock price sensitivity to negative and positive forecast revisions. The dependent variable is the stock price response, PCAR, which is the 5-day cumulative abnormal return [-5, -1] prior to analyst i's forecast revision for firm j in year k. Analyst forecast revision, Rev, is the difference between analyst i's forecast for firm j in year k and the analyst's prior forecast for the same firm-year, scaled by the absolute value of the latter. PILOT is an indicator variable that equals one if the stock was selected into the pilot program by the SEC, and zero otherwise. DURING is an indicator variable that equals one for the Reg SHO periods, and zero for the pre-Reg SHO periods. All other variables follow from the previous tables. HAC standard errors are clustered by firm-analyst and reported in parentheses. ***, **, and * denote significance levels of 1%, 5%, and 10%, respectively. Dependent Variable PCAR

Negative Revisions Positive Revisions

(1) (2) (3) (4) (5) (6) REV*PILOT*DURING 3.113** 3.298*** 3.462*** -0.876 -1.323 -1.143 (1.369) (1.155) (1.151) (1.008) (0.929) (0.996) REV*DURING 0.029 -0.310 -0.374 2.257 2.835* 2.600 (2.058) (1.962) (1.880) (1.557) (1.586) (1.613) REV*PILOT -2.136* -2.421*** -2.332*** 0.569 1.153 0.985

(1.248) (0.916) (0.814) (0.747) (0.887) (1.008) PILOT*DURING 0.193 0.199 0.184 0.130 0.208** 0.184*

(0.170) (0.166) (0.165) (0.104) (0.100) (0.110) REV 1.855 2.033 1.867 1.798 1.568 1.359

(2.021) (1.854) (1.433) (1.176) (1.090) (0.830) PILOT -0.261** -0.305*** -0.274*** -0.019 -0.082 -0.055

(0.131) (0.102) (0.089) (0.085) (0.089) (0.100) DURING -0.174** -0.182*** -0.198* -0.249*** -0.278*** -0.288*** (0.068) (0.058) (0.104) (0.017) (0.023) (0.048) Horizon -0.806* -0.765 -0.688 -1.252** -1.251** -1.197** (0.477) (0.497) (0.555) (0.546) (0.568) (0.578) Experience

-0.105*** -0.123*** 0.147** 0.142***

(0.025) (0.030) (0.064) (0.046) All Star

-0.284*** -0.331*** 0.042 0.053

(0.083) (0.067) (0.079) (0.048)

Coverage -0.186*** -0.178** -0.049 -0.053 (0.060) (0.076) (0.038) (0.049) Brokerage Size

0.002** 0.001*** -0.001** -0.001

(0.001) (0.001) (0.000) (0.000) Size

0.099* -0.010

(0.051) (0.081) Market/Book

-0.184*** -0.055*

(0.045) (0.032) Cashflow Volatility

1.053* 0.378

(0.618) (0.357) Institutional Holdings

0.315 0.662**

(0.469) (0.287) Constant 4.010 4.423 3.294 7.392** 7.272** 6.680* (2.585) (2.806) (3.718) (2.974) (3.258) (3.599) N 16377 15301 15301 21226 19711 19711 R-sq 0.002 0.004 0.007 0.003 0.004 0.005

36

TABLE 4 Trading Activity around the Permanent Removal of the Uptick Test

Table 4 reports the results of the multivariate difference-in-differences (DiD) tests that show how the permanent removal of the uptick test affected the trading volume of non-pilot stocks. The dependent variable is PVOL, which is the 5-day cumulative trading volume [-5, -1] prior to analyst i's forecast revision for firm j in year k (in logs). NONPILOT is an indicator variable that equals one if the stock was not selected into the pilot program by the SEC, and zero otherwise. REMOVAL is an indicator variable that equals one if the revision is announced during 2009:Q1 to 2009:Q4, and zero if it is announced during 2005:Q3 to 2007:Q2. Analyst forecast revision, Rev, is the difference between analyst i's forecast for firm j in year k and the analyst's prior forecast for the same firm-year, scaled by the absolute value of the latter. Exp Trade is the average trading volume of firm j 5 days [-10, -6] prior to the announcement. Spread is the bid-ask spread of firm j in the month prior to the announcement. All other variables follow from the previous tables. HAC standard errors are clustered by firm-analyst and reported in parentheses. ***, **, and * denote significance levels of 1%, 5%, and 10%, respectively.

37

Dependent Variable PVOL

Negative Revisions Positive Revisions

(1) (2) (3) (4) (5) (6)

PILOT*DURING 0.040*** 0.035*** 0.037*** -0.029*** -0.036*** -0.036***

(0.013) (0.011) (0.011) (0.007) (0.005) (0.005)

REV -0.003 0.000 -0.041*** 0.077*** 0.084*** 0.139***

(0.010) (0.008) (0.011) (0.018) (0.017) (0.040)

PILOT -0.012 -0.008 -0.015 0.007 0.011** 0.005

(0.013) (0.011) (0.011) (0.007) (0.005) (0.005)

DURING -0.018*** -0.015*** 0.038 -0.001 0.002 0.044*** (0.006) (0.005) (0.027) (0.005) (0.006) (0.016) Exp Trade 0.953*** 0.954*** 0.893*** 0.957*** 0.959*** 0.917***

(0.009) (0.007) (0.031) (0.001) (0.001) (0.016)

Spread -0.002 0.014 -0.140*** -0.148*** -0.139** -0.245***

(0.029) (0.040) (0.033) (0.056) (0.057) (0.058)

Horizon 0.169** 0.130* 0.041 0.004

(0.074) (0.067) (0.041) (0.063)

Experience 0.001 -0.002 0.013*** 0.008***

(0.005) (0.004) (0.000) (0.003)

All Star 0.010 -0.001 -0.008 -0.019

(0.007) (0.012) (0.017) (0.017)

Coverage -0.030*** -0.030*** -0.037*** -0.034*** (0.002) (0.002) (0.008) (0.007) Brokerage Size

-0.000*** -0.000*** -0.000** -0.000***

(0.000) (0.000) (0.000) (0.000) Size

0.078*** 0.051***

(0.028) (0.015) Market/Book

0.002 -0.003

(0.007) (0.005) Cashflow Volatility

0.161 0.140***

(0.110) (0.050) Institutional Holdings

0.142*** 0.083**

(0.049) (0.042) Constant 2.397*** 1.560*** 1.872*** 2.369*** 2.206*** 2.493*** (0.129) (0.400) (0.409) (0.006) (0.257) (0.400) N 11864 11092 11092 13206 12344 12344 R-sq 0.928 0.928 0.930 0.934 0.934 0.935

38

TABLE 5 Changes in Short Selling and Informational Efficiency around the Permanent Removal of the Uptick Test

Table 5 reports the results of the multivariate difference-in-differences (DiD) tests that show how the permanent removal of the uptick test affected the informational efficiency of non-pilot stocks. The dependent variable is the stock price response, PCAR, which is the 5-day cumulative abnormal return [-5, -1] prior to analyst i's forecast revision for firm j in year k. NONPILOT is an indicator variable that equals one if the stock was not selected into the pilot program by the SEC, and zero otherwise. REMOVAL is an indicator variable that equals one if the revision is announced during 2009:Q1 to 2009:Q4, and zero if it is announced during 2005:Q3 to 2007:Q2. All other variables follow from the previous tables. HAC standard errors are clustered by firm-analyst and reported in parentheses. ***, **, and * denote significance levels of 1%, 5%, and 10%, respectively.

Dependent Variable PCAR

Negative Revisions Positive Revisions

(1) (2) (3) (4) (5) (6) REV*NONPILOT* REMOVAL 1.716**

2.053*** 2.036*** -0.147 0.241 0.149

(0.599) (0.612) (0.612) (0.933) (0.962) (0.958) REV* REMOVAL -2.505*** -2.853*** -2.870*** 1.350* 1.012 0.939 (0.513) (0.524) (0.525) (0.754) (0.779) (0.779) REV*NONPILOT -1.639*** -1.671*** -1.654*** -0.021 -0.020 0.013

(0.474) (0.481) (0.481) (0.702) (0.723) (0.721) NONPILOT* REMOVAL 0.070 -0.059 -0.085 0.989*** 1.118*** 1.108***

(0.322) (0.333) (0.333) (0.310) (0.321) (0.320) REV -0.067 -0.027 -0.016 0.049 0.049 0.049

(0.187) (0.193) (0.193) (0.170) (0.176) (0.175) NONPILOT 0.678*** 0.799*** 0.717*** 0.183 0.129 -0.070

(0.263) (0.272) (0.274) (0.251) (0.261) (0.261) REMOVAL 2.046*** 2.030*** 1.986*** 2.624*** 2.637*** 2.441*** (0.412) (0.418) (0.420) (0.573) (0.589) (0.593) Horizon 1.067** 0.980** 1.160** 3.989*** 4.057*** 3.990*** (0.470) (0.488) (0.490) (0.451) (0.467) (0.467) Experience

-0.088 -0.120 0.378*** 0.366***

(0.124) (0.124) (0.113) (0.113) All Star

0.446** 0.367* 0.058 0.165

(0.218) (0.219) (0.189) (0.189)

Coverage -0.287* -0.270* -0.243* -0.352** (0.156) (0.157) (0.141) (0.141) Brokerage Size

0.003* 0.002 -0.004*** -0.004**

(0.002) (0.002) (0.001) (0.001) Size

0.152*** -0.286***

(0.047) (0.040) Market/Book

-0.229*** -0.156***

(0.065) (0.052) Cashflow Volatility

0.791 -3.453***

(0.658) (0.620) Institutional Holdings

0.978*** 0.803**

(0.327) (0.322) Constant -6.606*** -5.423** -7.894*** -21.023*** -21.184*** -18.115*** (2.559) (2.696) (2.756) (2.459) (2.571) (2.610) N 11785 11091 11091 13140 12342 12342 R-sq 0.008 0.011 0.013 0.032 0.035 0.044

39

TABLE 6 Cross-sectional Analysis

Table 6 reports the multivariate DiD regression results of two groups of stocks sorted by firm size and the level of analyst forecast dispersion. Firms are sorted into two groups (above or below the median) by their firm size or forecast dispersion each year. Analyst forecast dispersion is the standard deviation of analysts’ forecasts for firm j in year k-1. The HIGH (LOW) group includes stocks above (below) the median level of forecast dispersion. In Panel A, the dependent variable is PVOL, which is the trading volume prior to the analyst forecast announcement. In Panel B, the dependent variable is PCAR, which measures the stock price reaction prior to the announcement. All variables follow the definitions in Table 1. HAC standard errors are clustered by firm-analyst and reported in parentheses. ***, **, and * denote significance levels of 1%, 5%, and 10%, respectively.

Panel A: Trading Activity across Firm Size and Forecast Dispersion

Dependent Variable PVOL

Sorted by Size Dispersion

Level of Size/ Dispersion SMALL LARGE HIGH LOW

(1) (2) (3) (4)

PILOT*DURING 0.120*** -0.033 0.070** -0.033*

(0.028) (0.025) (0.028) (0.017)

REV -0.693*** -0.799*** -0.622*** -1.048***

(0.089) (0.042) (0.098) (0.175)

DURING -0.058*** 0.022** -0.021* 0.011

(0.018) (0.010) (0.012) (0.020)

PILOT -0.072*** 0.014 -0.032** 0.017

(0.014) (0.021) (0.016) (0.030)

Exp Trade 0.846*** 0.868*** 0.879*** 0.852***

(0.016) (0.011) (0.014) (0.016)

Spread -0.601*** -0.116** -0.152*** -0.052

(0.219) (0.050) (0.048) (0.034)

Constant 2.812*** 2.157*** 1.765*** 2.522***

(0.389) (0.249) (0.418) (0.295)

N 4653 10093 4645 4600

R-sq 0.837 0.891 0.915 0.908

Analyst Characteristics Yes Yes Yes Yes

Firm Characteristics Yes Yes Yes Yes

40

Panel B: Price Sensitivity across Firm Size and Forecast Dispersion

Dependent Variable PCAR

Sorted by Size Dispersion

Level of Size/ Dispersion SMALL LARGE HIGH LOW

(1) (2) (3) (4)

REV*PILOT*DURING 4.249*** -1.704 8.694*** -8.629**

(1.390) (1.451) (0.511) (3.618)

REV*DURING -0.032 0.366 -3.935*** 1.815

(1.009) (0.967) (0.397) (1.324)

REV*PILOT -3.094** 1.415** -6.044*** 8.797***

(1.299) (0.685) (0.299) (1.488)

PILOT*DURING 0.142*** -0.457*** 0.447* -0.630***

(0.045) (0.099) (0.246) (0.106)

REV 0.824 4.375*** 4.594*** 2.680**

(0.834) (0.869) (0.227) (1.112)

DURING -0.074 -0.015 -0.521*** 0.074

(0.104) (0.143) (0.055) (0.124)

PILOT -0.166*** 0.211* -0.405** 0.386***

(0.034) (0.109) (0.192) (0.097)

N 4653 4793 4645 4600

R-sq 0.837 0.891 0.915 0.908

Analyst Characteristics Yes Yes Yes Yes

Firm Characteristics Yes Yes Yes Yes

41

TABLE 7 Short Selling around Stock Recommendation Changes

Table 7 compares trading volume and the informational efficiency around analysts’ stock recommendation changes. 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 is the numeric value of the stock recommendations, where strong buy =1, buy =2, hold =3, underperform =3, and sell =4. A negative recommendation change, Downgrade, equals one if the stock recommendation 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 is smaller than 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖−1, and zero otherwise. Similarly, a positive recommendation change, Upgrade, equals one if the stock recommendation 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖 is greater than 𝐷𝐷𝑅𝑅𝑅𝑅𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖𝑖𝑖−1 , and zero otherwise. In Panel A, the dependent variable is PVOL, which is the 5-day cumulative trading volume [-5, -1] prior to analyst i's stock recommendation change for firm j in year k. In Panel B, the dependent variable is PCAR, which is the 5-day cumulative abnormal return [-5, -1] prior to analyst i's stock recommendation change for firm j in year k. All other variables follow from the previous tables. HAC standard errors are clustered by firm-analyst and reported in parentheses. ***, **, and * denote significance levels of 1%, 5%, and 10%, respectively.

42

Panel A: Trading Volume

Dependent Variable PVOL

Type of Recommendation Downgrade Upgrade

(1) (2) (3) (4) (5) (6)

PILOT*DURING 0.048** 0.058** 0.047* -0.004 -0.013 -0.010

(0.022) (0.029) (0.026) (0.020) (0.026) (0.025)

PILOT -0.021 -0.026 -0.016 -0.021* -0.009 -0.005

(0.014) (0.018) (0.017) (0.012) (0.015) (0.015)

DURING -0.012 -0.004 -0.013 -0.002 0.008 -0.002

(0.018) (0.022) (0.022) (0.018) (0.018) (0.015)

Exp. Trade 0.773*** 0.776*** 0.687*** 0.779*** 0.775*** 0.691***

(0.007) (0.008) (0.011) (0.005) (0.006) (0.010)

Spread 0.203*** 0.197*** -0.003 0.352*** 0.345*** 0.149**

(0.056) (0.063) (0.075) (0.074) (0.087) (0.075)

Experience -0.002 -0.016** -0.022*** -0.029***

(0.008) (0.007) (0.008) (0.008)

All Star -0.051** -0.062*** 0.056*** 0.035*

(0.023) (0.020) (0.019) (0.019)

Coverage 0.036** 0.034* 0.002 -0.007

(0.018) (0.018) (0.018) (0.018)

Brokerage Size 0.000** 0.000 0.000** 0.000*

(0.000) (0.000) (0.000) (0.000)

Size 0.113***

0.111***

(0.009)

(0.009)

Market/Book -0.015***

-0.023***

(0.005)

(0.005)

Cashflow Volatility 0.416***

0.418***

(0.082)

(0.060)

Institutional Holdings 0.121***

0.052

(0.038)

(0.042)

Constant 5.280*** 5.123*** 5.399*** 5.203*** 5.246*** 5.557***

(0.105) (0.116) (0.108) (0.071) (0.093) (0.100)

N 5311

4289 4289 5614

4550 4550

R-sq 0.862 0.863 0.870 0.867 0.866 0.874

43

Panel B: Price Sensitivity

Dependent Variable PCAR

Type of Recommendation Downgrade Upgrade

(1) (2) (3) (4) (5) (6)

PILOT*DURING -1.067*** -1.066*** -1.098*** -0.385 -0.181 -0.188

(0.335) (0.327) (0.355) (0.385) (0.267) (0.278)

PILOT 0.194 0.244* 0.252* 0.080 -0.071 -0.074

(0.159) (0.142) (0.141) (0.161) (0.157) (0.175)

DURING 0.860 0.941** 1.045*** -0.278 -0.283 -0.278

(0.537) (0.440) (0.395) (0.387) (0.380) (0.356)

Experience -0.048 -0.006 -0.057 -0.068

(0.154) (0.131) (0.126) (0.103)

All Star 0.336 0.430 0.246 0.176

(0.298) (0.316) (0.292) (0.294)

Coverage 0.251 0.082 0.150 0.093

(0.194) (0.180) (0.095) (0.113)

Brokerage Size -0.001 -0.000 0.002** 0.002

(0.002) (0.002) (0.001) (0.001)

Size -0.206***

0.066

(0.064)

(0.075)

Market/Book -0.173

-0.201*

(0.123)

(0.119)

Cashflow Volatility -0.087

-0.030

(1.089)

(1.435)

Institutional Holdings -1.095

0.231

(0.710)

(0.536)

Constant -0.012 -0.604 2.407* -0.116 -0.597*** -0.643

(0.401) (0.843) (1.238) (0.194) (0.223) (0.997)

N 5320

4299 4299 5653

4557 4557

R-sq 0.005 0.006 0.017 0.002 0.003 0.007

44

TABLE 8 Short Selling and Informational Efficiency around Earnings Announcements

Table 8 reports the results of the multivariate difference-in-differences (DiD) tests that show how Reg SHO affected trading activity and the informational efficiency of pilot stocks during firm earnings announcement periods. In Panel A, the dependent variable is PVOL, which is the 5-day cumulative trading volume [-5, -1] prior to the earnings announcement of firm j in quarter q. In Panel B, the dependent variable is PCAR, which is the 5-day cumulative abnormal return [-5, -1] prior to the earnings announcement of firm j in quarter q. Earnings surprise, Surp, is the difference between analyst i's forecast for firm j in quarter q and actual earnings for the same firm-quarter, scaled by prior fiscal year ending price. PILOT is an indicator variable that equals one if the stock was selected into the pilot program by the SEC, and zero otherwise. DURING is an indicator variable that equals one for the Reg SHO periods, and zero for the pre-Reg SHO periods. All other variables follow from the previous tables. HAC standard errors are clustered by firm and reported in parentheses. ***, **, and * denote significance levels of 1%, 5%, and 10%, respectively. Panel A: Trading Volume

Dependent Variable PVOL

Type of Earnings News Negative Positive

(1) (2) (3) (4)

PILOT*DURING 0.016*** 0.020*** 0.017*** 0.015***

(0.004) (0.004) (0.003) (0.003)

PILOT 0.008*** 0.006 -0.006* -0.002

(0.003) (0.004) (0.003) (0.003)

DURING 0.008 -0.010 0.018 -0.009

(0.007) (0.007) (0.014) (0.009)

SURP 3.671 -25.391*** 0.373 17.425***

(5.943) (6.643) (1.111) (2.004)

Exp. Trade 0.965*** 0.873*** 0.969*** 0.892***

(0.002) (0.006) (0.003) (0.002)

Spread -0.127*** -0.222*** -0.097*** -0.158***

(0.011) (0.020) (0.026) (0.026)

Size 0.099***

0.074***

(0.007)

(0.005)

Market/Book 0.001

0.009***

(0.004)

(0.002)

Cashflow Volatility 0.351***

0.258***

(0.041)

(0.016)

Institutional Holdings 0.240***

0.241***

(0.024)

(0.008)

Constant 2.060*** 2.284*** 2.029*** 2.234***

(0.017) (0.015) (0.043) (0.034)

N 8848 8848 21151 21151

R-sq 0.924 0.928 0.928 0.931

45

Panel B: Price Sensitivity

Dependent Variable PCAR

Type of Earnings News Negative Positive

(1) (2) (3) (4)

SURP*PILOT*DURING 0.169*** 0.170*** -0.137 -0.135

(0.037) (0.037) (0.087) (0.088)

SURP*PILOT -0.223*** -0.226*** 0.031 0.030

(0.026) (0.027) (0.084) (0.085)

SURP*DURING -0.050 -0.051 0.010 0.009

(0.049) (0.049) (0.024) (0.025)

PILOT*DURING -0.002* -0.002** 0.000 0.000

(0.001) (0.001) (0.001) (0.001)

SURP 0.150*** 0.155*** 0.060** 0.057**

(0.044) (0.044) (0.024) (0.024)

PILOT 0.000 0.000 0.001 0.001

(0.001) (0.001) (0.001) (0.001)

DURING 0.003*** 0.004*** 0.001* 0.002**

(0.001) (0.001) (0.001) (0.001)

Size 0.000

-0.000

(0.000)

(0.000)

Market/Book -0.001***

0.000*

(0.000)

(0.000)

Cashflow Volatility -0.001

0.002

(0.003)

(0.003)

Institutional Holdings 0.000

-0.002**

(0.002)

(0.001)

Constant -0.004*** -0.004 0.002*** 0.005**

(0.000) (0.004) (0.001) (0.002)

N 8859 8859 21177 21177

R-sq 0.003 0.004 0.001 0.001