Why do Investors Pay Attention to Stock Spam?

Why do Investors Pay Attention to Stock Spam?
Karen K. Nelson, Richard A. Price* and Brian R. Rountree
Jones Graduate School of Management, Rice University, Houston, TX 77005
March 2009
Abstract
This study examines the role of attention in the market response to spam messages and in the
selection of targeted stocks. We show that returns and abnormal volume are highest in the subset
of stocks targeted by spam emails that catch investors’ attention with extreme target prices and
stale press release information. The use of these attention-grabbing devices in stock spam has
become significantly more common in recent years. Attention effects also contribute to the
spammers’ selection of stocks to target. Spammed stocks are significantly more likely to have an
unusual level of recent trading volume and news.
JEL classifications: G11; G12; G14
Keywords: Investor attention; Spam; Target prices; Press releases
We appreciate helpful comments and suggestions of Patty Dechow, Gustavo Grullon, Per
Olsson, Greg Sommers, James Weston, and workshop participants at University of California at
Berkeley, Rice University, the 2008 American Accounting Association Financial Accounting
and Reporting Section Midyear Meeting, and the 2008 Lone Star Accounting Research
Conference. We also appreciate the research assistance of Tammy Knots, Ritesh Kumar, Ben
Perry, Dan Smith, and Luke Stadel. Richard Jones and Leonard Richardson generously provided
some of the data for this study.
* Corresponding author.
E-mail addresses: nelsonk@rice,edu, [email protected], [email protected]
1
1. Introduction
We examine the effects of investor attention using a unique sample of firms that exist in
relative anonymity until becoming the target of a mass email stock spam campaign. Prior
research focuses on larger stocks that likely already attract a fair amount of attention (e.g.,
Grullon et al. 2004; Engelberg et al. 2007; Barber and Odean, 2008; Hou et al. 2008), making it
difficult to isolate the effects of heightened attention associated with any particular event.
Moreover, these studies use relatively coarse measures of attention, whereas we develop proxies
that identify specific attention-grabbing attributes of spam email messages.1
Using internet archives of spam messages between January, 1999 and May, 2006, we
identify 658 unique spam events largely targeting companies trading on the Pink Sheets (54% of
the sample) or the OTC Bulletin Board (44%); in other words, small stocks that are unlikely to
garner investors’ attention under ordinary circumstances. We find that, on average, share prices
and volume respond positively to stock spam, at least in the short run. Mean one-day (three-day)
returns surrounding the spam date are a positive and significant 4% (8%). These returns reverse,
however, within a few days. Median returns are not significantly different from zero, however,
suggesting that stock spam does not affect share prices of the majority of targeted companies.
With 100 million stock spam messages sent each week according to Securities and
Exchange Commission (SEC) estimates (Lerer, 2007), what prompts investors to trade in the
stocks touted by some spam messages and not the stocks touted by others? We posit that
investors are attracted to stocks targeted by spam messages that catch their attention. To
1
Proxies for investor attention used in prior work include abnormal trading volume (Barber and Odean, 2008; Hou
et al., 2008), extreme returns (Barber and Odean, 2008), up/down markets (Hou et al., 2008), the existence of a news
story about the firm (Barber and Odean, 2008), and advertising expense (Grullon et al. 2004). In an effort to provide
a more precise measure of attention, Engelberg et al. (2007) employs television viewership data for the CNBC show
Mad Money to capture viewers’ characteristics, such as household wealth. Our study is unique in focusing on
message content as a means of attracting investor attention.
1
examine the effect of message content on investor attention and the market response to stock
spam, we manually collect detailed information about each spam message and investigate
whether certain message attributes are correlated with the market response.
We document that stock spam messages often include quoted information from an old
press release, typically one that was originally issued by the company targeted in the spam. On
average, the original press release precedes the spam message by 25 trading days. This spam
strategy likely attracts attention by imparting an air of credibility to spam recommendations.
Prior research suggests that investors overreact to attention-grabbing news events containing
stale information (Huberman and Regev, 2001; Tetlock, 2007).
Another prominent attribute of many stock spam messages is a target stock price. In our
sample, target prices greatly exceed share prices at the spam date, as evidenced by an attentiongrabbing mean (median) implied return of 412% (221%). Brav and Lehavy (2003) shows that
abnormal returns are increasing in the favorableness of analysts’ target prices. In a setting more
similar to ours, however, Antweiler and Frank (2004) finds that stock prices of 45 large
companies do not respond strongly to the bullishness of internet stock message boards.2
We report that the use of previously issued press releases and target prices in stock spam
has become significantly more frequent in recent years, suggesting spam has evolved over time
to include these attention-grabbing devices. Findings from regression analyses show that spam
messages with a target price trigger a significant market response, but only when that target price
is accompanied by information from a previously issued press release. Moreover, the larger the
return implied by the target price, the larger the market reaction. The results are robust to
2
Although internet bulletin board messages often contain a predicted price change, the measure of bullishness in
Antweiler and Frank (2004) is not the implied return but rather a trichotomous classification of messages as bullish,
bearish, or neither based on a computational linguistic algorithm analyzing the text of the bulletin board messages.
2
controls for other proxies for investor attention, including abnormal volume and the number of
duplicate messages sent for each spam event, and to alternative return and volume measures.
Our results indicate that stale press release information in spam messages does not by
itself generate a market response. However, we find a significant market reaction to the original
issuance of the press releases, particularly for earnings-related information, suggesting the
information was viewed as important when first released. Thus, our findings suggest that stale
information does not directly trigger a market reaction, but instead serves to reinforce or validate
the new target price information.
Finally, we provide evidence on the determinants of stock spam, focusing on whether
spammers themselves are affected by attention-grabbing events in selecting stocks to target. We
investigate whether there are systematic differences in returns, abnormal volume, and press
releases between our sample of stocks targeted by spam and two control samples: (i) random
non-spam dates for our spam sample, and (ii) random dates for a sample of stocks not identified
as spammed securities trading on the OTC Bulletin Board or PinkSheets. Logit regression
results show that targeted stocks have unusually high trading volume in the month preceding the
spam event and are more likely to have issued a press release. Overall, the results show that
attention can trigger a cascade effect, whereby attention-grabbing events influence the selection
of stocks targeted by spam emails and ultimately investors’ response to spam messages.
Our paper leverages a unique sample of firms to further our understanding of investor
attention. We find several similarities between our results and prior evidence, suggesting that the
influence of investor attention on investment decisions generalizes beyond samples of larger,
more widely-known stocks. We extend this literature to show that attention depends on the
nature of the information conveyed and on how that information is bundled.
3
Although our results indicate that stock spam is an attention-grabbing event for some of
the small stocks that are targeted, one might question why investors are willing to purchase these
stocks at all. Prior research suggests that some investors trade in stocks with lottery-type
features, such as our sample of spam stocks, as a form of gambling (e.g., Friedman and Savage,
1948; Markowitz, 1952; Statman, 2002; Barberis and Huang, 2007; Kumar, 2009). Our finding
that the market response to spam is increasing in the exorbitant returns implied by target prices
included in some spam messages (i.e., the jackpot) suggests a link between attention and
gambling-motivated investment decisions.
Our paper is related to other recent work on stock spam showing that, on average, spam
increases share prices and trading volume around spam events, and that these effects are
generally increasing in the frequency of spam messages (Böhme and Holz, 2006; Frieder and
Zittrain, 2007; Hanke and Hauser, 2008).3 Our study differs from these papers by investigating
specific attributes of spam messages that drive trading in targeted stocks. We show that spam
messages have become more sophisticated over time and that spam generally does not work
unless the message is crafted to catch investors’ attention by including both a target price and old
press release. Finally, we show that attention effects not only explain the market reaction to
spam but also which stocks are likely to be the target of spam.
Section 2 describes the data. Section 3 presents our main empirical tests of the market
response to attributes of spam messages. In section 4, we examine the market reaction to the
original issuance of press releases repeated in spam to calibrate the reaction on the spam date.
Section 5 investigates the determinants of stocks targeted by spam. Section 6 concludes.
3
The samples used in these related papers are smaller and concentrate on a shorter time period than our sample.
Böhme and Holz (2006) examines 93 stocks touted in spam emails between November, 2004 and February, 2006;
Frieder and Zittrain (2007) limits their sample to 304 Pink Sheet stocks targeted from January, 2004 to July, 2005;
and Hanke and Hauser (2008) uses a data set of 235 stocks spammed between January 1, 2005 and December 31,
2005.
4
2. Data and descriptive statistics
2.1. Spam sample
We obtain spam messages from two internet sources which archive spam sent to both
personal email addresses and to email addresses created specifically for the purpose of attracting
spam. The first archive (www.annexia.org) includes spam sent during 1997-2004 and the second
(www.crummy.com) spans the years 2005-2006. Because these archives also include non-stock
spam messages, we obtain the full text of each spam message, and scan for words commonly
included in stock spam, such as “OTC,” “Pink Sheets,” “strong buy,” and “undervalued,” among
others.4 This procedure identifies a sample of stock spam from January, 1999 through May,
2006.5
Spam messages received within a short period of time are generally duplicates of other
spam messages sent during the same short window. Accordingly, we identify each unique spam
and count the number of duplicate messages. If more than one unique spam for a given target
company occurs within a seven day period, we exclude all but the first unique spam. In all
analyses, we use the date of the first spam message received as the spam date (day 0) for each
unique spam event.
As shown in Table 1, Panel A, our final sample consists of 658 unique spam messages for
442 target companies. The number of unique spam messages (target companies) increases
dramatically over the sample period, from 41 (35) during 1999-2001, 139 (105) during 20022004, and finally 478 (321) during the final two years of the sample period, 2005-2006. Panel A
4
A recent trend in spam is to send messages as image files. These messages are generally excluded from the sample
because of the difficulty of automating the collection of the information, i.e., scanning the text of a spam message
sent as an image file does not identify the terms associated with stock spam unless the keywords are in the title.
Thus, our sample of stock spam includes messages that were sent in text or html format.
5
We do not have information on who sent the messages, or who (in addition to the data providers) received the
spam messages.
5
also shows a corresponding increase in the number of duplicate messages associated with each
unique spam event. Panel B of Table 1 reveals that the majority of the sample spam (54%)
targets companies trading exclusively on the Pink Sheets, followed by companies trading on the
OTC Bulletin Board (44%). Less than 2% of sample spam targets companies trading on AMEX
or NASDAQ. Finally, Panel C shows that most spam messages are sent on weekdays, with the
highest incidence occurring on Monday (23%), and declining monotonically throughout the
week to a low of 4% on Saturday.
2.2. Spam attributes
The content of stock spam messages varies considerably, as illustrated by the
representative examples in the Appendix. The first spam targeting New Millennium Media is
brief, with little information about the company and several “puffing” statements implying the
spammer has private information about future company sales and contractual developments that
are projected to increase trading volume and share price. The spammer quotes a target price of
$2.00 when shares in this company were trading at $0.42.
The second spam targeting eUniverse resembles an analyst research report, with select
financial statement data, summary bullet points highlighting company developments, a
description of the company, a fundamental valuation analysis, and a target stock price well above
the 52-week price range. The spam message also includes a disclaimer by the spammer
acknowledging the risks of investing in eUniverse, and directs the reader to the company’s SEC
filings for more information concerning these risks. Finally, the spam message discloses that the
spammer was not hired by the target company and did not receive any compensation from or on
behalf of eUniverse.
6
The final spam targeting Digilava Inc. is fairly sensational in format and tone, but does
contain background information on the company. An interesting feature of this spam is that it
embeds a prior press release from the targeted company, stating that Digilava “today announced
that the Company has executed a Letter of Intent” to acquire a strategic partner. No mention is
made of the fact that the original announcement occurred in a press release on March 9, 2004,
even though the spam was first sent on March 11, 2004. This spam message also includes a
disclaimer regarding the risks of investment, but further indicates that the spammer may “own,
buy, or sell any securities mentioned at any time,” and was paid $12,350 from an unidentified
third party to send the message.
To capture the variation in the content of spam messages, we manually code several
features of each unique spam. We present summary descriptive information for these
characteristics in Table 2, Panel A for the pooled sample and for three separate time periods,
1999-2001, 2002-2004, and 2005-2006. Most of the spam messages (74% in the pooled sample)
provide a description of the company, which typically involves a discussion of the history of the
company and the products/services provided by the company. Relatively few spam messages
include a discussion of fundamental analysis focusing on valuation metrics as support for
recommending the stock. However, the frequency of this disclosure increased significantly in
the last two years of the sample period, up to 14% in 2005-2006 compared to 5% in 2002-2004.
The majority of the spam messages quote information from a previously issued press
release.6 Specifically, in the pooled sample 62% of spam messages include a directly quoted
press release (i.e., a press release that acknowledges the source of the news) and an additional
11% of spam messages include an indirectly quoted press release (i.e., a press release that does
6
We provide additional information on characteristics of the press releases (e.g., source, type of news, and the time
lag to the spam message) in Table 3.
7
not attribute the information to a source). We are able to verify the existence of most of the press
releases by tracing the information back to the original announcement. It is possible that the
press releases we are unable to verify do in fact exist, but are no longer available through public
archives. The practice of directly quoting a press release has increased significantly over time,
from 34% of spam messages in 1999-2001 to 49% in 2002-2004 (significant at the 0.10 level),
and finally to 68% in 2005-2006 (significant at the 0.01 level). In contrast, indirectly quoted
press releases have become less common over time, falling from 20% in 1999-2001 to 9% in
2005-2006.
Many spam messages contain a separate disclaimer section. Approximately two-thirds of
the sample spam messages include safe harbor language consistent with the provisions of the
Private Securities Litigation Reform Act of 1995 (PSLRA) to caution that forward-looking
statements in the spam message may not be realized. An example of this language is shown in
the Digilava spam message reproduced in the Appendix. The use of safe harbor cautionary
language is peculiar in the context of spam because the PSLRA’s safe harbor only provides
protection for forward-looking statements made by the company or its representatives.
Nevertheless, the use of safe harbor cautionary language has increased dramatically, perhaps to
lend a false air of legal legitimacy to the spam message. A smaller proportion of sample spam
messages contain a generic disclaimer without the usual safe harbor cautionary language;
however, these generic disclaimers were more frequent in the early years of the sample period.
In the same section as the disclaimer, many spam messages disclose compensation
received, either monetary (46%) or shares in the targeted company (14%). The proportion of
spam messages disclosing the receipt of monetary compensation doubled between 1999-2001
and 2002-2004, although the median dollar amount has remained relatively constant. The
8
significant increase in these disclosures is likely in response to enforcement actions taken by the
SEC. Sending stock spam does not constitute fraud, but failing to adequately disclose
compensation received is actionable. Often included in the disclaimer section is the name of the
group or individual responsible for sending the message. This disclosure, however, has become
significantly less common in recent years.
Finally, our sample spam messages commonly include the current share price of the
targeted company, a practice that has increased significantly over the sample period. The use of
a target share price has also significantly increased over time; none of the spam messages in
1999-2001 include a target price while 41% of the 2005-2006 spam messages contain a target
price. Spammers also increasingly reference another company or companies they claim to have
previously recommended and have since performed well.
Table 2, Panel B provides additional information on target share prices and the related
implied return. There are 192 unique spam messages containing a target share price and a
current price. Target prices greatly exceed current shares prices, as evidenced by an implied
mean (median) return of 412% (221%). To put these findings in context, Brav and Lehavy
(2003) reports that over the period 1997-1999 equity analysts’ mean (median) target price is
higher by 32.9% (25.5%) relative to the current stock price; even for stocks rated a ‘strong buy’
by analysts, the target price exceeds the current stock price by 41%, on average, substantially
below the implied return of spammers’ target prices.
Taken together, the evidence presented in Table 2 suggests that spammers adapt message
content over time to include more information that appears credible, most notably prior
company-issued press releases, while at the same time disclaiming legal responsibility through
the use of safe harbor cautionary language and the disclosure of compensation received for
9
sending the spam messages. These disclosures, coupled with high target price projections and
claims of past stock-picking success, arguably increase the attention-grabbing nature of stock
spam. In the analyses reported in Section 3, we examine the extent to which these attributes
affect the market reaction to the spam messages.
2.3. Quoted press releases
In addition to coding the content of the spam messages, we collect data on the press
releases contained within the spam messages. As shown in Table 3, Panel A, of the 658 unique
spam messages, 468 directly or indirectly quote a previously issued press release. We searched
for each press release to obtain the source and original announcement date and to verify its
content. We were able to verify the content of the press releases contained in 438 of the spam
messages, and to find the actual press releases for 421 spam messages.
Panel B of Table 3 presents information on the characteristics of press releases contained
in the spam messages. Because our subsequent empirical tests examine returns and volume
around the original press release dates, Panel B reports both the total number of press releases
contained in sample spam and the number of press releases with returns data available. We
focus our discussion on the latter subsample of 632 observations. Although some spam
messages quote more than one press release, the mean (median) number of press releases quoted
per spam message is 0.96 (1.00). In almost all cases (623 out of 632), the source of the press
release is the targeted company, with the few remaining news articles generally originating from
journalists. The mean (median) number of trading days between the original press release and
the first spam message incorporating that release is 25 (10), suggesting a substantial lag after the
news first became publicly available.
10
In Panel C of Table 3, we classify the news contained in each of the press releases into
the following five groups: (i) Contracts, including the formation of partnerships and the
execution of other contracts; (ii) Earnings, including earnings announcements and management
earnings forecasts; (iii) Expansion, including acquisitions, exploration, and external financing;
(iv) Products, including marketing, products and service arrangements; and (v) General,
including changes in management and other events that do not fit in one of the other categories.
If the press release contained more than one news item, we coded only the most significant item
discussed. The most common categories of news are Products, Contracts, and Expansion;
Earnings and General news are relatively infrequent by comparison.
3. Market response to spam
In this section, we examine the distribution of returns and abnormal volume surrounding
the spam date for the pooled sample (section 3.1) and for different partitions of the sample based
on attention-grabbing spam attributes (section 3.2). We also present regression analyses of
returns and abnormal volume on spam attributes (section 3.3).
3.1. Pooled results
Table 4, Panel A, provides information on the distribution of raw returns for the sample
of 658 unique spam messages in our sample. The mean return in the 3-day window (−1,1)
surrounding the spam date is a positive and significant 8%. There is a significant run-up in stock
prices prior to the spam date (13% mean return in the (−10,−2) window) followed by a
statistically significant reversal from day 0 through day 10. The median return, however, is not
statistically different from zero except in the event windows extending at least 5 days after the
spam date, where the returns are significantly negative. These results indicate that spam does not
11
produce positive returns for most targeted stocks. In fact, on day 0 positive and negative returns
occur with almost equal frequency. Abnormal trading volume, defined as the volume on a given
day divided by the median volume over the window (−20, −2), is unusually high, however, in the
days immediately following the spam release, as shown in Table 4, Panel B. Median volume is
more than twice as high as usual through day 3, but begins to taper off subsequently.
The analyses in Table 4, Panel A use raw returns because there is not a suitable market
index for our sample.7 Therefore, we estimate the distribution of returns for sample firms on
randomly selected non-spam dates to provide a benchmark for the spam date results. For each
unique spam message, we randomly select one event date (day 0) preceding and one following
the spam date, excluding the 30 days surrounding the spam date. The results of this analysis,
reported in Table 5, Panel A, show that mean returns are insignificantly different from zero in all
event windows. Median returns are zero in the shorter event windows, and negative in the longer
event windows, and the proportion of positive returns is generally lower than for the spam event
dates reported in Table 4. Finally, Table 5, Panel B reveals no evidence of unusual volume
around the random event dates as the abnormal volume measure is close to one. Taken together,
the results in Tables 4 and 5 show that spam generally shifts the distribution of returns to the
right, but does not generate a positive return for the majority of targeted companies. On average,
spam temporarily improves liquidity without affecting stock prices.
3.2. Market response and spam attributes
We posit that the market response to stock spam is related to attention-grabbing attributes
of the spam messages. In particular, investors are more likely to notice spam messages quoting a
previously issued press release or containing a target price. Table 6 reports returns for the
7
Market indices such as the Russell 2000 Index or the Russell Microcap Index exclude stocks traded on the OTC
Bulletin Board and Pinksheets because of their failure to meet national exchange listing requirements. In general,
however, research finds that raw returns approximate abnormal returns over short windows.
12
sample partitioned on whether the spam message includes a press release (Panel A), target price
(Panel B), or both (Panel C).
The findings in Table 6, Panel A show that including an old press release does little to
explain the market reaction to spam, inconsistent with an investor attention argument. Although
the mean return in the one-day event window (0,0) is higher for the press release sample (4%)
than the no press release sample (2%), this is not the case in either the two-day (0,1) or three-day
(–1,1) windows. Moreover, in all three event windows, median returns are zero whether or not a
press release is included. Overall, these findings indicate that the attention paid to an old press
release included in spam does not by itself trigger a market response.
In contrast to the press release findings, Panel B of Table 6 suggests returns are higher for
spam messages including a target price. In all event windows, mean returns are approximately
three times higher for the target price sample, and the median return is also significantly positive
in the one-day event window. Note that this analysis considers only whether a target price is
included in the spam message; we investigate whether the magnitude of the return implied by the
target share price is associated with event window returns in Table 8.
Because spam messages often (explicitly or implicitly) justify a target price by reference
to firm-specific information, Table 6, Panel C sorts the sample based on both press release and
target price.8 Consistent with the univariate sorts, mean returns are higher when a target price is
included in the spam message while there is only weak evidence of a press release effect,
suggesting target prices are the primary factor attracting investor attention. However, the
findings also show that the combination of a target price and press release generally produces the
highest mean and median returns. Taken together, the findings reported in Table 6 show that
8
Only 69 (10%) spam messages include a target price without a press release, compared to 332 (50%) that include a
press release only and 136 (21%) that include both.
13
spam messages providing a target price attract investors’ attention, particularly when that target
price is accompanied by hard information from a previously issued press release.
Although our focus is on specific attention-grabbing attributes of spam messages, Barber
and Odean (2008) reports that abnormal trading volume is a strong indicator of investor
attention; when volume is unusually heavy, it is likely that investors are paying more attention
than usual to the stock.9 To test this alternative measure of attention, Table 7, Panel A sorts the
sample into terciles based on abnormal volume on day 0. We find that mean and median returns
are consistently positive and significant only in the high abnormal volume partition. However,
Panel B of Table 7 shows that even within the highest tercile of abnormal volume, returns are
driven by spam messages containing both press release and target price information. Mean
(median) returns on the spam date are 20% (14%), considerably higher than any other partition
examined thus far. The only other partition with statistically significant returns is the press
release only category; the target price only category has relatively few observations, as noted
above, making statistical inference difficult. Without a target price or a price release, however,
high abnormal volume does not generate significant returns.10
A second alternative measure of attention is the frequency of messages sent for each
spam event; sending more messages increases the number of potential investors aware of the
targeted stock. Table 7, Panel C sorts the sample into terciles based on the frequency of
duplicate spam messages for each spam event.11 The findings show that returns are increasing in
9
Barber and Odean (2008) investigates two other measures – news and extreme returns – that are also likely to be
associated with attention-grabbing events. They find that abnormal trading volume is the single best measure of
attention, followed by extreme returns, and lastly by whether a stock was or was not mentioned in that day’s news.
10
Mean and median returns are zero for a sample of randomly selected dates with high abnormal volume
(untabulated), providing further evidence that a partition on abnormal trading volume does not by itself produce
significantly positive returns absent a spam event.
11
Because of the discrete nature of this count data, sorting into three groups does not result in groups of equal size.
The low frequency group consists of spam events for which there was a single message in our sample; the number of
messages in the middle (high) frequency group ranges from 2–14 (15–858).
14
spam frequency, consistent with findings in Böhme and Holz (2006) and Frieder and Zittrain
(2007). We show in Table 7, Panel D, however, that spam frequency and message content are
related; high frequency touts tend to include either or both a press release or target price;
relatively few high frequency touts include neither.12 One explanation for this finding is that
more sophisticated spammers craft attention-grabbing messages that they then disseminate to a
broader audience. Within the high frequency partition, returns are once again driven primarily
by spam messages containing both press release and target price information, where both the
mean and median are positive and significant. Mean returns are also significantly positive for
high frequency spam touts containing a target price but not a press release, although median
returns are generally insignificant. Without a target price, high spam frequency generally does
not result in a significant market response.
3.3. Regression analyses
The results reported above suggest that only a subset of spam messages attract investors’
attention. To further test the market reaction to spam, we model spam date returns and abnormal
volume as a function of the attention-grabbing attributes of spam messages discussed above:
Return = α 0 + α1 Press + α 2 Implied Return_No Press
+α 3 Implied Return_Press + α 4 Abnormal Volume
(1)
+α 5 Spam Frequency + ε .
Abnormal Volume = α 0 + α1Press + α 2 Implied Return_No Press
+α 3 Implied Return_Press + α 4 Lagged Abnormal Volume
(2)
+α 5 Spam Frequency + ε .
Return is the raw return on day 0 and Abnormal Volume is the natural logarithm of 1 plus
the volume on day 0 divided by the median volume over the window (–20,–2). Press is set equal
12
Compared to Table 6, Panel C, the sample of high frequency spam events contains a disproportionate number of
spam messages with a target price only (37, or 17% versus 10% in Table 6, Panel C) and with both a target price and
press release (62, or 28% versus 21% in Table 6, Panel C).
15
to 1 if the spam message quotes an old press release about the targeted company, and 0
otherwise. Implied Return_Press (No Press) is the implied return for the sample of spam with
(without) a press release. The implied return is the natural logarithm of 1 plus the difference
between the target price included in the spam and the current price divided by the current price,
and is set equal to 0 for observations without target prices. By focusing on the magnitude of the
return implied by the target price, rather than merely the inclusion of a target price as in our
previous analyses, these regressions provide a strong test of the extent to which the bullishness of
spam messages attracts investors’ attention. Because the findings reported above suggest an
interaction effect between press releases and target prices, we report separate coefficient
estimates for implied returns with and without a press release. We expect the market response to
be increasing in the implied return, especially when the spam message also quotes an old press
release. Finally, we control for abnormal volume and Spam Frequency, the natural log of the
number of duplicate spam messages for each unique spam event, as alternative measures of
investor attention.
Table 8 reports summary statistics for the estimations of the regression equations. In the
first set of results for the spam date returns, Press and Implied Return_No Press are insignificant,
indicating that an old press release included in spam does not by itself trigger a market response.
However, Implied Return_Press is positive and significant (p-value = 0.047). Thus, the larger
the return implied by the spammer’s target price, the larger the return reaction on the spam date,
provided the spam message also includes an old press release about the targeted company.
Finally, of the two alternative measures of investor attention, only Abnormal Volume is positive
and significant.
16
The second set of results in Table 8 reports findings for the estimation of equation (2).
Similar to the returns results, Press and Implied Return_No Press are insignificant, but Implied
Return_Press is positive and significant (p-value < 0.001). The control variables Lagged
Abnormal Volume and Spam Frequency are also positive and significant. 13
Overall, the findings reported in this section show that the market response to spam is
concentrated in the subset of stocks targeted by spam emails using target prices and old press
releases to attract investors’ attention. Consistent with Barber and Odean (2008), we also find
that abnormal volume is associated with investor attention, although this evidence is not too
surprising because greater trading volume is typically driven by greater numbers of investors
trading in a stock. What is noteworthy is the evidence we document suggesting that investors
distinguish between targeted stocks based on salient attributes of the spam message. Moreover,
investors do not simply consider a single attribute of the message, but rather appear to be most
effectively swayed by a combination of attention-grabbing devices.
4. Market response to original press release
Although prior research suggests that investors overreact to stale information (Huberman
and Regev 2001; Tetlock, 2007), the results in Section 3 show that repeating an old press release
in spam does not by itself trigger a market response. One possible explanation for this finding is
that the press release does not contain meaningful information that would trigger a response,
either when originally released or when subsequently repeated in the spam email. Alternatively,
13
As additional sensitivity analysis in the return regression, we use mid-point returns (calculated as the average of
the daily high and low price because bid and ask data were not available from Pinksheets). Returns can be affected
by bid-ask bounce whereby stock prices bounce between the bid and ask price, biasing upwards the returns
calculated using transaction prices (Blume and Stambaugh, 1983). However, inferences are consistent using this
alternative measure of returns. Inferences are also unchanged in both the returns and abnormal volume regressions
when we use dollar volume rather than share volume to calculate abnormal volume.
17
investors could recognize the press release included in the spam as old information and hence not
react to it directly, but rather as validation for the new information in the target price. In this
section, we examine the market reaction to the original issue of the press release to calibrate the
reaction on the spam date. Taken together, these analyses provide relevant evidence on the stale
information hypothesis that does not rely on anecdotal data from a single event (Huberman and
Regev, 2001) or coarse proxies for stale information (Tetlock, 2007).14
The findings reported in Table 9 document a significant market reaction to the original
issuance of the press releases. Mean and median returns are significantly positive in the three
days surrounding the press release date, with a relatively high proportion of positive returns on
those days (Panel A). In addition, trading volume doubles on day 0 and day 1 (Panel B).
Separating the sample by the type of news contained in the press release reveals that earningsrelated information generates the strongest market reaction, and is the only category with
statistically and economically significant mean and median returns (Panel C). In untabulated
analysis, we examine spam date returns by press release category but fail to find any significant
patterns associated with news type. In addition, we find no evidence of a significant correlation
between returns on the original press release date and the spam date.15
Taken together, our results are inconsistent with stale information in spam eliciting a
market overreaction. We document a significant market reaction to the original press release
14
Huberman and Regev (2001) provides an interesting case study of a small biotechnology firm, EntreMed, whose
share price tripled after publication of a front-page New York Times article about the company’s new cancer-curing
drug, even though the substance of the story had been reported five months earlier in the scientific journal Nature
and in the popular press, including the Times itself. Tetlock (2007) examines the market response to public news
using several proxies to capture the degree to which a news event is stale, including the presence of another news
story or an extreme abnormal stock return in the prior week, high media or analyst coverage in the prior month, and
high stock liquidity in the prior month.
15
Anecdotal evidence suggests that some company insiders use spam to profit in their trading activities. We do not
find a significant association between spam activity and insider trading activity reported to the SEC for OTC
Bulletin Board firms. However, the level of SEC oversight for OTC Bulletin Board firms is likely limited and firms
with stocks trading on Pinksheets are not required to file with the SEC. Many firms publicly denounce stock spam
after they have been targeted by spammers.
18
issuance, and although the press releases are subsequently repeated in a spam email that also
generates a significant market response, the stale press release information does not by itself
trigger the spam response. Our results are not necessarily inconsistent with Tetlock’s (2007)
findings of a market reaction to stale news stories, however, as that paper uses proxies for the
degree to which a news story contains stale as opposed to new information. Rather, in our
setting we are better able to isolate the stale (press release) information from the new (target
price) information to determine the source of the market reaction at the spam date. Our results
show that stale information does not directly trigger a market reaction, but instead serves to
reinforce or validate new information.
5. Determinants of stocks targeted by spam
The stocks targeted by spammers are typically small, low-priced stocks not traded on an
organized exchange. From this set of potential targets, how does the spammer select stocks to
tout? On the one hand, target stocks may be randomly chosen if spammers simply engage in
spam activities for “entertainment purposes,” as claimed in some spam emails, or if spammers
believe they can craft an enticing tout for any of the potential targets. Alternatively, it is possible
that spammers focus on stocks that have attracted their attention. After all, spammers face the
same search and information processing limitations as any other individual. As a result,
spammers may target attention-grabbing stocks.
Following Barber and Odean (2008) and the findings reported above, we use returns,
abnormal trading volume, and press releases (news) as proxies for attention and investigate
whether there are systematic differences in our sample of stocks targeted by spam relative to two
control samples. The first control sample (Spam Control Sample) consists of random dates from
19
6 to 18 months before or after the spam date for our stock spam sample. The second control
sample (Random Control Sample) consists of a random sample of stocks not identified as
spammed securities trading on the OTC Bulletin Board or PinkSheets on randomly chosen dates.
If any of the random dates in either of the two control samples is in fact a spam date not
identified by our sample sources, then we bias against finding any differences between the spam
and control samples.
We obtain a list of firms trading on the OTC Bulletin Board and PinkSheets from
pinksheets.com, and stock price and volume data from nasdaq.com. We search for press releases
using Google Archived News Search which identifies press releases primarily from PR
Newswire, techweb, Primezone, PR Web, and Business Wire. We do not use the previously
gathered press releases quoted in the spam messages for the spam sample unless identified by the
Google Archived News Search as this could bias in favor of finding a significant difference
between the spam sample and the control samples.
Table 10, Panel A presents descriptive statistics for the returns, volume, and news
variables for the Spam Sample and each of the two control samples.16 The results show that
mean returns in the two months preceding the event date are significantly higher in the Spam
Sample than the Random Control Sample. Similarly, trading volume is significantly higher for
the Spam Sample relative to both control samples. Finally, there are significantly more news
events for the Spam Sample as both the mean number of press releases and the fraction of firms
with at least one press release are significantly higher for the Spam Sample.
Table 10, Panel B provides the results of logit regression analysis where the dependent
variable is equal to 1 for observations in the Spam Sample, and 0 for observations in the Control
16
The Spam Sample includes only the first spam event for each of the 442 firms in the sample. Because of missing
returns or volume data for some of the variables, this sample is reduced to 397 firms for the analyses in Table 10.
20
samples. Consistent with spammers targeting attention-grabbing stocks, the results of both
regressions show that high volume in the preceding month significantly increases the likelihood
of a stock being spammed (p-values < 0.001). The negative coefficient on volume two months
before may suggest that spammers target stocks with an unusual level of trading volume relative
to past experience. The news variable is also positive and significant in both regressions (pvalues < 0.001). Finally, there is only limited evidence past returns are higher for targeted stocks
relative to the Random Control Sample.
Overall, the results show that spammers do not randomly target stocks, but rather are
attracted to stocks that have recently caught their attention. Once a targeted stock is selected,
however, the response to the spam campaign depends on the ability of the spam message to
attract investors’ attention to the stock. Thus, our findings show that attention can have a chain
reaction effect.
6. Summary and conclusion
This study examines the role of attention in investment decisions using a sample of firms
targeted by stock spam and detailed data on the information contained in the spam messages.
We find that trading volume more than doubles in the days immediately following the spam
campaign, and the mean return is positive and significant. However, the median return is zero,
with nearly as many firms experiencing negative returns as positive on the spam date. We
conclude that touting the relatively unknown companies that are the favored targets of spammers
temporarily improves liquidity without having a significant effect on the stock price of most
targeted companies.
21
We posit that the market reaction to stock spam is related to specific attention-grabbing
attributes of spam messages. In particular, investors are more likely to notice messages quoting
a previously issued press release or containing a target stock price. These features of spam
messages have become significantly more common in recent years, suggesting the content of
spam messages has evolved over time. Results show that combining credible, but stale,
information from old press releases with a target price projection increases the return and volume
reaction to spam. Moreover, the larger the return implied by the target price, the larger the
market reaction.
We also show that attention affects the selection of targeted stocks. Using two control
samples of securities trading on the OTC Bulletin Board or Pinksheets, we find that in the period
preceding the spam event targeted stocks are characterized by unusually high trading volume and
more news (press releases). These results suggest that spammers do not randomly select stocks,
but rather are themselves influenced by attention-grabbing events.
Examining investor attention in the context of stock spam entails some trade-offs. On the
one hand, spam invariably targets small, illiquid securities with relatively little publicly-available
financial or other information. On the other hand, the relative obscurity of these stocks offers a
unique opportunity to examine attention effects unfettered by pre-existing knowledge of the
company and the possibility of confounding information events not captured by the analyses.
Moreover, our targeted investigation allows us to provide an in-depth analysis of both the
content of spam messages and the press releases they quote. Overall, the approach in our paper
and the findings we report complement and extend the growing literature on factors influencing
investor attention and its effect on asset prices.
22
Appendix. Examples of Stock Spam Messages
Target Company: New Millennium Media (OTCBB: NMMG)
Spam Date: February 10, 2002
Subject: NMMG - STRONG BUY- HUGE COVERAGE
Body: NMMG - New Millennium Media
(OTCBB : NMMG)
MAJOR CONTRACT ANNOUNCEMENTS AND HUGE NEWSLETTER COVERAGE FOR NMMG!
NMMG will be profiled by some major newsletters along with the release of
significant news regarding explosive sales for the Company. There will be
huge volume and a strong increase in price for several days. The same
newsletters that profiled IACP will begin coverage on NMMG. They brought
IACP from $ .50 to $4.35 in 10 days! We know for certain that the same
groups are going to profile NMMG.
We are very proud that we can share this information with you so that you can
make a profit out of it. It is highly advisable to take a position in NMMG
as soon as possible, today before the market closes, or Tomorrow.
NMMG shares have exploded in the last 30 days on huge volume. We are
forecasting significant contractual developments with the largest companies
in the world over the coming weeks.
The stock has been a breakout performer and will continue moving up
immediately. We think the stock can easily reach $2.00 in less than a month.
Good luck and watch NMMG fly this week!
23
Target Company: eUniverse (NASDAQ: EUNI)
Spam Date: March 18, 2002
Company Report
eUniverse, Inc. (NASDAQ: EUNI)
Revenue (millions)
Earnings per share (diluted)*
Six Month Target Price: $13.18
FY 03/00
FY 03/01
FY 03/02*
FY 03/03*
$1.8
($0.42)
$15.7
($1.50)**
$33.2
$0.18
$52.6
$0.40
52-Week Range
$1.031 - $8.50
Fiscal Year End
March 31
Shares Outst. (fully diluted)
30.0 million
Revenue/Share (TTM)
$1.03
Approx. Float
18.0 million
Price/Sales (TTM)
5.5X
40%
Price/Sales (2003)
3.7X
Insider Holdings
A Few Reasons to Consider EUNI:
•
Nielsen/NetRatings ranked the Company number 6 among the 10 most visited Web sites in the
world, right behind MSN, Yahoo, and AOL
•
Fiscal year Q3 ending Dec. 31, 2001, revenue topped $10 Million Mark for the First Time. Net
Profit Increased 131% sequentially to $2.0 Million. Higher Net Income than Yahoo!
•
Fiscal year ending March 31, 2001 revenue is estimated to be $33 Million with $5.4 Million in
Net Income or $0.18 per share. Several Wall Street firms follow EUNI and estimates for fiscal
year 2003 are $52.0 Million in revenue and an EPS of $0.40
•
EUNI entered into a strategic agreement with ATandT that is expected to reduce costs and
facilitate network expansion
•
Late last year, 550 Digital Media Ventures, a Sony Corp. company, took a $17 million equity
stake in eUniverse
About eUniverse, Inc.
EUNI operates a network of entertainment-related Websites focused on diversionary content and community offerings. EUNI's
network of Websites (www.euniverse.com) consists of diversion-oriented content and community properties revolving around
two main themes: family and entertainment. EUNI has also developed a proprietary email content delivery system
(www.flowgo.com) that is among the largest of its type on the Internet, reaching over 25 million subscribers per day and
delivering over 700 million emails per month.
People visit EUNI's sites for animated content, newsletters, online greeting cards, streaming media, and gaming information.
For its advertising partners, EUNI offers sponsorship, branding and direct marketing opportunities. With its knowledge of
when, where and how EUNI reaches its audience, eUniverse provides a rich environment to deliver information, foster longterm relationships and enable transactions online. EUNI works with its partner-advertisers to understand their goals for their
media campaign. Then, based on performance history, EUNI chooses the most relevant, persuasive and efficient contact
opportunities in its network to convey their message. EUNI constantly tests and reevaluates the media placement and creative
content within its network, working with its partners to achieve the desired result.
24
Valuation and Conclusion
Valuation
In our opinion, there are 2 ways to value EUNI, one a multiple of earnings, and two a multiple of revenues. According to
Multex, EUNIs peer group trades in the marketplace at 32.95 times earnings (for those few companies that earn money). Based
on $0.40 per share in earnings for the upcoming fiscal year (beginning next month), this would equate to a stock price of $13.18
per share.
If we look to value EUNI on a multiple of revenues, then we should look no further than Yahoo. Unlike Yahoo, EUNI is
growing in just about all aspects of their business, the only difference for the most part is, Wall Street has not heard of EUNI. If
EUNI were to trade today in the marketplace, at the same multiple of revenues as Yahoo (which is 8 times), then based on
analysts forecasts of $52.8 Million in revenues for the upcoming fiscal year, this would equate to a stock price of $14.08 per
share.
Conclusion
With the continued positive outlook in EUNIs revenue and earnings growth, we believe that EUNI is a compelling and
undervalued investment opportunity for risk oriented investors. We believe, as Wall Street catches on to the EUNI story, this
could have positive effect to EUNIs share price.
******* Important Notice and Disclaimer: Please Read *******
Investor Insights (II) publishes reports providing information on selected companies that II believes has investment potential. II is not a registered investment advisor or
broker-dealer. This report is provided as an information service only, and the statements and opinions in this report should not be construed as an offer or solicitation to
buy or sell any security. II accepts no liability for any loss arising from an investor's reliance on or use of this report. An investment in EUNI is considered to be highly
speculative and should not be considered unless a person can afford a complete loss of investment. II has not been hired by eUniverse to do this profile and did not receive
any compensation, of any kind, from or on behalf of eUniverse (EUNI) for this profile. II and/or its affiliates or agents have not discussed this profile with the management
of EUNI. Subsequently II may buy or sell shares of EUNI stock in the open market. This report contains forward-looking statements, which involve risks, and
uncertainties that may cause actual results to differ materially from those set forth in the forward-looking statements. For further details concerning these risks and
uncertainties, see the SEC filings of EUNI including the company's most recent annual and quarterly reports.
25
Target Company: Digilava Inc. (Pink Sheets: DGLV)
Spam Date: March 11, 2004
Watch DGLV - Don't Miss DGLV - March Top Stock Choice
******************************************************************************
!!!!!! Very tight float stock, Will move fast !!!!!!
******************************************************************************
Congratulations to all members who took advantage of our February Stock Choice.
GETC - The stock was profiled at 18 cents and reached 98 cents on Friday
******************************************************************************
--------------------------------------------------------------------------------------------------------------------The hottest stock for the Second week of March
DGLV * DGLV * DGLV * DGLV * DGLV * DGLV
++ Top March Trading Alert ++
Company Profile: Digilava Inc.
Symbol: DGLV
Current Price: $0.19
Once again - Very Tight Float Stock
Speculative NEar Term Target Price - $0.79
Speculative Long Term Target Price - $1.50
- DGLV is a leading San Francisco-based Application Service Provider (ASP) of permission-based rich media direct
response marketing and communications that offers empowering technology to direct marketers, web publishers and
advertisers. The Company's core products and services enable clients to optimize their marketing and advertising
campaigns by effectively delivering and tracking online video-enhanced rich media promotions, newsletters, PR and
IR releases, and distance learning, including purchases, pass-along and new customer acquisition.
- DigiLava offers client application software for editing and processing complex video sequences and automatically
processing and delivering this video content directly to their online account.
- Their Software Products Include
* Bannerforge - premiere banner graphics creation website
* VideoMail Studio - VideoMail Studio (VMS) offers consumers an on-line video streaming solution to publish
video contents on websites.
* VScope - VScope is a software vectorscope/waveform analyzer and video test pattern image generator that
provides basic video measurement capabilities for a video capture/playback card in a PC equipped with Video for
Windows compliant video capture/playback hardware...
26
-------------------READ READ READ-------------------------------------------------------------------------------DigiLava Signs Letter of Intent To Acquire Strategic Partner -- Local Area Yellow Pages
----------------------------------------------------------------------------------------------------------------------------DigiLava, Inc. (Pink Sheets: DGLV), a leading San Francisco based Application Service Provider of rich-media
direct response marketing, today announced that the Company has executed a Letter of Intent to acquire Local Area
Yellow Pages (LAYP), which is currently utilizing DigiLava's proprietary technology platform.
LAYP is a comprehensive and invaluable Business Portal that allows consumers to quickly access local and online
business information, including maps, links to featured business websites and other valuable tools and resources. In
addition, LAYP offers a Business Internet Package (BIP), which consists of Unlimited Internet Access, Featured
Listing, Business Web Sites, Business Email System, and much more to business owners across the nation.
"The intent to acquire Local Area Yellow Pages is part of the company's strategic business plan to expand both its
technology offerings and business customer base. The continuous role-up strategy positions the Company to take a
proactive leading role in the Media and Internet market," stated Baldwin Yung, CEO of DigiLava. "Subject to final
due diligence and Board approval, the acquisition entails the retention of key business customer channel and a
significant revenue base."
"Based on the rapid growth in customer acquisition and escalating financials, we plan to leverage the capital and
expertise of DigiLava to expand our business customer channel and to integrate the next generation enabledapplication, which will provide superior technology solutions to benefit our current and prospective business
customers," stated Chris Chen, Vice President of Local Area Yellow Pages.
-----------------------------------------------------------------------------------------------------------------------------Disclaimer
-----------------------------------------------------------------------------------------------------------------------------Please be advised that nothing within this email shall constitute a solicitation or an offer to buy or sell any security
mentioned herein. This newsletter is neither a registered investment advisor nor affiliated with any broker or dealer.
All statements made are our express opinion only and should be treated as such. We may own, buy and sell any
securities mentioned at any time as well as parties related to this newsletter. This email includes forward-looking
statements within the meaning of The Private Securities Litigation Reform Act of 1995. These statements may
include terms as "expect", "believe", "may", "will", "move", "undervalued" "crazy" "going" "Speculative" "Target
Price" and "intend" "hot" "explode" or similar terms. This newsletter was paid $12350 from third party to send this
report. Please do your own due diligence before investing in any profiled company.
27
References
Antweiler, W., Frank, M. Z., 2004. Is all that talk just noise? The information content of
internet stock message boards. Journal of Finance 59, 1259-1294.
Barber, B. M., Odean, T., 2008. All that glitters: The effect of attention and news on the buying
behavior of individual and institutional investors. Review of Financial Studies 21, 785818.
Blume, M., Stambaugh, R., 1983. Biases in computed returns: An application to the size effect.
Journal of Financial Economics 12, 387-404.
Böhme, R., Holz, T., 2006. The effect of stock spam on financial markets. Unpublished
working paper, Technishe Universitat Dresden.
Brav, A., Lehavy, R., 2003. An empirical analysis of analysts’ target prices: Short-term
informativeness and long-term dynamics. The Journal of Finance 58, 1933-1967.
Engelberg, J., Sasseville, C., Williams, J., 2007. Attention and Asset Prices: The Case of Mad
Money. Unpublished working paper, Northwestern University.
Frieder, L. L., Zittran, J., 2007. Spam works: Evidence from stock touts and corresponding
market activity. Unpublished working paper, Purdue University.
Friedman, M., Savage, L. J., 1948. The utility analysis of choices involving risk. Journal of
Political Economy 56, 279-304.
Hanke, M., Hauser, F., 2008. On the effects of stock spam e-mails. Journal of Financial
Markets 11, 57-83.
Huberman, G., Regev, T., 2001. Contagious speculation and a cure for cancer: A nonevent that
made stock prices soar. Journal of Finance 56, 387-396.
Kumar, A., 2009. Who gambles in the stock market? Journal of Finance, forthcoming.
Lerer, L., 2007. Why the SEC can’t stop spam. Forbes.com, March 8.
Markowitz, H., 1952. The utility of wealth. Journal of Political Economy 60, 151-158.
Statman, M., 2002. Lottery players/stock traders. Financial Analysts Journal 58, 14-20.
Tetlock, P. C., 2007. All the news that’s fit to reprint: Do investors react to stale information?
Unpublished working paper, Yale University.
28
Table 1
Sample description
Panel A: Spam activity
Pooled
1999-2001
2002-2004
2005-2006
Number of unique spam events
Number of unique target companies
658
442
41
35
139
105
478
321
Number of duplicate messages
Mean
Median
29.9
4.0
1.6
1.0
8.7
1.0
38.5
9.0
Panel B: Stock Exchange
Exchange
Number
Percent
AMEX
NASDAQ
OTCBB
Pink Sheets
6
4
292
356
0.91%
0.61
44.38
54.10
Total
658
100.00%
Panel C: Weekday of spam delivery (first message for each unique spam)
Day
Number
Percent
Sunday
Monday
Tuesday
Wednesday
Thursday
Friday
Saturday
42
154
119
116
109
89
29
6.38%
23.40
18.09
17.63
16.56
13.53
4.41
Total
658
100.00%
This table presents description information about the frequency of spam over time and by unique
spam message (Panel A), the stock exchange target spam companies are listed on (Panel B), and
the frequency of unique spam by day of the week (Panel C).
29
Table 2
Attributes of spam messages
Panel A: Content
Pooled
Number of unique spam messages
658
(1)
(2)
1999-2001
2002-2004
41
(1)
vs.
(2)
139
(3)
2005-2006
(2)
vs.
(3)
478
Percentage of Unique Spam Messages Including (all figures represent % unless otherwise noted):
Company description
Fundamental analysis
Directly quoted press release
Indirectly quoted press release
Verifiable press release
Safe Harbor language
Disclaimer not using Safe Harbor language
Claim to have received monetary compensation
(Median dollars received)
Claim to have received share compensation
(Median number of shares received)
Name of group/individual responsible for spam
Quoted current price
Quoted recent price
Target price
Reference to another company touted by spammer
74
11
62
9
67
65
13
46
$10,000
14
200,000 shrs
40
70
6
31
6
30
63
5
34
17
49
24
44
27
$10,000
15
95,000 shrs
59
44
22
0
0
75
4
49
13
59
71
10
52
$12,000
17
100,000 shrs
55
54
9
9
3
*
***
***
***
***
*
**
75
14
68
8
70
66
12
46
$10,000
13
375,000 shrs
33
76
4
41
8
***
***
**
**
***
***
**
***
**
Table 2 − continued
Panel B: Target prices and implied returns (n = 192)
Quoted target price
Current price (observations with non-missing target price)
Implied return
Horizon of implied return (in days)
Mean
Median
Q1
Q3
$2.37
0.75
412%
23
$1.28
0.31
221%
3
$0.45
0.13
121%
1
$3.50
1.06
408%
5
Panel A presents descriptive evidence of spam characteristics along with test of differences over time of the frequencies in which
spam include a particular characteristic. Characteristics include company description, fundamental analysis which indicates whether
the spam includes a valuation model or some discussion of the drivers of value, press release related information either directly
attributed to a previous company specific press release (directly quoted) or not attributed to any particular source (indirectly quoted),
the number of press releases collected that agreed to the information in the spam (verified), safe harbor language or some other
disclaimer about the accuracy of the information in the spam, whether the spammer received monetary or share compensation and if
so, the amount, the name of the company/group sending the spam, a quotation of the current or recent price of the stock, the inclusion
of a target price which the spam indicates the price will reach over a certain horizon, and whether or not the spam includes the name of
another company the spammer had recommended in the past. Panel B provides further information on the frequency of spam that
include target prices along with descriptive statistics for target prices, current prices, returns implied by the difference between the
target price and current price (implied return), and the number of days until the stock will reach the target price (Horizon).
*, **, *** indicates significance at or below the 10%, 5%, 1% level, respectively.
31
Table 3
Characteristics of press releases quoted in spam messages
Panel A: Frequency of spam messages with quoted press release(s)
Number
Percent
Number of unique spam messages
658
100%
Directly quoted press release
Indirectly quoted press release
Total
406
62
468
62
9
71
Verified press release
Actual press release located
438
421
67
64
Panel B: Characteristics of press releases quoted in spam messages
Total
With Returns
Data
Actual press releases located
Number of press releases per spam message
Mean
Median
810
632
1.23
1.00
0.96
1.00
Source of press release
Company
Other
801
9
623
9
Weekdays between press release and spam message
Mean
Median
23.5
10.0
25.0
10.0
Panel C: Classification of news contained in quoted press releases
Total
Contracts
Earnings
Expansion
Products
General
Total
206
58
209
226
111
810
32
With Returns
Data
167
50
143
183
89
632
Table 3 − continued
This table presents information about the frequency and content of spam that include press
release information. Panel A presents frequencies of spam that included information from a
press release that was directly attributed to a particular source within the spam (directly quoted)
or that was identified as including press release information with attributing it to any source
(indirectly quoted). We verified the press release information via tying it back to an actual press
release (actual press release located) or identifying the information in an alternative source like a
journalist citation. Panel B documents the number of press releases cited, source, and time lapse
between the press release date and initial spam date for spam that include press release
information. Panel C categorizes the information within the press release cited by the spam into:
(i) Contracts, including the formation of partnerships and the execution of other contracts; (ii)
Earnings, including earnings announcements and management earnings forecasts; (iii)
Expansion, including acquisitions, exploration, and external financing; (iv) Products, including
marketing, products and service arrangements; and (v) General, including changes in
management and other events that do not fit in one of the other categories. If the press release
contained more then one news item, only the most significant item is coded.
33
Table 4
Spam date returns and abnormal volume
Panel A: Raw returns (n = 658)
Window
(–10,–2)
(–1,1)
(0,0)
(0,1)
(0,5)
(0,10)
Mean
0.13***
0.08***
0.04***
0.04***
–0.01
–0.05***
Median
0.00
0.00
0.00
0.00
–0.03***
–0.09***
Q1
–0.14
–0.10
–0.06
–0.09
–0.20
–0.27
Q3
0.14
0.14
0.08
0.11
0.11
0.09
Percent
Positive
Negative
43
48
47
44
43
42
44
44
39
54
33
61
Panel B: Abnormal volume (n = 573)
Day
–3
–2
–1
0
+1
+3
+5
Median
1.03
1.23
1.46
2.60
2.66
2.03
1.86
Q1
0.47
0.48
0.53
0.97
0.94
0.68
0.69
Q3
2.68
3.12
4.05
8.23
8.64
6.22
5.59
Panel A (Panel B) presents raw returns (abnormal volume) surrounding the first date a unique
spam is released (day 0). Percent Positive (Negative) represents the percentage of returns for a
given interval that are positive (negative). Summations within a row are less than 100% because
of returns of zero. Abnormal Volume is calculated as the share volume on a given day divided
by the median volume over the window (–20,–2); values greater (less) than 1 indicates relatively
high (low) volume. The total number of observations in Panel B, 573, is less than the 658 unique
spam observations in the sample because 85 observations have zero trading volume in the
benchmark period.
*, **, *** indicates significantly different from zero at the 10%, 5%, 1% level, respectively,
using two-tailed tests.
34
Table 5
Random date returns and abnormal volume
Panel A: Random date returns (n = 1,273)
Window
(–10,–2)
(–1,1)
(0,0)
(0,1)
(0,5)
(0,10)
Mean
0.00
0.01
0.00
0.00
0.00
0.01
Median
–0.04***
0.00***
0.00***
0.00***
0.00***
–0.05***
Q1
–0.18
–0.10
–0.06
–0.08
–0.15
–0.21
Q3
0.07
0.05
0.03
0.04
0.09
0.10
Percent
Positive
Negative
32
56
33
47
30
39
32
44
37
50
35
56
Panel B: Random date abnormal volume (n = 1,160)
Day
–3
–2
–1
0
+1
+3
+5
Median
1.00
0.97
0.92
0.98
1.02
1.04
1.03
Q1
0.37
0.33
0.25
0.29
0.29
0.33
0.34
Q3
2.63
2.47
2.53
2.74
2.92
3.13
2.89
Panel A (Panel B) presents raw returns (abnormal volume) surrounding randomly selected dates.
For each unique spam message, we randomly select one event date (day 0) preceding and one
following the spam date, excluding the 30 days surrounding the spam date. Returns (abnormal
volume) data are missing for 43 (156) of the 1,316 (= 658 × 2) randomly selected event dates.
Percent Positive (Negative) represents the percentage of returns for a given interval that are
positive (negative). Summations within a row are less than 100% because of returns of zero.
Abnormal Volume is calculated as the volume on a given day divided by the median volume
over the window (–20,–2); values greater (less) than 1 indicate relatively high (low) volume on
the date.
*, **, *** indicates significantly different from 0 at the 10%, 5%, 1% level, respectively, using
two-tailed tests.
35
Table 6
Effect of press releases and target prices on spam returns
Panel A: Press release
Press
Release
No
Yes
N
190
468
(–1,1)
Mean
Median
0.11***
0.00
0.06***
0.00
Return Period
(0,0)
Mean
Median
0.02
0.00
0.04***
0.00
(–1,1)
Mean
Median
0.05***
0.00
0.14***
0.00
Return Period
(0,0)
Mean
Median
0.02**
0.00
0.08***
0.01**
(0,1)
Mean
Median
0.04**
0.00
0.04***
0.00
Panel B: Target price
Target
Price
No
Yes
N
453
205
(0,1)
Mean
0.02*
0.07***
Median
0.00
0.00
Panel C: Press release and target price
Press
Release
No
Yes
No
Yes
Target
Price N
No 121
No 332
Yes
69
Yes 136
Return Period
(0,0)
Mean
Median
0.00
0.00
0.02**
0.00
0.05*
0.00
0.09***
0.03**
(–1,1)
Mean
Median
0.08**
–0.01
0.04**
0.00
0.17**
0.00
0.13***
0.02
(0,1)
Mean
Median
0.03
0.00
0.02
0.00
0.05
0.00
0.08*** 0.03
This table sorts the sample based on whether the spam message includes information from a
previously issued press release (Panel A), a target price (Panel B), or a combination of press
release and target price (Panel C). The return period windows surround the spam date (day 0).
*, **, *** indicates significantly different from 0 at the 10%, 5%, 1% level, respectively, using
two-tailed tests.
36
Table 7
Alternative measures of investor attention and spam returns
Panel A: Abnormal volume
Median
Abnormal
Abnormal
(–1,1)
Volume N Volume
Mean
Median
Low 191
0.70
–0.01
–0.02**
Middle 191
2.60
0.04**
0.01
High 191 15.08
0.16***
0.06**
Return Period
(0,0)
Mean
Median
–0.01
0.00***
0.01
–0.01
0.11*** 0.06***
(0,1)
Mean
Median
–0.01
–0.01**
0.01
0.00
0.10*** 0.03**
Panel B: Sorting on press release and target price, conditional on high abnormal volume
Press Target
Release Price N
No
No 36
Yes
No 82
No
Yes 21
Yes
Yes 52
(–1,1)
Mean
Median
0.06
–0.02
0.13***
0.03
0.15
–0.03
0.28***
0.22***
Return Period
(0,0)
Mean
Median
0.03
–0.01
0.09*** 0.05*
0.07
0.00
0.20*** 0.14***
(0,1)
Mean
Median
0.02
–0.05
0.08**
0.00
0.05
0.00
0.20*** 0.16***
(–1,1)
Mean
Median
0.02
–0.02*
0.08***
0.00
0.12***
0.02**
Return Period
(0,0)
Mean
Median
–0.01
–0.01***
0.05*** 0.00
0.06*** 0.00
(0,1)
Mean
Median
0.00
–0.02**
0.05*** 0.00
0.06*** 0.00
Panel C: Spam frequency
Median
Spam
Spam
Frequency N
Freq.
Low
197
1
Middle 242
4
High
219
43
Panel D: Sorting on press release and target price, conditional on high spam frequency
Press Target
Release Price N
No
No 23
Yes
No 97
No
Yes 37
Yes
Yes 62
Return Period
(0,0)
Mean
Median
0.04
0.00
0.03
0.00
0.10**
0.00
0.09*** 0.07***
(–1,1)
Mean
Median
0.20**
0.08
0.04*
0.00
0.27**
0.04
0.13**
0.04*
37
(0,1)
Mean
Median
0.09
0.03
0.00
0.00
0.13**
0.03**
0.09**
0.05*
Table 7 - continued
This table presents raw returns surrounding the first date a unique spam is released (day 0).
Panel A sorts the sample into terciles based on abnormal volume, calculated as the volume on the
spam date (day 0) divided by the median volume over the window (–20,–2). Panel B partitions
the sample with High abnormal volume on whether the spam message includes information from
a previously issued press release or target price. Panel C sorts the sample into terciles based on
the frequency of duplicate spam messages. Panel D partitions the sample with High spam
frequency on whether the spam message includes information from a previously issued press
release or target price.
*, **, *** indicates significantly different from 0 at the 10%, 5%, 1% level, respectively, using
two-tailed tests.
38
Table 8
Spam date returns and abnormal volume regressions
Variable
Intercept
Press
Implied Return_No Press
Implied Return_Press
Abnormal Volume
Lagged Abnormal Volume
Spam Frequency
Adj. R2
Nobs.
Return
Coeff. Est.
p-value
–0.068
0.001
0.018
0.309
0.007
0.766
0.035
0.047
0.046
< 0.001
0.005
0.225
0.106
573
Abnormal Volume
Coeff. Est.
p-value
0.665
< 0.001
–0.066
0.496
0.080
0.371
0.253
< 0.001
0.645
0.078
< 0.001
0.003
0.345
573
This table presents results of regressions of raw returns and abnormal volume surrounding the
first date a unique spam is released (day 0) on spam characteristics. Abnormal Volume is the
natural logarithm of 1 plus the volume on day 0 divided by the median volume over the window
(–20,–2). Press is an indicator variable equal to 1 if the spam message refers to information in a
previously released company specific press release, 0 otherwise. Implied Return_Press (No
Press) is the implied return for the sample of spam with (without) a press release. The implied
return is the natural logarithm of 1 plus the difference between the target price included in the
spam relative to the stock’s current price, and is set equal to 0 for observations without target
prices. Lagged Abnormal Volume is abnormal volume on spam day –1. Spam Frequency is the
natural log of the number of duplicate spam messages in the dataset for each unique spam event.
39
Table 9
Original press release returns and abnormal volume
Panel A: Raw returns (n = 632)
Window
(–10, –2)
(–1,1)
(0,0)
(0,1)
(0,5)
(0,10)
Mean
0.19***
0.08***
0.04***
0.05***
0.03**
0.00
Median
0.00
0.00**
0.00***
0.00*
-0.005**
-0.06***
Q1
–0.12
–0.07
–0.04
–0.07
–0.13
–0.21
Q3
0.22
0.16
0.08
0.12
0.12
0.10
Percent
Positive
Negative
0.47
0.46
0.49
0.41
0.47
0.32
0.47
0.40
0.41
0.50
0.37
0.59
Panel B: Abnormal volume (n = 574)
Day
–3
–2
–1
0
+1
+3
+5
Median
1.10
1.14
1.27
2.00
2.03
1.41
1.49
Q1
0.40
0.48
0.47
0.88
0.69
0.55
0.52
Q3
2.66
2.93
3.54
5.56
4.77
4.22
4.06
Panel C: News type
News
Type
Contracts
Earnings
Expansion
General
Products
N
167
50
143
89
183
(–1,1)
Mean
Median
0.07***
0.00
0.16***
0.06
0.07**
0.00
0.10***
0.00
0.07***
0.00
Return Period
(0,0)
Mean
Median
0.04***
0.00*
0.10***
0.04***
0.04***
0.00**
0.01
0.00
0.05***
0.00**
40
(0,1)
Mean
Median
0.04**
0.00
0.13***
0.04*
0.04**
0.00
0.05**
0.00
0.06***
0.00
Table 9 – continued
This table presents raw returns (Panel A) and abnormal volume (Panel B) surrounding the
original release date (day 0) for press releases subsequently quoted in the spam sample. Percent
Positive (Negative) represents the percentage of returns for a given interval that are positive
(negative). Abnormal Volume is calculated as the share volume on a given day divided by the
median volume over the window (–20,–2). Because some spam messages quote more than one
press release, the number of original press releases exceeds the number of spam messages with
an old press release (468, per Table 3). The total number of observations in Panel B, 574, is less
than the 632 original press release observations in Panel A because 58 observations have zero
trading volume in the benchmark period. Panel C partitions the sample based on the content of
the news according to the classifications in Table 3.
*, **, *** indicates significantly different from 0 at the 10%, 5%, 1% level, respectively, using
two-tailed tests.
41
Table 10
Determinants of stocks targeted by spam
Panel A. Descriptive statistics for spam and control samples
Variable
Ret(–1)
Ret(–2)
logDolVol(–1)
logDolVol(–2)
NumPR(–1)
NumPR(–2)
PR
(1)
Spam Sample
(n = 397)
Mean
Median
0.23
0.00
0.22
0.00
12.30
12.63
11.55
12.04
0.39
0.00
0.44
0.00
38.0%
(2)
(3)
Spam Control Sample Random Control Sample
(n = 397)
(n = 397)
t-tests of Means
Mean
Median
Mean
Median
(1) vs. (2) (1) vs. (3)
0.14
0.00
0.06
0.00
***
0.19
0.00
0.04
0.00
***
9.26
10.76
7.39
8.62
***
***
9.43
10.97
7.81
9.06
***
***
0.23
0.00
0.12
0.00
***
***
0.20
0.00
0.09
0.00
***
***
18.9%
12.1%
***
***
Panel B. Logit regression results of spam sample versus control samples
Spam Sample vs. Spam Control
Variable
Intercept
Ret(–1)
Ret(–2)
logDolVol(–1)
logDolVol(–2)
PR
Coeff. Est.
–2.33
–0.01
0.07
0.31
–0.12
0.59
Std. Err.
0.298
0.080
0.074
0.046
0.041
0.177
p-value
< 0.001
0.943
0.356
< 0.001
0.004
< 0.001
42
Spam Sample vs. Random Control
Coeff. Est.
–2.99
0.24
0.19
0.33
–0.07
0.89
Std. Err.
0.301
0.127
0.125
0.046
0.041
0.205
p-value
< 0.001
0.055
0.125
< 0.001
0.084
< 0.001
Table 10 – continued
This table presents descriptive statistics (Panel A)and logit regression analysis (Panel B) for the spam sample relative to two control
samples. The Spam Sample includes the first spam message for each ticker. The Spam Control Sample consists of random dates from
6 to 18 months before or after the spam date for our stock spam sample. The Random Control Sample consists of a random sample of
stocks not identified as spammed securities trading on the OTC Bulletin Board or PinkSheets on randomly chosen dates. Returns
(Ret), the log of dollar volume (logDolVol), and the number of press releases (NumPR) are calculated for each of the two months
preceding the event date (designated with subscripts –1 and –2), excluding days –1 and 0 in the event month. Returns are winsorized
at the top and bottom 1% because of the existence of extreme outliers. PR is an indicator variable equal to 1 if the firm issued a press
release in the two months preceding the event date, and 0 otherwise.
*, **, *** indicates significantly different from 0 at the 10%, 5%, 1% level, respectively, using two-tailed tests.
43