ArticlePDF Available

How Much Do Investors Trade Because of Name/Ticker Confusion?

Authors:
  • Rutgers, The State University of New Jersey, Camden, United States

Abstract and Figures

We conduct a search for pairs of companies with similar names/ticker symbols. Between 12% and 25% of such pairs exhibit co-movements in trading turnover, which we attribute to investor confusion. We estimate that trades made by mistake contributed to 5% of the trading turnover. The three-hour CARs for the company chosen by mistake around the time intervals with extreme returns for the paired company are 0.5%. The confusion is highest for large companies and around time intervals with high turnover. We show that when the cause of confusion disappears, the co-movement in turnover also disappears. Forthcoming in the Journal of Financial Markets JEL classification: G10, G14.
Content may be subject to copyright.
How Much Do Investors Trade Because of Name/Ticker
Confusion?
Vadim S. Balashov
Andrei Nikiforov
Rutgers School of Business-Camden
May 25, 2019
Abstract
We conduct a search for pairs of companies with similar names/ticker symbols.
Between 12% and 25% of such pairs exhibit co-movements in trading turnover, which we
attribute to investor confusion. We estimate that trades made by mistake contributed
to 5% of the trading turnover. The three-hour CARs for the company chosen by mistake
around the time intervals with extreme returns for the paired company are 0.5%. The
confusion is highest for large companies and around time intervals with high turnover.
We show that when the cause of confusion disappears, the co-movement in turnover
also disappears.
Forthcoming in the Journal of Financial Markets
JEL classification: G10, G14.
Keywords: Ticker Symbol; Investor Confusion; Similarly Identified Securities;
Market Inefficiency; Investor Irrationality
Corresponding author contact information: Vadim S. Balashov, Assistant Professor of Finance, Rutgers
School of Business-Camden, Rutgers University-Camden, 227 Penn Street, Camden, NJ 08102, work (856)
225-6706, cell (225) 202-8432, e-mail: vadim.balashov@rutgers.edu
Andrei Nikiforov, CFA, Clinical Assistant Professor of Finance, Rutgers School of Business-Camden,
Rutgers University-Camden, 227 Penn Street, Camden, NJ 08102, work (856) 225-6594, e-mail: and-
nikif@camden.rutgers.edu
We are grateful to Tarun Chordia, Jennifer Conrad, Wei Xiong, Wei Jiang, Ron Kaniel, Tyler Shumway,
Stephen J. Brown, Bing Han, Richard Warr, Suzanne S. Lee, participants of the 2018 SFA conference in
Asheville, NC, participants of the finance workshop at the Rutgers School of Business-Camden, David Ped-
ersen, Zhanel DeVides, Eugene Pilotte, Ivo Jansen, Jun Guo, and anonymous referees for helpful comments
and discussion. We thank Brad Barber and Terrance Odean for generously sharing their discount brokerage
dataset with us. All errors remain our own. We would like to thank the Whitcomb Center for Research in
Financial Services at Rutgers University for funding the data for this paper.
1
1 Introduction
On October 4, 2013, the stock of a long-closed electronics retailer Tweeter Home Entertain-
ment Group (Ticker Symbol: TWTRQ) suddenly soared 1,400%. The sudden increase in
investor interest in the company and the unusual return happened the day after another,
unrelated, company Twitter Inc. (proposed Ticker Symbol: TWTR) filed papers detailing
its plans for a 1 billion initial public offering (as reported by Isidore (2013) to CNN). The
phenomenon can be attributed to investor confusion, as investors apparently mistook the
penny stock for Twitter’s upcoming stock offering.
In another example, on May 2, 2007, Graco Children’s Products Inc., a subsidiary of
Newell Companies Inc. (Ticker Symbol: NWL) announced a recall of its soft block tower
toys. In the upcoming days, the stock for an unrelated fluid-handling systems manufacturer,
Graco Inc. (Ticker Symbol: GGG) dropped by over 2.5%. Even popular news aggregators
such as Google Finance incorrectly listed the news about Graco Children’s Products alongside
the Graco Inc. stock quote, showing that confusion is not limited to individual investors.
Besides sharing the name, these two companies have nothing in common.
There are other instances of stock confusion among investors. Ford Motor Company
(F), the automobile manufacturer, is often confused with Forward Industries Inc. (FORD),
a designer and distributor of custom carry and protective solutions. Hewlett Packard Co.
(HPQ), an IT company, is often confused with Helmerich & Payne Inc. (HP), a petroleum
contract drilling company.1
In a well-researched case study, Rashes (2001) documents that between 1996 and 1997,
two unrelated companies, MCI Communications (MCIC), a telecommunications company,
and Massmutual Corporate Investors (MCI), a closed-end mutual fund, showed unusual co-
movements in turnover and prices. He investigates the issue and attributes the co-movements
to investor confusion between the two stocks. Moreover, he shows that deviations from the
1See also the following news about name/ticker confusion: Ewing (1999), Levenson (2014), and Levine
(2019), among others.
2
intrinsic value in the smaller company, MCI, lasted days as investors realized their mistakes.
These and other similar examples are often used to point to investors’ irrationality and as
evidence against market efficiency. The academic study by Rashes (2001) can now be found
on syllabi of graduate classes on psychology and economics alongside other papers that
constitute the fundamentals of the behavioral school of thinking in finance.2However, it
is important to recognize that Rashes (2001) is not a systematic study of the phenomenon,
but rather a well-executed singleton case study. The main question of how common this
behavioral phenomenon is in the U.S. stock market has remained unanswered. Is his finding
just a small anomaly with no larger significance or is it a systematic component of investors’
decision making? Therefore, it is important to study how prevalent this particular behavior
is.3
This paper attempts to quantify the presence of confusion trading and investigates re-
search questions that remain open in the literature: How common is it for companies to have
identifiers similar to those of other companies? Are the examples documented by Rashes
(2001) anecdotal or do they represent ordinary occurrences? How often do investors confuse
companies with similar identifiers and what are the determinants of such confusion? How
much trading is due to confusion? How much do confusion trades cost investors?
We start by casting a wide ‘net’ through screening for companies that share parts of their
names and/or ticker symbols with another company. Even after excluding the most common
parts of the names, such as ‘Corp.’, ‘Inc.’, and so forth, we find that the phenomenon of
having similar names/tickers is quite common. Companies that share some meaningful part
of their names and/or ticker symbols with another company constitute over 55% of publicly
traded firms. To limit our sample of firms to a more manageable size that will allow us
to conduct a detailed intraday analysis, we apply several criteria to zoom in on pairs of
2E.g. at the UC Berkley’s “Psychology and Economics 219” course. The syllabus can be found here:
http://eml.berkeley.edu/˜webfac/dellavigna/e219b s12/219BSyllabusSpring2012 Jan12-2.pdf
3Davies et al. (2007) applied Rashes (2001)’s analyses to 29 pairs in the UK market. In contrast to our
study, they did not use intraday data. They find statistically insignificant results and conclude that any
trade by confusion, if present, can be confined to shorter, intraday, intervals. Because of the use of daily
regressions, the applicability of their results to the larger universe of securities is unclear.
3
companies that represent the greatest likelihood of confusion. Such companies are the best
candidates with which to document trades made by mistake.
We use five types of possible similarities between the names and tickers of companies to
identify the pairs most likely to be mistaken by investors. After running our identification
algorithm we are able to zero in on 254 pairs with the greatest similarities. Unlike Rashes
(2001), we use intraday data to capture possible co-movements in turnover and returns.
Using intraday data increases the chances of detecting confusion because it may be too
short-lived to be captured by daily closing prices (Chordia et al. (2014)).
We document that, out of the 254 candidate pairs with the highest possible similarities,
31 pairs (or 12.2%) seem to exhibit a statistically significant co-movement in turnover that,
when controlled for other factors, can be attributed to trades made by investors who mistake
one company for another. We find that in some cases, the phenomenon happens only around
time intervals containing extreme trading turnover in the bigger company and is not present
for the rest of the time (this is true for 15 additional pairs in the sample). We also show that
in 18 additional cases, the investor reaction involving the smaller company occurs with a
one- to three hour delay after news concerning the bigger company is released in our sample
period. Including these 33 pairs brings our estimate of pairs that cause investor confusion
up to 25% of identified potential candidates. Given how widespread similarities between
companies’ names and tickers are, the finding suggests that trades made by mistake are not
anecdotal and have a more systematic nature.
We show that such confusion is most evident around periods with extreme trading
turnover and extreme returns for the bigger company of the pair and is less evident around
more common events such as analysts’ forecasts, recommendations, and release of earnings
reports.
For 56 out of the original 254 and for 18 out of the 31 significant turnover pairs, we
document that returns also exhibit significant co-movements. The cumulative abnormal
return for the smaller (“mistaken”) company around times with extreme returns in the
4
bigger company ranges between 0.1% and 0.6% for positive news (between -0.1% and -0.6%
for negative news) for the three-hour window around the event. We document that, on
average, it takes over a week for returns to reverse. We also show that confusion trading is
not just a retail investor phenomenon. Institutional investors are also subject to confusion
trading.
The market-wide setup of our study and screening for companies with the highest degree
of similarities allows us to investigate determinants of the strength of investor confusion. We
find that the larger the size of the smaller company in a pair, the stronger the confusion.
We also show that the probability of confusion is higher if two firms are listed on the same
exchange and for companies with lower institutional holdings. Analyst coverage also increases
the probability of confusion. At the same time, return co-movements are more evident in
pairs for which stock returns are less volatile.
By our estimates, the trades made by mistake, on average, cost investors 1.1M per
pair per year in transaction costs. This figure typically constitutes around 5% of the total
transaction costs for the smaller company of the pair. The fraction of trades due to confusion
in the smaller company is similar at 5%. Compared to the documented number of trades
made by retail investors of just 2% (Evans (2009)), the reported number is economically
large and indicates that confusion trading is not driven only by retail investors.
For robustness, we also look at pairs in which one (or both) of the companies change
their names/tickers so that they are no longer confusing, and we show that when the cause
of confusion disappears, the phenomenon also goes away.
Our study has broad policy implications and is of interest to the broad community of
market participants. First of all, we draw investors’ attention to the existing phenomenon
of mistaking one company for another and urge them to be careful when placing orders
for companies with similar tickers and/or names. As noted byLevine (2019), even fast
algorithmic traders are not free of this type of error, let alone individual retail investors. Our
study also points out that brokerage houses can do more to prevent trades made by mistake.
5
We provide a list of companies that are most likely to be subject to confusion trading and
recommend using additional checks and/or warnings when investors try to execute a trade
for those stocks, especially for large trades and high transaction costs. We would also like
to draw attention away from the companies on the list themselves or any other company
that possibly shares part of its name/ticker symbol with another firm to focus on general
awareness of the potential confusion. The impact may be especially big for the smaller
company in a pair. Firms might be willing to avoid any price movements not associated
with new information related to the firm itself. One approach for these firms is to change
their name and/or ticker. Our study shows that when firms do change their names and/or
tickers, confusion disappears.4
The documented short-term predictability of returns and a sizable reaction ( 0.5%) in
the SMALL stock to news related to the BIG stock indicates inefficiencies in the markets.
These inefficiencies should be of interest to arbitrageurs. In the modern era of high-frequency
trading and low transaction costs, computer algorithms seek opportunities to exploit even
the tiniest profits actively. Actively trading off the mispriced confused stock will generate
profits for arbitrageurs and eliminate the inefficiency. Our study could also be of interest to
the regulators of financial markets. Rashes (2001) documents that the NYSE was contacted
about changing the ticker symbol for MCI/MCIC. Yet, the NYSE did not believe the amount
of confusion generated was significant enough to justify the change. Our study shows that
at least for some pairs, such confusion is sizable and persistent. Finally, this paper is of
interest to behavioral and financial researchers. We document that companies with similar
ticker symbols and/or names represent more than one-half of all publicly traded stocks. If
confusion trading is a phenomenon that is not limited to just the sample of stocks described
in this study, but is indeed a much broader phenomenon, it may have significant implications
for asset pricing. To our knowledge, our paper is the first to measure the presence, scale,
4On October 8, 2013, FINRA changed Tweeter’s ticker symbol to THEGQ. There are other cases when
companies changed their names and/or tickers because of potential confusion with another stock. For
example, AppNet Systems was confused with Appian Technology and changed its ticker symbol from APPN
to APPG on March 31, 1999.
6
and economic significance of investor confusion.
2 Data and Variables
2.1 Sample
We start by searching for pairs of publicly traded companies on the CRSP database which
have similar names and/or ticker symbols. First, we identify companies with similar ticker
symbols. This list includes companies for which the ticker symbol of one company contains
the entire ticker of another company (with at least three letters), plus an extra letter or
two. For example, ABM-ABMD. We were able to identify 5,703 pairs of that nature with
7,589 unique companies (some companies’ tickers were matched to the tickers of several
other companies). Another example of similar ticker symbols includes pairs for which both
companies have ticker symbols of at least four letters in length. Here, the ticker symbol
of one company contains all the letters of the ticker symbol of another company with the
same two first letters but with the last two letters switched. For example, TGCI-TGIC. The
search provides 375 pairs of that nature for 739 unique companies.
We then move on to find companies with similar names. This process is more complicated
because many companies share common words in their names such as “First,” “American,” or
“Financial.” We want companies to share the “meaningful” parts of their names. Therefore,
we proceed with screening for companies that share parts of their names and exclude the
most commonly shared words. Even after we filter out the first 100 most common words in
company names, the search still provides 10,493 companies.
Finally, we look for companies for which a ticker symbol of one company is part of the
name of another company. There are 84 pairs like that, attached to 165 unique companies.
For all pairs in our original search, we require that the companies in each pair had overlapping
periods of existence. Combined together, our search shows that there are some 14,437
companies that share some part of their identifying information (name or ticker symbol)
7
with another company. This constitutes over 55% of all stocks that were being publicly
traded during the sample period 1993-2013. 5We conclude that the majority of companies
on the market share some part of their name or/and trading symbol with some other company
and that the phenomenon is common.
Of course, not all identified companies will be confused by investors. We proceed with
narrowing down our search and selecting company pairs with the highest degree of similari-
ties. We believe that these companies can be confused by investors more easily than others.
We require that the intraday stock prices are available on the NYSE TAQ Trade database
between 1993 and 2013. To avoid any co-movement caused by intra-industry factors, we
require that companies in each pair belong to two different two-digit SIC-groups.
We apply five filters and identify five sources (“types”) of confusion. The first type
includes pairs for which the ticker symbol of the smaller company of the two is part of the
name of another company (the bigger one of the pair). The search gives us 26 pairs of this
type. The infamous example of a confused pair that falls into this type is the Ford Motor
Company (F) and Forward Industries Inc. (FORD).
The second type is similar to the first one, with the difference being that the ticker
symbol of the bigger company of the two is part of the name of the smaller company. There
are 34 pairs that could fall prey to this second type of confusion. In some cases, there is
double-confusion: a ticker for one company is part of the name of another company, and, at
the same time, the ticker of the paired company is part of the name of the first company.
An example of a pair exhibiting these traits would be Witco Chemical Corp. (WIT) and
Wit Capital Group Inc. (WITC). Besides sharing their names and ticker symbols, these two
companies have nothing in common.
For the third type of confusion, we require that three conditions are met. First, both
companies’ ticker symbols should consist of at least three letters. Second, the ticker symbol
of one company is the ticker of another company plus an extra letter or two. These two
5We use the number of stocks available on CRSP to get the estimate, 26,236.
8
requirements by themselves are very broad and create too many matches to be useful. At
the same time, we believe that the requirements by themselves are not representative of the
true confusion. Therefore, we add the extra requirement that the two companies should
share parts of their names. We further manually check that the shared part of the name
is meaningful and excluded matches that are simply generated by common words such as
“Inc.” or “Corp.” The search provides 182 pairs. Rashes (2001)’s MCI-MCIC pair is in this
category.
The fourth type of confusion includes pairs for which both companies have ticker symbols
containing at least four letters. Here, the ticker symbol of one company is the ticker symbol of
another, with the last two letters switched. Again, we further require that the two companies
share a meaningful part of their names. There are 8 pairs in this group, and an example is
the Victoria Bankshares Inc. (VICT) and Victoria Creations Inc. (VITC) pair.
Finally, we search the media for news about confused stocks that were not identified by
our algorithm. The search gives us two pairs: Hewlett Packard Co. (HPQ and its previous
ticker symbol HWP) is often confused with Helmerich & Payne Inc. (HP), and Newell
Companies (the owner of the Graco Children’s products brand, NWL) with its confusing
pairing, Graco Inc. (GGG).
We also check companies’ PERMNOs and PERMCOs on the CRSP database to make
sure they are not different issues of the same company. In total, this gives us 254 candidate
pairs of companies.
We recognize that there are possibly other types of confusion between companies. For
example, Davies et al. (2007) document that companies may be confused when their names
are spelled differently but sound similar. Even though other types of confusion are not
included in our search, we believe that our sample size is large enough to test the major
hypotheses of the paper and that it can provide a lower bound for investor confusion.
Two possible kinds of erroneous trades may contribute to co-movements in turnover. The
first kind, “true” investor confusion, occurs when investors mistake one company for another
9
and trade irrationally believing that they are buying or selling stocks in the right company.
The second kind is due to the phenomenon called “fat fingers,” which occurs when investors
enter the wrong ticker symbol into their trading system as a genuine mistake (a typo). Rashes
(2001) argues that the number of shares being traded via mistakes of the first kind is much
greater than the number being traded due to the second type. Davies et al. (2007) note
correctly that any mistakes that are a result of fat fingers are more likely to be evident over
a short horizon and are less likely to be a reason for co-movements in daily studies. Often,
investors will realize their mistake within minutes and correct themselves. This will result
in two transactions being recorded: one for the mistaken trade, and one when the trade is
corrected. In many cases, however, as noted by Rashes (2001), there are safeguards against
this type of mistake. The goal of our study is not to differentiate between the two kinds
of mistakes. Instead, we are trying to identify how much of the trading turnover can be
attributed to both kinds, collectively. Internet Appendix A provides the list of all pairs in
our sample.
2.2 Intraday Prices, Returns, and Turnover
We use the NYSE TAQ Trade database to obtain intraday stock prices. To build relative
announcement returns, we create 10-minute interval prices by converting the TAQ trade-by-
trade prices. There are 39 10-minute intervals in a trading day. We form intraday prices
at interval times every 10 minutes (P9:40,P9:50,. . . ,P15:40,P15:50 ,P16:00 ) using the nearest
TAQ price within 5 minutes of the interval time. In cases when there are multiple prices in
the specified second, we compute the volume-weighted average price. If no trades occurred
for some 10-minute interval, we set the corresponding interval return to zero. However, we
exclude observations for which more than 10% of the 10-minute interval prices are missing to
exclude possible liquidity effects. We employ the standard filters used in the microstructure
literature: we exclude the first and last trades as well as trades that occur outside of regular
trading hours. For each 10-minute interval, we compute the total volume for each company
10
in the pair and the market. To calculate turnover, we divide the total volume obtained by
the number of shares outstanding on the same day from CRSP. Using turnover instead of raw
share volume helps in controlling for events such as stock splits. Finally, for easy comparison
between different pairs, we standardize dependent and independent turnover variables by
subtracting the mean and dividing by the standard deviation within each pair.
2.3 Additional Variables
For each pair, we document the market capitalization of each company (the proxy for the size
of the company, measured on the latest date of the coexisting period for each pair), the start
and end dates for the period when the confusion was present, and the two-digit SIC group.
For each company in our pair, we then search for corporate events that can be identified from
electronic sources. We recognize four kinds of such events: earnings announcements (EAs),
company issued guidance (CIG), forecast revisions, and recommendation changes issued by
financial analysts. For earnings announcements, forecasts, and recommendations, we search
the I/B/E/S Actuals file augmented by the FirstCall Historical Database (FCHD) Actuals
file (for the years 1993-2011). CIGs are from the FCHD CIG file (up until 2011). Using
the time-stamp, the found events are merged with the TAQ sample for the corresponding
10-minute interval. We classify events as “positive” news if the cumulative abnormal return
(CAR) in the 210-minute window (21 intervals) centered around the event is positive, and
as “negative” news if it is negative. We also identify the exchange on which each company
in our sample is listed, and we calculate the firm volatility as the standard deviation of the
10-minute stock return for the sample period.
We also calculate the mean percentage of institutional holdings for each company in
the pair during the co-existing period. In order to do that, we obtain the total number
of shares held by institutional investors at the end of each quarter from the Thompson
13F Institutional Holdings data set and divide this number by the total number of shares
outstanding. Finally, we calculate the average analyst coverage for each company as the
11
mean number of analysts who issue forecasts on the I/B/E/S Detailed U.S. file during each
year of co-existence.
3 Identifying Significant Co-Movers
3.1 Turnover Co-Movements
We believe that any confusion in the name between two companies, if present, is easier to
identify when it concerns the smaller company. We do not claim that the other direction
is not possible, especially when the two companies are of a similar size. However, following
Rashes (2001) and Davies et al. (2007), we will conduct our analysis searching for the
evidence of confusion concerning the smaller company in each pair. Throughout the study,
we will denote the bigger of the two companies in a pair as “BIG,” and the smaller one as
“SMALL.”
Similar to the daily tests in Rashes (2001) and Davies et al. (2007), we start by identifying
pairs which exhibit significant co-movements in turnover. We also believe that trading
turnover regressions are better at capturing erroneous trades made out of confusion than
return regressions for several reasons. Turnover is a one-direction variable, meaning that
any time a trade occurs, it increases in value. Stock prices move only when there is a
significant imbalance in the buy and sell trade orders. If uninformed investors do not create
the imbalance, the returns will not exhibit a co-movement6. Alternatively, an investor may
quickly realize his/her mistake and correct it immediately (with a similar trade, but in reverse
order). If both transactions are within minutes of each other, the total return reaction for
the 10-minute interval will be negligible. Both trades, however, will increase the total volume
for the interval and will increase the turnover co-movement between the two similar firms.
Finally, one of the goals of our study is to obtain an estimate of how common trades made
6We test the hypothesis that return correlations are lower for highly volatile firms later in the paper, and
find strong empirical support for it.
12
as a result of confusion are. Turnover equations are most useful for this goal, and we will
test return correlations later in the paper.
To gauge trading turnover correlations within a pair, we first estimate the following
baseline regression separately for each pair iin our sample:
T urnoverSM ALLi,t =β0+β1T urnoverB IGi,t +β2T urnoverM arketi,t +
+β3T imei,t +β4Y eari,t +β5EventsS M ALLi,t +εi,t (1)
where T urnoverBI Gi,t is the turnover for the concurrent trading interval for the BIG
company in the pair, and T urnoverM arketi,t is the turnover for the market in the same 10-
minute interval t. Studies (Harris (1986); Jain and Joh (1988); McInish and Wood (1990a);
McInish and Wood (1990b); Wood et al. (1985)) have documented the U-shaped structure
of intraday volume levels. To avoid observing correlation due to the U-shaped volume, we
add time interval fixed effects throughout (T imei,t is a set of dummy variables, one for each
10-minute time interval). Y eari,t is a set of year fixed effects. We add year fixed effects to
capture the change of turnover over longer horizons. Further, Eventsi,t is a set of dummy
variables (fixed effects), with one for each of the four corporate events: EAs, CIGs, forecasts,
and recommendations. A dummy variable equals one if a corporate event in the company
coincides with time interval t. To eliminate any opening and closing effects, we exclude the
first (9:30-9:40 AM) and last (15:50-16:00) intervals.
Because of the multiple-hypotheses testing nature of the empirical setup in our study,
we should expect 25 pairs out of 254 to produce significant correlations by chance alone (at
the 10% level, see Hou et al. (2017)). To test if our results are indeed driven by confused
trades, we need to show that pairs from the matched sample have a higher likelihood of
co-movement than a randomly assigned pair. To do that, we create two samples of matched
pairs. The first sample consists of the original 254 pairs produced by our matching criteria.
The second sample is created by randomly matching a SMALL company to any of the 254
13
BIG companies in the original list.7We call this sample the control group. We find that in
the control group, 24 pairs have significant co-movements in turnover, which is predicted by
the multiple-hypotheses testing. Figure 1 compares the distributions of the t-values for the
original 254-pair treatment group and for the 254-pair control group. We also run a t-test
for the difference in the means for the distribution of t-values for the two groups and find
that they are significantly different at the 5% level.
[Insert Figure 1 around here]
Coefficients β1are estimated with error. While the error may by zero, on average, it is
almost certainly positive once we condition on the largest values of β1. Therefore, we will
overestimate the effects of confusion when stating conclusions based on the largest values of
β1. To avoid bias, we use a two-step approach (Faraway (1998)).8For each pair, we first
randomly split the sample into two sub-samples. We identify pairs with significant β1in
the first step. For the identified pairs, we re-estimate β1in the second step and use these
values for future inference. Note that with this approach, we are expected to see just three
(=254x0.1x0.1) significant pairs in the second step by chance only.
Table 1, column 2, presents the results of estimating β1in step one. We were able to
identify 41 significant pairs in the first step. For 31 out of the 41 pairs, β1is significant
in the second step. The results are presented in Table 1, column 3. Among the pairs with
the highest degree of confusion, are NAFC/NASH, MCIC/MCI, JPM/JPMX, as well as
WIT/WITC and PSX/PSXP. We conclude that for the identified 31 pairs, the co-movement
is driven by confusion trading. For the remainder of the paper, unless stated otherwise, we
compare results obtained for the 31-pair treatment group with the control group of randomly
assigned pairs.
[Insert Table 1 around here]
7We still make sure that the two companies in the randomly assigned sample co-existed during the same
period and that they have different SIC codes.
8We thank an anonymous referee for suggesting this method.
14
3.2 Extreme Turnover and Delayed Reaction
If the majority of the trades made because of confusion are done by uninformed and careless
retail investors, then an argument can be made that such investors do not follow companies
on an uninterrupted intraday basis. Instead, these noise traders react only to noticeable
news around the BIG company and by mistake make trades related to the SMALL one. If
this is the case and the co-movement happens only around high turnover intervals for the
BIG company, then our specification in regression 1 may result in an insignificant β1. To
test this, we follow Rashes (2001) and Davies et al. (2007) and look for evidence of confusion
around time intervals with unusually high trading turnover for the BIG company.
We define high turnover time intervals as intervals for which the abnormal trading
turnover (the residual from regression 4, see Section 3.3) is above a certain percentile for the
BIG company, over the sample period. We use the 99th, 99.9th, and 99.95th percentiles,
which correspond to 1% (one interval in 2.5 trading days, on average), 0.1% (one in 25 trad-
ing days), and 0.05% (one in roughly 50 trading days) of the observations in the sample,
respectively. We call intervals with unusually high abnormal turnover “extreme turnover”
(ET) intervals. We then estimate the following specification:
T urnoverSM ALLi,t =β0+β1T urnoverB IGi,t +β2T urnoverB IGi,t xETBIGi,t +
+β3T urnoverMarket i,t +β4T imei,t +β5Y eari,t +
+β6EventsSM ALLi,t +εi,t (2)
where ETB I Gi,t is a dummy variable, which equals one if the time interval corresponds
to the extreme turnover in the BIG company. T imei,t ,Y eari,t, and Eventsi,t are time, year,
and events fixed effects, respectively.
The results of estimating regression 2 using the two-step approach are presented in Table
2, columns 3-5. The plus sign indicates that β2is positive and significant in specification 2 in
both steps. For each pair we verify that the sum of the coefficients β1+β2>0. Twenty-five
15
out of the 31 original pairs have a positive and significant coefficient β2, indicating that
the confusion is indeed higher around events with extreme turnover in the BIG company.
Besides the original pairs, the coefficient β2in front of T urnoverB IGi,t xETBIG i,t is positive
and significant for 12 pairs when the 99 percentile is used as a cutoff, for an additional two
pairs when 99.9% is used, and for an additional pair for the 99.95% cutoff, giving a total of
15 pairs.
It may be the case that the 10-minute interval is too narrow to capture any co-movements
between the stocks due to confusion. If investors react to the news concerning the BIG
company with a delay of more than 10-minutes, our regression specification 1 will fail to
capture any trades of that nature. It should be noted that widening the window has a two-
fold result. On the one hand, it will allow us to capture trades made with a delay. On the
other hand, if trades are done by mistake and investors almost immediately realize that a
mistake has been made, and correct themselves within minutes (if not seconds), widening
the window will lower the significance of the coefficient β1. To test this, we run regression 1
with an extended window for T urnoverSM ALL:
T urnoverSM ALLi,4t=β0+β1T urnoverB IGi,t +β2T urnoverM arketi,t +
+β3T imei,t + +β4Y eari,t +β5EventsS M ALLi,t +εi,t (3)
where T urnoverSM ALLi,4tis the total trading turnover for the SMALL company for the
window that starts with interval tand spans 4tintervals. For 4t, we try 6, 12, and 18
intervals, which corresponds to one, two, and three trading hours, respectively.
The results of estimating regression 3 using the two-step approach are presented in Table
2, columns 6-8. Again, the original 31 pairs are significant in most specifications. By
extending the window to one hour, we gain nine pairs. Extending the window to two hours
gives an additional six pairs. Finally, for the three-hour window, we obtain three more pairs
for which the co-movement is significant. In unreported results, we find that extending the
16
window beyond the three trading hours does not gain any significant co-movements. We
conclude that if any delayed reaction to confusion occurs, it occurs within three trading
hours. In total, we identify 18 pairs for which there is evidence of delayed reactions.
[Insert Table 2 around here]
Another reasonable question to ask is whether confusion lessens over time. Davies et
al. (2007) argue about Rashes (2001)’s paper that “the very publication of the paper might
have alerted investors to be more careful when executing their trades.” (p. 695). If investors
learn about confusion pairs and the mistakes made in the past, we should see less confusion
with time. Another factor that may contribute to investor confusion could be the advent of
high-frequency trading. If high-frequency algorithms are less prone to making mistakes, we
also should see less confusion over time. To test this hypothesis, we add an interaction term
for time and T urnoverB I Gi,t in 1. The coefficient is negative and significant for 12 out of the
31 original pairs but is positive and significant for 15 out of the 31 pairs (MCIC/MCI is one
of these pairs, so we reject Davies et al. (2007)’s conjecture). We conclude that there is no
systematic evidence that confusion lessens with time.
3.3 Abnormal Turnover Around Extreme Turnover Intervals, Ex-
treme Returns, and Events
For each company iin our sample, we first run the following regression of turnover:
T urnoveri,t =β0+β1T urnoverMarket i,t +β2T imei,t +
+β3Y eari,t +β4Eventsi,t +εi,t (4)
The residual from estimating equation 4 is referred to as the “abnormal” turnover. To
illustrate our findings from the previous section, consider Figure 2. Figure 2 shows the
abnormal trading turnover for the SMALL company around the time intervals with ex-
treme (99.95th percentile) turnover in the BIG company. To be able to aggregate abnormal
17
turnover across companies of different sizes, we normalize it by dividing it by the corre-
sponding standard deviation (the mean is already zero). The resulting average normalized
abnormal turnover (NAT) is plotted around ETs for the BIG company. Time zero of each
panel corresponds to the time interval of the ET for the BIG company. The figure demon-
strates the NAT in the window [-20,+20] for 10-minute time intervals around ETs, which
corresponds roughly to one trading day. The figure shows the average abnormal turnover
for the 31 significant pairs from baseline regression 1 and for the control group of randomly
assigned pairs. Observe that for the times when the BIG company experiences extremely
high turnover, the SMALL company has a well-defined positive spike in turnover as well.
When compared with the control group, the spike is much higher for the sub-sample of the
31 identified pairs of significant co-movers.
[Insert Figure 2 around here]
Now consider Figure 3. Similar to Figure 2, Figure 3 plots the average normalized
abnormal turnover of the SMALL company around the four types of corporate events for
the BIG company. Panel A does so for EAs, Panel B for CIGs, Panel C for forecasts, and
Panel D for recommendations.
[Insert Figure 3 around here]
The two figures allow us to compare the relative investor activity visually due to confusion
surrounding different types of events. Out of the four events (EAs, CIGs, forecasts, and
recommendations), trading activity is highest around CIGs issued for the BIG company (the
spike is around 0.4), followed by EAs (roughly 0.12), and then by much smaller values for
forecasts and recommendations. Unlike around ETs, there is a noticeable delayed response
around EAs and CIGs.
To explore these findings further, we calculate the average cumulative normalized abnor-
mal turnover (CNAT) around different types of events. In order to do so, we use a [-10,+10]
window for the 10-minute intervals and compute the CNAT inside the window.
18
Table 3 provides the results. The average CNAT is positive and highly significant around
time intervals with extreme trading activity for the BIG company. The CNAT is highest
for the 99.95th percentile ETs (4.6 for the 31 significant pairs), followed by the 99.9% ETs
(CNAT is 3.9) and 99% ETs (1.9). In every case, the values are statistically different from
those of the control group. We also test to see if abnormal trading activity for the SMALL
firm around ET intervals for the BIG firm is higher for pairs that show stronger confusion
as estimated by equation 1 and find strong support for that idea. For example, a simple
regression of CNAT values from around 99% ETs for the 31 significant pairs on the esti-
mates of the coefficients ˆ
β1from the second step yields the following result (p-values are in
parentheses): CN ATi= 2.18(0.080)+ 33.63(0.000) ˆ
β1i0.01(0.164)# of E T si(N= 31). We
conclude that abnormal trading activity in the SMALL company around ETs in the BIG
company is higher for pairs with stronger confusion.
We then look at the CNAT around time intervals with extreme returns in the BIG
company. Extreme returns are defined as returns that exceed the 99th, 99.9th, and 99.95th
percentile for positive returns and are below the 1st, 0.1st, and 0.05th percentiles for negative
returns. We calculate the CNAT for all extreme returns, and then separately for negative
and positive returns. The bigger the CNAT, the larger the return. The CNAT for the sub-
sample of the 31 pairs is statistically different from zero and that of the control group. We
also tested to see if the reaction differs between positive and negative news and find that
for the most extreme returns (99.9th (0.1st) and 99.95th (0.05th) percentile) the reaction is
greater for positive news (untabulated). This finding is consistent with the hypothesis that
following negative announcements concerning the BIG company, confused investors will try
to sell stocks that they do not own in the smaller company, but short selling rules require
more steps than outright purchases (like locating loanable shares) do, which should limit the
probability of an erroneous trade. Also, Barber and Odean (2008) show that investors are
influenced differently when making buying versus selling investment choices.
The CNAT allows us to formally compare investor trading activity around the four types
19
of corporate events. Among the four types of events, the average CNAT is highest around
CIGs (3.2), followed by EAs (1.5), forecasts (0.5), and recommendations (0.3).9The results
are consistent with Figure 3.
[Insert Table 3 around here]
If erroneous trades are performed by confused investors who mistake one company for
another, then we can predict that such confusion is more likely to occur around unexpected
corporate events involving the BIG company and/or releases of more information. Barber
and Odean (2008) document that individual investors are influenced by high-impact news
when making investment decisions. An example of such news is CIGs. These corporate an-
nouncements are issued infrequently because companies are not required to provide earnings
guidance. Companies usually provide CIGs in the form of expectations for earnings, but
can also provide guidance on other aspects of their financial activities, such as inventory,
units sold, or cash flow. CIGs are events that are associated with information release and
large turnover and return reactions (Skinner (1994), Rogers et al. (2009)). More often than
not, guidance leads to analysts revising their forecasts and changing their recommendations
(Altinkilic et al. (2013), Altinkilic and Hansen (2009)) as well as investors buying or selling
the stocks of the company (Das et al. (2007), Skinner (1994)). Noise traders who hear
such news are more likely to execute a trade involving a matching company by mistake.
On the other hand, EAs are regular, scheduled events. It is less likely that an investor will
read the earnings report for one company and execute a trade accidentally involving another
company.
Analyst forecasts and recommendations are not scheduled events. However, they often
follow other corporate news, such as EAs and guidance. Further, they are much more nu-
merous than either EAs or CIGs (Altinkilic and Hansen (2009)). On average, there are over
100 forecasts issued per firm per year (Altinkilic et al. (2013)). Even though inattentive
9In accordance with our earlier findings, we also document that the 31 pairs are significantly different
from the pairs in the control group.
20
investors may occasionally make trades by mistake following analyst forecasts or recommen-
dations, such activity is less likely to occur on a regular basis due merely to the commonness
of these events.
3.4 Retail Versus Institutional Investors
It can be argued that retail investors are more likely to trade out of confusion. Extant
studies have documented statistics on retail investors’ participation in the U.S. stock market.
According to French (2008), retail investors’ ownership of U.S. stocks was 36.2% in 2000
and had declined to 21.5% by 2007. Retail investors account for an even lower fraction of
total trading volume (Evans (2009) reports 2%). In this section, we investigate whether
confusion trading is predominately a retail or institutional investor phenomenon. In order to
do so, for each time interval, we split the total turnover for the SMALL company into retail
and institutional components. To identify retail turnover for the 2007-2013 era, we follow
Boehmer et al. (2017).10 To identify retail turnover during the 1993-2006 era, we follow the
approach of Barber et al. (2009), as modified by Hvidkjaer (2008). Specifically, for each
month, we sort stocks into quintiles based on the NYSE/AMEX firm-size cut-off points. We
then use the following retail-trade cut-off points within the firm-size quintiles: 3,400 for
the smallest firms, and 4,800, 7,300, 10,300, and 16,400 for the largest firms. Dividing
the cut-off points by the share price at the end of the prior month gives the retail volume
cut-off points. Campbell et al. (2009) document that, between 1993 and 2000 trades under
2,000 are correlated with institutional trading because institutional investors tend to break
up their orders in order to reduce transaction costs. Following Campbell et al. (2009), we
classify trades under 2,000 as institutional trades. Our estimates show that retail volume
was stable at under 10% for every year in our sample.
We then calculate the CNAT for each SMALL company separately for retail and institu-
tional turnovers around extreme turnover (ETs) intervals, extreme returns (ERs), and events
102007 was the first year when TAQ started reporting exchange code ‘D,’ which indicates retail trades in
FINRA Trade Reporting Facility.
21
related to the BIG company. We document the results for retail turnovers in Table 3, column
2. Table 3, column 3, documents results for institutional turnovers. The CNAT values show
that abnormal trading due to confusion is present in both retail and institutional trades
and that the phenomena are surprisingly similar in magnitude. We test for the difference
between retail and institutional confusion and see no obvious pattern. We conclude that
both retail and institutional investors are subject to confusion trading and that the effects
are very similar between the two groups. We investigate the issue of retail trading further
in later sections.11
3.5 Return Co-Movements
In this section, we test to see if co-movements in trading turnover for the 31 identified pairs
translate into co-movements in returns. In order to do so, for each of the 254 significant
pairs that we identified, we estimate the following regression equation:
ReturnSM ALLi,t =β0+β1ReturnB IGi,t +β2ReturnBI G i,t xN eg RetBI G Dummyi,t +
+β2ReturnM arketi,t +β3E ventsSMALLi,t +εi,t (5)
where ReturnBIG i,t is the contemporaneous return for the BIG company, and
Neg RetB IG Dummyi,t is a dummy variable which equals one if ReturnBI G i,t is negative.
Following Rashes (2001), we include this interaction term because we predict that trades
by mistake are more likely to occur around positive news because it is easier to buy than
sell a stock one does not own. EventsSM ALLi,t represent a set of eight dummy variables:
four for the positive and four for the negative corporate events (EAs, CIGs, forecasts, and
recommendations).
We follow the same two-step approach we followed for turnover regressions. The coef-
11The proxies for retail volume for the 1993-2006 period are noisier than the more precise proxies of 2007-
2013. We, therefore, re-run our tests in this section using the 2007-2013 sub-period, and find no difference
between retail and institutional confusion.
22
ficient β1is positive and significant in both steps for 56 out of the 254 pairs, including 18
pairs out of the 31 pairs that showed significant co-movements in returns. Table 1, columns
4 and 5, provides β1estimations for each of the steps for the significant turnover co-movers.
To illustrate the behavior of returns of the SMALL company around events involving the
BIG company, consider Figures 4 and 5 and Table 4. For each company in our pair and
for each time interval, we first calculate the abnormal return as the residual resulting from
running the following regression:
Returni,t =β0+β1ReturnM arket i,t +β2Eventsi,t +εi,t (6)
Figure 4 demonstrates the CARs for the SMALL company around extreme returns (ERs)
for the BIG company. The 0.05th and 99.95th percentiles are used to define extreme returns.
Time zero of each panel corresponds to the time interval coinciding with an ER involving
the BIG company. The figure plots graphs for the sub-sample of 31 significant co-movers
and for the control group.
Panel A of Figure 4 shows the behavior of the CAR in the [-200,+200]-min window
around ETs, which corresponds roughly to one trading day. A modest, virtually trivial CAR
in the periods preceding the ER involving the BIG company is followed by a sizable return
for the SMALL company in the 10-minute interval coinciding with the BIG company’s ER.
The CAR remains constant in the post-period for at least 20 intervals (200 minutes). This
finding indicates that any return reactions involving the SMALL company due to confusion
are present for at least three hours, on average. The CAR for the control group, albeit
visible, is smaller in magnitude.
The estimated average CAR around events in the three windows centered around the
main event: [-2,+2], [-5,+5], and [-10,+10], each involving 10-minute intervals, are presented
in Table 4. The CAR return is highest around the most extreme positive and negative
returns (99.95th percentile). It is around 0.5% (-0.6%) for positive (negative) returns for the
23
31-pair sub-sample. The returns for the 31 pairs are significantly different from those of the
control group. Finally, the return reaction is much smaller for the 99th percentile extreme
returns and intraday stock price jumps.12 It is around 0.11% (-0.17%) for positive (negative)
returns for the 31-pair sub-sample.
[Insert Table 4 around here]
Panel B of Figure 4 shows the behavior of CARs for the SMALL company over a ten-day
period following an extreme return in the BIG stock. We see reversal patterns that take, on
average, around nine days as the SMALL firm’s stock price returns to its fundamental value.
[Insert Figure 4 around here]
Figure 5 plots the CAR for the SMALL company around earnings announcements, CIGs,
forecasts, and recommendations for the BIG company over a ten-day period. At first, it
seems that there are significant reactions regarding the SMALL company to EAs and CIGs
involving the BIG company. However, further analysis in Table 5 shows that these CARs for
the significant pairs when exemining the EAs are not significantly different from zero or from
the values for the control group. This finding explains why Davies et al. (2007) failed to
find any significant reactions around EAs in daily tests. CARs around CIGs are significant
only for positive news.
[Insert Figure 5 around here]
The return reactions to forecasts are significant for the 31 pairs only when the forecasts
are positive (associated with a positive reaction regarding returns for the BIG company).
The reaction to positive recommendations is also significant. The average CARs for both
forecasts and recommendations are between 0.1% and 0.2%.
12The stock price jumps are defined using the algorithm developed by Lee and Mykland (2008). To
use the algorithm, we first estimate the instantaneous bi-power variation using the previous 291 10-minute
intervals (as recommended by the simulation studies done in Lee and Mykland (2008)). We then use the
0.1% significance level to identify time intervals associated with an intraday stock price jump in the BIG
company. The average detection rate is 1.5% for the entire sample and a total of 152,420 stock price jumps
are produced.
24
The observed results and their statistical significance depend largely on the different
sample sizes for different events. Forecasts and recommendations are much more numerous
than EAs and CIGs. The sample of forecasts consists of 15,869 observations, and the sample
of recommendations consists of 3,721 observations. The sample of EAs, on the other hand,
includes only 1,006 events. The sample of CIGs contains just 300 observations. Obtaining
statistical significance for smaller samples is difficult. Overall, we conclude that there is a
significant price reaction involving the SMALL company when there are extreme returns for
the BIG company that may be attributed to investor confusion. The reaction, on average, is
within the (-0.6%; 0.5%) range and lasts for at least 200 minutes. The full reversal, however,
on average takes over a week.
The sizable reaction of the unrelated SMALL stock around ER events involving the BIG
stock indicates markets inefficiencies and the temporary mispricing of the confused stock.
Therefore, the path the reversal takes is predictable, posing an opportunity for a potentially
profitable trading strategy involving the SMALL firm around events dealing with the BIG
firm (and, to a smaller degree, vice versa). There are two possibilities. First, a swift investor
could execute, as soon as possible, an order involving the SMALL stock whenever a noticeable
event pertaining to the BIG stock occurs, be it positive or negative, before confusion sets
in and then reverse the transaction after the confusion drives the SMALL stock price up
or down. The implementation and profitability of this strategy would depend upon acting
quickly and the degree of confusion involved. Alternatively, a smart investor could observe
mispricing for the SMALL stock and execute a reverse order (“sell” for positive news, “buy”
for negative news) after confusion affects the price and realize positive returns when the stock
price returns to its intrinsic value through the reversal path. The profitability of any trading
strategy also depends on transaction costs. We show that any of the above strategies, if
profitable, would earn at most 0.5%. Transaction costs estimated by Lesmond et al. (1999);
Barber and Odean (2000); Hasbrouck (2009) show a much higher round-trip value between
1.2% and 3%, and an even higher value for smaller stocks. An obstacle to implementing a
25
trading strategy based on confusion trading is that such a strategy usually involves trades
of small, illiquid stocks with wide bid-ask spreads. It should be noted, however, that in
recent years, transaction costs have dropped significantly. Moreover, high-frequency trading
algorithms have direct access to exchanges and do not trade through a traditional broker.
Therefore, they operate at fractions of transaction costs (or avoid them altogether, paying
only exchange fees or collecting rebate fees when they supply liquidity) and are able to
exploit any tiny positive alpha. Such algorithms should be able to implement a profitable
strategy based on name/ticker confusion.
3.6 Determinants of Significant Co-Movers
In this section, we investigate what makes the 31 significant pairs different from the rest
of the sample and identify variables that may affect the magnitude of investor confusion.
We start by reporting mean values for the key variables for the entire sample and the 31
significant pairs in Table 5.
An argument can be made that if two companies are vastly different in size, then it is
easier for investors to spot the difference and not execute an erroneous trade. Also, much
smaller companies may be characterized by lower liquidity, and it may be harder for the
investor to execute an erroneous trade if he/she cannot locate a buyer. 13
We start by comparing the size of the BIG and SMALL companies. We see that the
significant pairs are the pairs for which both companies, BIG and SMALL, are significantly
larger in size (measured by market capitalization). This finding is true for both turnover and
return co-movements. We also measure the difference in size between the BIG and SMALL
companies in each pair as the ratio of market capitalizations. The difference in size is not
statistically different between the sub-sample of the 31 pairs and the remaining 223 pairs or
for return regressions.
13Our data are consistent with this observation: in unreported t-tests we find that the mean size of the
SMALL companies is statistically smaller for the sub-sample of the 18 pairs with delayed reaction identified
in section 3.2 than that of the original 31 pairs.
26
One could argue that it is harder for investors to execute a trade in confusion if the two
stocks are listed on different exchanges. To test this, we create a dummy variable for each
pair which equals one if the two firms are listed on the same exchange and zero otherwise.
The mean value of the variable is higher for both turnover and return significant co-movers.
Thus, it is easier to trade out of confusion when both companies are listed on the same
exchange.
We also calculate the mean institutional holdings for each group (Table 5, rows 7-8).
We observe that the sample mean is no different for the BIG companies between the two
sub-samples and is higher for the for the SMALL companies in the 31 significant pairs. This
finding is true for both turnover and return co-movements. We also compare the average
volatility for the BIG and the SMALL companies across sub-samples. The volatilities for the
BIG and SMALL companies are no different for significant turnover co-movers. For return
co-movements, volatility seems to be a much more important factor. For significant pairs in
this case, the volatilities of the BIG and SMALL firms are much smaller than those found
for the rest of the sample. Finally, we compare average analyst followings and find that the
significant pairs, on average, have higher analyst followings for both the BIG and SMALL
companies.
The last two columns of Table 5 provide mean values for the 18 pairs (out of the original
31) for which the return co-movements are also significant versus the return co-movements
for the remaining 13 pairs. Again, we see that the significant pairs are the ones for which both
companies are larger in size. The difference in size, institutional holdings for the SMALL
company, and analyst coverage for the BIG company, and the Same E xchange dummy are
the same for the two sub-samples. Institutional holdings of the BIG company and the analyst
following of the SMALL company are higher for the significant pairs. The volatilities of the
BIG and SMALL companies are not significantly different for the 18 pairs with significant
return co-movements.
[Insert Table 5 around here]
27
We turn further to multivariate tests. To estimate the determinants of significant co-
movers in trading turnover and returns, we first calculate the partial (excluding the effect of
the market) correlation betweenT urnoverSM ALL and T ur noverBIG. We then run a simple
OLS regression of the set of 254 partial correlation coefficients on the characteristics of each
pair and estimate the following equation:
T urnover C orrelationi=γ0+γ1SizeB I Gi +γ2S izeSM ALLi +γ3S ame Exchangei+
+γ4Inst.HoldB IGi +γ5I nst.HoldSMALLi +
+γ6V olatilityB IGi +γ7V olatilitySM ALLi +
+γ8AnalystC ov.BI G i +γ9AnalystCov.SM ALL i +γ10 Overlapi+
+γ11T y pe E ff ectsi+εi(7)
SizeBIG and S izeSM ALL are the natural logarithms of the market capitalization of the
BIG and SMALL company of each pair, respectively. Overlapiis the total duration of the
existence of the confusion for the pair (the length of the time period when the companies
shared parts of their names/ticker symbols), measured in 10-minute intervals. T y peEf f ects
is a set of dummy variables for different types of confusion used to identify the pairs in
our original search algorithm. Table 6, column 1, lists the regression results of estimating
equation 7 for total turnover.
SizeS M ALL is positive and significant, confirming our conjecture that the bigger the
SMALL company in the pair, the more likely it is that investors will make erroneous trades.
SameE xchange dummy is positive and significant. Confusion is more likely when both stocks
are listed on the same exchange. I nst.H oldSM ALLi is negative and significant, consistent
with the idea that companies that are mostly held by retail investors are more likely to
become subjects of confused trading. Volatility for either of the companies in the pair is
not a significant determinant of turnover co-movements. Analyst coverage is positive and
significant for BIG and SMALL companies. Companies with more analysts following become
28
subjects of confused trading more often. Overlap is insignificant, so our results are not
driven by the longer periods of co-existence for significant pairs. For the set of type dummy
variables, we used type 5 as the reference group. In unreported tests, we checked if there is
any difference between the remaining four types of confusion and found none. We conclude
that the type is irrelevant to investor confusion.
We repeat the test separately for retail and institutional turnover in columns 2 and 3,
respectively. We find no major differences between total, retail, and institutional turnover
results, consistent with our findings in section 3.4. We also redo the test for the control
sample in column 4. We find that, except for the marginally significant analyst coverage for
the SMALL firm, none of the variables are significant.
We then turn to identify determinants of significant return co-movers. We estimate an
equation, similar to equation 7, substituting T urnov erC orrelationifor ReturnCorrelationi.
The results are presented in column 5 of Table 6. Neither of the size variables is insignificant.
Again, being listed on the same exchange significantly increases the chances of co-movement
between securities. Institutional holdings are marginally positive for the SMALL company.
V olatilityB IG and V olatilityS M ALL are both negative and significant. It is easier to identify
any co-movements in returns when both companies in the pair experience lower volatility.
The length of the time period is again insignificant. We see that although some determinants
of turnover co-movers are the same as those for the return co-movers (Same Exchange
dummy), others are different (mainly, volatility). This finding helps us explain why some
of the significant co-movers in turnover exhibit return correlations, while others do not.
The pairs that are not significant co-movers in returns are the ones in which the firms are
characterized by volatile returns. Again, we repeat the test for the control group in column 6.
Volatility remains the major explanatory variable when explaining co-movement in returns
along with the size effect.
[Insert Table 6 around here]
Overall, we conclude that even though we were able to identify some factors that influence
29
confusion (exchange and analyst coverage), it is hard to predict which pairs will exhibit strong
co-movement in turnover and/or returns out of candidate pairs with similar names/tickers.
4 Costs of Trades Made by Mistake
In the previous sections, we established that for 31 pairs of stocks, there is evidence of
investors conducting trades out of confusion between two companies. Each trade is associated
with transaction costs. In this section, we estimate how much in transaction costs trades
made by mistake cost investors on average.
We start by identifying the trades that can be attributed to investor confusion. In order
to do so, we use the second step estimates from equation 1 and calculate the proxy for the
volume created by confusion ˆ
V CO ST SM ALLi,t as:
ˆ
V CO ST SM ALLi,t =ˆ
β1T urnoverBI Gi,t ShroutSM ALL,i,t (8)
where ˆ
β1is the step-two estimate of ˆ
β1and ShroutSM ALL,i,t is the number of shares
outstanding for the SMALL firm. This approach allows us to estimate how many, on average,
of the trades can be attributed to investor confusion as a fraction of the total volume involving
the SMALL company. In order to do so, we first find the sum of ˆ
V CO ST SM ALLi,t across all
31 significant pairs and all time periods. The resulting value is 565M shares (Table 7, Panel
A, column 3). We then find the total trading volume for all 31 pairs for all periods. The
total trading volume is 11.6B shares (Table 7, Panel A, column 5). The ratio of these two
values gives us a rough estimate of how much of the trading volume can be due to trades
made by mistake: 565M/11.6B '5%.14
14It can be shown that among the 31 individual pairs, this percentage depends on the size of the BIG
and SMALL companies. The larger the BIG company and the smaller the SMALL company, the larger the
fraction of trades made by mistake. We obtained the following regression equation with just 31 data points
(p-values are in parentheses):
%V COSTSM ALLi = 0.73(0.323) + 0.05(0.087)SizeBI Gi 0.09(0.031)S izeSM ALL i + 0.12(0.534)Same Exchangei
9.90(0.667)V olatilityBI G i 18.98(0.161)V olatilityS M ALL i + 0.00(0.575)Overlapi(N= 31)
30
Evans (2009) documents that trades by individual investors represent, on average, less
than 2% of the trading volume for NYSE-listed firms. Therefore, when compared to the
average amount of retail trading, the documented 5% confusion-driven trading appears to
be driven by more than just retail investors, confirming our findings in Section 3.4.
We then follow Hasbrouck (2009)’s algorithm to identify transaction costs as a percentage
of a closing share price (Table 7, Panel A, column 2). Our estimate, 1.54% is close to that
reported in Barber and Odean (2000). Finally, for each SMALL stock iin the 31 significant
pairs and for each 10-minute time interval t, we calculate transaction costs due to confusion
ˆ
$COSTSM ALLi,t as:
ˆ
$COSTSM ALLi,t =ˆ
V CO ST SM ALLi,t P riceSM ALL i,t T rans. C osts% (9)
We find the total transaction costs due to confusion across all intervals over the period
of co-existence for the 31 pairs. Table 7, Panel A, column 6, provides our estimates. Similar
to the volume estimates, we can find how much trades due to confusion cost investors as a
percentage of the total transaction costs. Our estimates show that total transaction costs
for the sample period for the 31 pairs were 1.37B (Table 7, column 8). At the same time,
the total estimated transaction costs due to confusion were 66M (Table 7, column 6). We
arrive at '4.8% of total transaction costs being due to confusion.
Looking at the average annual numbers (Table 7, Panel A, columns 4 and 7), we see that,
on average, trades by mistakes involve 8.2M shares and 1.1M in transaction costs per pair
per year. These values represent the lower bounds of the true costs associated with confusion
trading for three reasons. First, they do not include any effects of reverse confusion, i.e.,
when events surrounding the SMALL company cause price and turnover reactions in the BIG
company of the pair. Second, they do not include the additional 33 pairs exhibiting reactions
to extreme turnovers only and delayed reactions. More importantly, our estimates do not
include the effect of the price impact. When investors execute erroneous trades and correct
31
themselves afterwards, there may be capital losses involved. According to our findings in
Section 3.5, these capital losses may be up to 0.5% (-0.6%) for positive (negative) news and
could be more sizable than transaction costs. Even though the sum of these two parts is
small compared to the total market, it may have a large impact on the individual investors
who made these confusion trades. We address this issue in Internet Appendix B.
We further calculate total transaction costs around different events involving the BIG
company. In order to do so, we use the 10-minute interval around the event and three
alternative windows: [-2,+2], [-5,+5], and [-10,+10] for the 10-minute intervals. The total
transaction costs for different corporate events for the entire sample period for all 31 pairs
are shown in Table 7, Panel B. The highest total transaction costs occur around times of
extreme trading turnover within the 99th percentile for the BIG company. The [-10,+10]
window (which corresponds to±100 minutes) provides an estimate of 52.99M (or almost
80% of the total transaction costs due to confusion, Table 7, Panel B, column 9)15.
Comparing total costs around different events is not representative due to the differ-
ent frequencies of events. For example, using the 99.9th percentile for identifying extreme
turnover intervals instead of the 99thpercentile decreases the number of ETs ten times over.
Panel C of Table 7 shows the average transaction costs attributed to investor confusion on
aper event basis. Making this transition allows us to compare relative events in terms of
associated transaction costs to investors.
Consistent with our earlier findings in Table 3, the highest transaction costs and num-
ber of wrongful trades occur around time intervals with the most extreme trading turnovers
(99.95th percentile) for the BIG company. On average, for the [-10,+10] window, the trans-
action costs are over 5,080 per event (column 9) and constitute approximately 44,300 traded
shares (column 5). These values are followed by those for the extreme turnovers in the 99.9th
and 99th percentiles with over 4,500 and 2,800 in costs and 39,300 and 24,200 in trading
volumes, respectively. Among the four types of corporate events (Table 7, Panel C, column
15Given that our detection technique generates the 99% EVs with the frequency of one percent of the
time, the [-10,+10] windows around those EVs cover approximately 20% of the data.
32
9), the CIGs have the highest transaction costs per event (over 2,130), followed by EAs (over
1,470), recommendations ( 1,250), and forecasts ( 900). These values are consistent with
uninformed investors reacting more strongly to irregular and unexpected corporate events
(such as CIGs) than to scheduled and expected corporate events (such as EAs).
[Insert Table 7 around here]
5 Stopped Confusion
If our hypothesis is correct and the observed co-movements in turnover and stock prices are
indeed due to investor confusion, then it must be the case that any such co-movement should
disappear when the cause of the confusion disappears. Such a result is possible when one
company of the pair changes its name and/or ticker symbol to a different one. Incidentally,
for 10 out of the 31 pairs, we are able to identify such changes.
For five pairs, one company of the two in the pair changes its ticker (and usually its
name at the same time), and both companies continue to co-exist. This allows us to re-
estimate equation 1 for the new period during which both companies are still publicly traded
concurrently, but the source of confusion has disappeared. For five pairs, two companies
co-existed before the estimation period, and then one of the companies changed its name
and/or ticker, and that change started the possible confusion. In this case, we can re-estimate
equation 1 for the pre-period during which the source of confusion has not started yet. In
four cases, the change of the name and ticker took place in the BIG company. In six cases,
it occurred in the SMALL company.
Table 8 documents the 10 pairs in this study, their old and new tickers and names, as
well as original and new estimation periods with a short description of the origins of the
changes that we were able to identify. Column 8 of Table 9 re-estimates equation 1 for each
of the pairs for the new estimation period. For nine pairs, the newly estimated coefficient is
33
not significant anymore (in one case, it is actually significant but with a negative sign).16
[Insert Table 8 around here]
We believe that this evidence is consistent with our conjecture that any identified co-
movements in trading turnover are due to mistakes made by investors.
6 Discussion and Conclusion
This paper investigates the economic and statistical significance of trades made out of con-
fusion in the U.S. market. Using a sampling approach, we document that confusion trading
is not anecdotal and has a more systematic nature. After conducting a market-wide search
for companies with similar names and/or ticker symbols, we first document that the ma-
jority of publicly traded firms share part of their name and/or ticker symbol with another
company. To narrow our search, we concentrated on the companies for which the likelihood
of confusion is highest, and we were able to identify 254 pairs that can be easily mistaken
for each other by investors.
Out of the 254 pairs, 31 pairs (12.2%) show statistically and economically significant
co-movements in trading turnover within the same 10-minute interval. That co-movement
is hard to attribute to anything but investor confusion. Fifteen more pairs show significant
co-movements only when the trading turnover in the bigger company is especially large.
Finally, for 18 pairs, the co-movement is delayed, with delays being contained within three
trading hours, which brings the total estimate of how common significant confusion pairs are
to over 25% of pairs with similar tickers/names.
Our paper is the first to quantify the percentage of trades associated with confusion and
the effects of these “confusion” trades. Trades made by mistake constitute roughly 6.1M
shares per pair per year and cost investors, on average, 1.1M per pair per year. We find
16Observe that because we do not use the two-step approach here, we tend to over-find confusion, and, by
chance only, are expected to see one significant pair (10x0.1).
34
that such “confusion” trades account for around 5%, on average, of the total trading turnover
and total annual transaction costs involving the smaller company. We also show that investor
confusion is especially great around time intervals with large, abnormal trading turnovers
and/or extreme returns for the bigger company in the pair. Erroneous trades involving the
smaller firm are easier to execute around positive news for the bigger company than around
negative news for the bigger company. Finally, investors who make investment mistakes seem
to be more likely to make trades out of confusion around irregular events (such as company
issued guidance) rather than regular, scheduled events (such as earnings announcements).
We also document that confusion trading is a phenomenon that is attributable to both retail
and institutional investors alike.
We test our hypothesis that these co-movements are due to investor confusion on pairs
for which the cause of confusion ceases due to name and/or ticker changes on the part of
one of the companies and show that any co-movement disappears in such cases.
We do not claim the observed co-movements are confined only to the 31 (or 64) pairs.
Quite to the contrary, the goal of this study is to show that confusion trading is potentially
much more prevalent than previously believed. Our original sub-sample represents a small
fraction of all firms that are potentially exposed to confusion trading. A broader study that
tests for co-movements between all stocks that share parts of tickers or names is left for
future research. Such a study would give a better understanding of how widespread the
phenomenon is, gauge the possibility for a trading strategy that exploits this behavioral
anomaly, and investigate the implications for asset pricing.
We also urge the regulators to study the issue and publicize the list of the tickers which
are most often confused (listed in the Appendix). We urge brokers to implement simple
technical solutions (akin to using a spell checker) to intervene with the order process when a
trader enters a ticker from the list of highly confused pairs. And for investors, our message
is simple, “Always double-check before you trade.”
35
References
Altinkilic, O., Balashov and V. S., Hansen, R. S., 2013. Are analysts’ forecasts informative
to the general public? Management Science, 59, 2550–2565.
Altinkilic, O., Hansen, R.S., 2009. On the information role of stock recommendation revi-
sions. Journal of Accounting and Economics 48, 17–36.
Barber, B.M. and Odean, T., 2000. Trading Is Hazardous to Your Wealth: The Common
Stock Investment Performance of Individual Investors. Journal of Finance, LV (2000),
773–806.
Barber, B.M. and Odean, T., 2001. Boys will be Boys: Gender, Overconfidence, and Common
Stock Investment. The Quarterly Journal of Economics, 116 (1), 261–292.
Barber, B.M. and Odean, T., 2008. All that Glitters: The Effect of Attention on the Buying
Behavior of Individual and Institutional Investors. Review of Financial Studies, 21:
785-818.
Barber, B. M., Odean, T., and Zhu, N., 2009. Do Retail Trades Move Markets? The Review
of Financial Studies, 22(1), 151-186.
Boehmer, E., Jones, C.M., and Zhang, X., 2017. Tracking Retail Investor Activity, working
paper. Available at SSRN: https://ssrn.com/abstract=2822105.
Campbell, J. Y., Ramadorai, T., and Schwartz, A., 2009. Caught on tape: Institutional
trading, stock returns, and earnings announcements. Journal of Financial Economics,
92(1), 66-91.
Chordia, T., Subrahmanyam, A., and Tong, Q., 2014. Have capital market anomalies atten-
uated in the recent era of high liquidity and trading activity? Journal of Accounting
and Economics, 58, 41-58.
36
Isidore, C., 2013. Twitter-hungry investors rush to wrong Tweet shares.
http://money.cnn.com/2013/10/04/investing/twitter-tweet-shares/.
Das, S., Kim, K., and Patro, S., 2007. Management Earnings Forecasts and Subsequent Price
Formation. Working Paper, University of Illinois–Chicago.
Davies, J.R., Hillier, D., and Thamm, J., 2007. Investor confusion and similarly identified
securities. Accounting and Finance, 47(4), 693-711.
Evans, A.D., 2009. A Requiem for the Retail Investor? Virginia Law Review 95(4), 1105-
1129.
Ewing, T., 1999. Mistaken Identity Bolsters Shares of Tiny Company, The Wall Street
Journal. https://www.wsj.com/articles/SB922967682606972316.
Faraway, J.J., 1998. Data splitting strategies for reducing the effect of model selection on
inference. Computer Science Statistics, 30, 332–341.
French, K.R., 2008. The Cost of Active Investing. Journal of Finance 63, 1537-1573.
Jain, P.C. and Joh, Gun-Ho, 1988, The Dependence between Hourly Prices and Trading
Volume. The Journal of Financial and Quantitative Analysis, 23, No. 3, 269-283.
Harris, L., 1986. A transaction data study of weekly and intraday patterns in stock returns.
Journal of Financial Economics, 16, 99-118.
Hasbrouck, J., 2009. Trading costs and returns for US equities: Estimating effective costs
from daily data. Journal of Finance 64 (3), 1445-1477.
Hirshleifer D., Lim, S. S., and Teoh, S. H., 2009. Driven to Distraction: Extraneous Events
and Underreaction to Earnings News. The Journal of Finance, 64: 2289–2325.
Hou, K., Xue, C., and Zhang, L., 2017. Replicating Anomalies. Manuscript, Ohio State
University.
37
Hvidkjaer, S., 2008. Small Trades and the Cross-Section of Stock Returns. Review of Finan-
cial Studies 21(3), 1123-1151.
Lee, S. Mykland, P., 2008. Jumps in financial markets: a new nonparametric test and jump
dynamics. Review of Financial Studies 21, 2535-2563.
Levenson, E., 2014. Investors Buy Up Worthless Stock After Con-
fusing It for an Actual Successful Company, The Atlantic.
https://www.theatlantic.com/business/archive/2014/01/investors-buy-worthless-
stock-after-confusing-it-successful-company/357081/
Lesmond, D.A., Ogden J.P., and Trzcinka, C.A., 1999. A New Estimate of Transaction Costs.
The Review of Financial Studies, 12 (5), 1113-1141.
Levine, M., 2019. Sometimes the Algos Buy the Wrong Stock, Bloomberg.
https://www.bloomberg.com/opinion/articles/2019-02-15/who-won-from-the-ticker-
mix-up-after-j-j-acquisition
McInish, T.H. and Wood, R.A., 1990a. A transaction data analysis of the variability of
common stock returns during 1980-1984. Journal of Banking and Finance, 14, 99-112.
McInish, T.H. and Wood, R.A., 1990b. An analysis of transaction data for the Toronto stock
exchange return patterns and the End-of-day effect. Journal of Banking and Finance,
14, 441-458.
Rashes, M. S., 2001. Massively confused investors making conspicuously ignorant choices
(MCI-MCIC). Journal of Finance 56, 1911–1927.
Rogers, J.L, Skinner, D. J., and A. Van Buskirk, 2009. Earnings guidance and market un-
certainty. Journal of Accounting and Economics, vol. 48, pp. 90–109.
Skinner, D., 1994. Why firms voluntarily disclose bad news. Journal of Accounting Research,
vol. 32, pp. 38–60.
38
Wood, R., McInish, T., and Ord J., 1985. An investigation of transaction data for NYSE
stocks. Journal of Finance, 40, 723-739.
39
Fig. 1 . D istrib ution of t -v alues.
Notes: The figure shows the histograms of t-values for the coefficients for Turnover_Big in
equation 1 for the 254 treatment and 254 control pairs.
0
10
20
30
40
50
60
70
80
-3
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6
6.5
7
7.5
8
8.5
9
9.5
10
>10
Distribution of t-values for the Treatment and Control Groups
254 Pair Treatment Group 254 Pair Control Group
Fig. 2 . A b norm al Turnove r A round ETs.
Notes: The figure shows the average normalized abnormal trading turnover in the smaller company
of each pair around extreme turnover (ET, 99.95 percentile) time intervals in the bigger company
of the pair. EVs coincide with time zero on the horizontal axis.
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
-200 -150 -100 -50 0 50 100 150 200
min
Abnormal Turnover around ETs (99.95 percentile)
31 Pair Treatment Group (N=967 ETs) 254 Pair Control Group (N=4,981 ETs)
Fig. 3 . A bnorm al Turno v er A round Even ts.
Notes: The figure shows the average normalized abnormal trading turnover in the smaller company
of each pair around corporate events (earnings announcements, CIGs, forecasts, and
recommendations) in the bigger company of the pair. Events coincide with time zero on the
horizontal axis.
-0.1
-0.05
0
0.05
0.1
0.15
0.2
-200 -100 0100 200
min
Panel A. Abnormal Turnover Around EAs
31 Pair Treatment Group (N=1,006 EAs)
254 Pair Control Group (N=4,329 EAs)
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
-200 -100 0100 200
min
Panel B. Abnormal Turnover Around CIGs
31 Pair Treatment Group (N=300 CIGs)
254 Pair Control Group (N=707 CIGs)
-0.02
-0.01
0
0.01
0.02
0.03
0.04
-200 -100 0100 200
min
Panel C. Abnormal Turnover Around Forecasts
31 Pair Treatment Group (N=15,869 forecasts)
254 Pair Control Group (N=55,694 forecasts)
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
-200 -100 0100 200
min
Panel D. Abnormal Turnover Around
Recommendations
31 Pair Treatment Group (N=3,721 recommendations)
254 Pair Control Group (N=13,341 recommendations)
Fig. 4 . Cu mu lative A bnorm al R etu rn A round ER s.
Notes: The figure shows the average cumulative abnormal return in the smaller company of each
pair around extreme return (ER, 99.95 percentile) time intervals in the bigger company of the
pair. ERs coincide with time zero on the horizontal axis.
-0.80%
-0.60%
-0.40%
-0.20%
0.00%
0.20%
0.40%
0.60%
0.80%
-200 -150 -100 -50 0 50 100 150 200
min
Panel A. CAR Around Extreme Returns (ERs) (99.95 percentile)
Negative ERs, 31 Pair Treatment Group (N=1,000 ERs) Positive ERs, 31 Pair Treatment Group (N=970 ERs)
Negative ERs, 254 Pair Control Group (N=5,019 ERs) Positive ERs, 254 Control Group (N=5,006 ERs)
-0.80%
-0.60%
-0.40%
-0.20%
0.00%
0.20%
0.40%
0.60%
0.80%
-2 0 2 4 6 8 10
days
Panel B. CAR Around Extreme Returns (ERs) (99.95 percentile)
Negative ERs, 31 Pair Treatment Group (N=1,000 ERs) Positive ERs, 31 Pair Treatment Group (N=970 ERs)
Negative ERs, 254 Pair Control Group (N=5,019 ERs) Positive ERs, 254 Control Group (N=5,006 ERs)
Fig. 5. C u mu lative A bnorm al R eturn A round Ev e n ts.
Notes: The figure shows the average cumulative abnormal return in the smaller company of each pair around extreme return (ER) time
intervals in the bigger company of the pair. ERs coincide with time zero on the horizontal axis.
-0.80%
-0.30%
0.20%
0.70%
-2 0 2 4 6 8 10
days
Panel A. CAR Around Earnings Announcements (EAs)
Negative EAs, 31 Pair Treatment Group (N=504 EAs)
Positive EAs, 31 Pair Treatment Group (N=502 EAs)
Negative EAs, 254 Pair Control Group (N=5,019 EAs)
Positive EAs, 254 Control Group (N=5,006 EAs)
-2.00%
-1.00%
0.00%
1.00%
2.00%
-2 0 2 4 6 8 10
days
Panel B. CAR Around Company Issued Guidance (CIGs)
Negative CIGs, 31 Pair Treatment Group (N=135 CIGs)
Positive CIGs, 31 Pair Treatment Group (N=165 CIGs)
Negative CIGs, 254 Pair Control Group (N=351 CIGs)
Positive CIGs, 254 Control Group (N=339 CIGs)
-0.80%
-0.30%
0.20%
0.70%
-2 0 2 4 6 8 10
days
Panel C. CAR Around Forecasts
Negative Forecasts, 31 Pair Treatment Group (N=7,912 Forecasts)
Positive Forecasts, 31 Pair Treatment Group (N=7,957 Forecasts)
Negative Forecasts, 254 Pair Control Group (N=27,442 Forecasts)
Positive Forecasts, 254 Control Group (N=27,237 Forecasts)
-0.50%
0.00%
0.50%
-2 0 2 4 6 8 10
days
Panel D. CAR Around Recommendations
Negative Recommendations, 31 Pair Treatment Group (N=1,820 Recommendations)
Positive Recommendations, 31 Pair Treatment Group (N=1,901 Recommendations)
Negative Recommendations, 254 Pair Control Group (N=6,392 Recommendations)
Positive Recommendations, 254 Control Group (N=6,571 Recommendations)
Tab le 1.
Turnov e r and R etu rn R egr e ssions for Significant P airs . Notes: The table shows turnover
regressions (columns 2 and 3) and returns regressions (columns 4 and 5) for SMALL firms on the
turnover (returns) of the corresponding BIG firm. The sample is randomly split into two equal-
sized subsamples. Pairs that show significant coefficients in turnover regressions in step one are
shown (column 1) along with the results of the second subsample estimation (step two, column
3). The same procedure is repeated for return regressions as well. The regressions include year,
10-min interval time, and events fixed effects. P-values are in parentheses. ***, (**, *) indicate
significance at 1% (5%, 10%) level. 31 pairs with significant Turnover_Big coefficients in both
steps for turnover regressions are highlighted in grey. All variables are standardized.
TICKER PAIRS
Turn Big St 1
Turn Big St 2
Ret Big St 1
Ret Big St 2
(1)
(2)
(3)
(4)
(5)
1
ACG/ACGY
0.015**
0.024***
0.268***
0.345***
(0.016)
(0.000)
(0.000)
(0.000)
2
ALD/ALDV
0.010*
-0.008
0.002
-0.021
(0.077)
(0.873)
(0.391)
(0.652)
3
ALLC/ALCL
0.012*
0.009
0.017*
0.004
(0.063)
(0.106)
(0.073)
(0.357)
4
AMSC/AMS
0.004*
-0.003
-0.019
0.021***
(0.063)
(0.695)
(1.000)
(0.000)
5
AMSWA/AMS
0.007*
0.009**
-0.003
-0.003
(0.087)
(0.043)
(0.770)
(0.737)
6
AREL/ARLCF
0.021**
0.017*
0.022
0.027
(0.011)
(0.075)
(0.205)
(0.161)
7
BBBY/BED
0.013***
0.018***
0.008***
0.008**
(0.000)
(0.007)
(0.000)
(0.049)
8
BEL/BELT
0.192**
0.017*
-0.089
0.295***
(0.021)
(0.086)
(0.864)
(0.000)
9
CAS/CASL
0.086*
0.019**
0.110**
0.150***
(0.073)
(0.019)
(0.017)
(0.005)
10
CHK/CHKM
0.038***
0.029***
0.031***
0.110***
(0.000)
(0.000)
(0.006)
(0.000)
11
CINF/CIN
0.044***
0.023***
0.031***
0.027***
(0.000)
(0.000)
(0.000)
(0.000)
12
EEM/EEME
0.033**
0.050***
0.494***
0.647***
(0.000)
(0.015)
(0.000)
(0.000)
13
EEM/EEMV
0.081***
0.026***
0.625***
0.665***
(0.000)
(0.026)
(0.000)
(0.000)
14
F/FORD
0.005*
0.006*
0.005*
0.131***
(0.089)
(0.095)
(0.076)
(0.000)
15
FBS/FBSI
0.010**
-0.004
0.013
0.020
(0.042)
(0.657)
(0.236)
(0.149)
16
FCFS/FCF
0.037***
0.039***
0.045**
0.026***
(0.000)
(0.000)
(0.000)
(0.000)
17
FSS/FSSB
0.005*
-0.001
0.004
0.009
(0.056)
(0.544)
(0.321)
(0.140)
18
GLUU/GLU
0.014***
0.004
0.009***
0.001
(0.008)
(0.292)
(0.000)
(0.317)
19
HOMB/HOM
0.854**
0.050*
-0.148
0.202**
(0.020)
(0.092)
(0.820)
(0.044)
20
HPQ/HP
0.037***
0.039***
0.135***
0.025***
(0.000)
(0.000)
(0.000)
(0.000)
21
HWP/HP
0.031***
0.028***
0.004**
0.030***
(0.000)
(0.000)
(0.024)
(0.000)
Tab le 1 (co n tinue d )
TICKER PAIRS
Turn Big St 1
Turn Big St 2
Ret Big St 1
Ret Big St 2
(1)
(2)
(3)
(4)
(5)
22
ISIP/ISIS
0.132***
0.033***
0.003
0.003
(0.000)
(0.000)
(0.175)
(0.169)
23
JPM/JPMX
0.114***
0.134***
-0.096
0.091
(0.000)
(0.000)
(0.890)
(0.108)
24
KSWS/KSW
0.008**
0.036***
-0.007
0.008
(0.019)
(0.000)
(0.703)
(0.274)
25
MCIC/MCI
0.079***
0.151***
0.015*
0.015***
(0.000)
(0.000)
(0.058)
(0.000)
26
MDC/MDCA
0.016***
0.018***
0.636***
0.042***
(0.000)
(0.002)
(0.000)
(0.000)
27
NAFC/NASH
0.025***
0.341***
-0.015
0.122***
(0.000)
(0.000)
(0.814)
(0.000)
28
NTI/NTIC
0.022*
0.007
-0.094
-0.114
(0.061)
(0.316)
(0.999)
(1.000)
29
NWL/GGG
0.037***
0.044***
0.089***
0.015*
(0.000)
(0.000)
(0.002)
(0.097)
30
PSX/PSXP
0.059*
0.068***
0.026
0.309***
(0.089)
(0.004)
(0.288)
(0.000)
31
RCM/RCMT
0.032***
-0.010
0.151*
-0.099
(0.000)
(0.776)
(0.088)
(0.852)
32
REA/REOGF
0.009*
-0.004
0.098***
0.062**
(0.076)
(0.679)
(0.004)
(0.035)
33
SLE/SARA
0.031**
0.031**
0.123*
0.184**
(0.045)
(0.023)
(0.092)
(0.031)
34
TCP/TCPC
0.013**
0.027*
0.019
0.028
(0.029)
(0.098)
(0.248)
(0.130)
35
UHT/UHTS
0.014**
0.014***
-0.022
0.053
(0.045)
(0.005)
(0.702)
(0.214)
36
USB/USBC
0.048***
0.017**
0.090*
0.086*
(0.000)
(0.043)
(0.061)
(0.068)
37
USB/USBE
0.050**
0.035***
0.220***
0.066
(0.014)
(0.005)
(0.000)
(0.102)
38
USLD/USDL
0.067***
0.054***
0.030**
0.018
(0.000)
(0.000)
(0.043)
(0.160)
39
USPH/USHP
0.008**
0.000
0.040**
0.044**
(0.035)
(0.492)
(0.047)
(0.030)
40
VNQ/VNQI
0.053***
0.024**
0.534***
0.460***
(0.000)
(0.042)
(0.000)
(0.000)
41
WIT/WITC
0.141***
0.073***
0.382*
0.411**
(0.000)
(0.001)
(0.065)
(0.016)
Tab le 2.
Extended Regre ssions for Tu rnover . Notes: This table shows additional pairs that gain
statistical significance conditional on extreme turnover for a BIG firm or extending the time
interval around the event. Columns 3-5 show pairs that are statistically significant for three types
of extreme turnover for a BIG firm (see regression 2). Columns 6-8 show pairs that are statistically
significant when the interval around a SMALL firm event is extended and is equal to 60 minutes,
120 minutes, or 180 minutes (see regression 3). The two-step approach is used in each regression.
“+” indicates that the coefficient of interest in each regression is statistically significant at 10% in
both steps.
TICKER PAIRS
Original
Regression
ET 99
Regression
ET 99.9
Regression
ET 99.95
Regression
Delayed
60 min
Regression
Delayed
120 min
Regression
Delayed
180 min
Regression
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
1
ACG/ACGY
+
+
+
+
+
+
2
AMSWA/AMS
+
+
+
+
+
3
AREL/ARLCF
+
+
+
+
+
+
4
BBBY/BED
+
+
+
+
+
+
5
BEL/BELT
+
+
+
+
+
+
6
CAS/CASL
+
+
7
CHK/CHKM
+
+
+
+
+
+
+
8
CINF/CIN
+
+
+
+
+
+
+
9
EEM/EEME
+
+
+
+
10
EEM/EEMV
+
+
+
11
F/FORD
+
+
+
+
+
12
FCFS/FCF
+
+
+
+
+
13
HOMB/HOM
+
14
HPQ/HP
+
+
+
+
+
15
HWP/HP
+
+
+
+
+
16
ISIP/ISIS
+
+
+
+
+
+
+
17
JPM/JPMX
+
+
+
+
+
18
KSWS/KSW
+
+
+
+
+
19
MCIC/MCI
+
+
+
+
+
+
+
20
MDC/MDCA
+
+
+
+
+
21
NAFC/NASH
+
+
+
+
+
22
NWL/GGG
+
+
+
+
+
23
PSX/PSXP
+
+
+
+
24
SLE/SARA
+
+
+
+
+
25
TCP/TCPC
+
+
+
+
26
UHT/UHTS
+
+
27
USB/USBC
+
+
+
+
+
+
28
USB/USBE
+
+
+
+
+
29
USLD/USDL
+
+
+
+
+
+
+
30
VNQ/VNQI
+
+
+
+
+
31
WIT/WITC
+
+
+
+
Tab le 2 (co ntinue d )
TICKER PAIRS
Original
Regression
ET 99
Regression
ET 99.9
Regression
ET 99.95
Regression
Delayed
60 min
Regression
Delayed
120 min
Regression
Delayed
180 min
Regression
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
32
AIPC/AIP
+
+
+
+
+
+
33
HOM/HOMZ
+
+
+
+
+
+
34
PION/PIO
+
+
+
+
+
+
35
RCM/RCMT
+
+
+
36
EWZ/EWZS
+
+
+
+
+
37
HTLD/HTL
+
+
+
+
+
38
IIS/IISLF
+
+
+
+
39
EEM/EEMS
+
+
40
ABI/ABII
+
+
+
41
NEG/NEGX
+
+
42
UICI/UIC
+
+
43
NTI/NTIC
+
44
NET/NETN
+
+
45
PCMS/PCM
+
+
46
GLUU/GLU
+
47
IASI/IAIS
+
+
+
48
UST/USTB
+
+
+
49
RELI/REL
+
+
+
50
EMC/EMCI
+
+
51
BEL/BELM
+
+
52
AURE/AUR
+
53
CCB/CCBL
+
54
VISX/VIS
+
55
ZION/ZN
+
56
ALLC/ALCL
+
+
57
LWIN/LEAP
+
+
58
FCGI/FCG
+
+
59
HMGD/HMG
+
+
60
NEBS/NESB
+
+
61
TODDA/TOD
+
+
62
ICOS/IVIS
+
63
TATT/TAT
+
64
SIE/SIER
+
Tab le 3.
Cum ulativ e A bnorm al Turnove r for SM A LL Firm around Extrem e V olu m e an d C orporate
Events for B IG F irm . Notes: The table shows average cumulative normalized abnormal Turnover (CNAT) for
the SMALL firm around extreme Turnovers (ET), extreme returns (ER) as well as around corporate events for the
BIG firm. The columns show the total, retail and institutional CNAT for the window of [-10, +10] 10-min intervals
centered on the event. The table presents the data for the 31 statistically significant pairs. EV BIG 99, 99.9, and 99.95
corresponds to 99, 99.9, and 99.95 percentiles of largest (extreme) Turnovers for the 10-min intervals in the sample. ER
BIG 99, 99.9, and 99.95, corresponds to largest (extreme) returns corresponding to 99, 99.9, and 99.95 percentiles. -
and “+” correspond to negative and positive extreme news, correspondingly. EA are Earnings Announcements, CIG are
Corporate Issued Guidance, FORECAST is the analyst forecast, and RECOMM is the recommendations issued by
analysts. P-values are in parentheses. There are three categories of significance in the table. Asterisks (***,**,*)
correspond to statistically significant values from zero at 1%, 5%, and 10% correspondingly. Small letters (a, b, and c)
in column 1 correspond to the significant difference between the 31 significant pairs and the 254 pair control group at
1%, 5%, and 10% correspondingly. Small roman numbers (iii, ii, i) in column 2 correspond to the significant difference
between the retail and institutional turnover at 1%, 5%, and 10% correspondingly.
EVENTS
Total Turnover
[-10, +10]
Retail Turnover
[-10, +10]
Inst. Turnover
[-10,+10]
N
(EVENTS)
(1)
(2)
(3)
(4)
EV BIG 99
1.9107***, a
2.2109***, iii
1.8340***
19,020
(0.000)
(0.000)
(0.000)
EV BIG 99.9
3.9274***, a
3.1221***, i
3.8959***
1,917
(0.000)
(0.000)
(0.000)
EV BIG 99.95
4.6193***, a
3.6926***
4.5559***
967
(0.000)
(0.000)
(0.000)
ER BIG 99
0.5988***, a
0.4777***, i
0.5758***
38,272
(0.000)
(0.000)
(0.000)
ER BIG 99(-)
0.5736***, a
0.5063***
0.5576***
19,132
(0.000)
(0.000)
(0.000)
ER BIG 99(+)
0.6239***, a
0.4490***, i
0.5940***
19,140
(0.000)
(0.000)
(0.000)
ER BIG 99.9
1.5722***, a
1.1108***
1.4254***
3,899
(0.000)
(0.000)
(0.000)
ER BIG 99.9(-)
1.3055***, a
1.0425***
1.1925***
1,959
(0.000)
(0.000)
(0.000)
ER BIG 99.9(+)
1.8414***, a
1.1797***, i
1.6607***
1,940
(0.000)
(0.000)
(0.000)
ER BIG 99.95
2.2554***, a
1.5626***
2.0069***
1,970
(0.000)
(0.000)
(0.000)
ER BIG 99.95(-)
1.9773***, a
1.6548***
1.8117***
1,000
(0.000)
(0.000)
(0.000)
ER BIG 99.95(+)
2.5422***, a
1.4676***
2.2080***
970
(0.000)
(0.000)
(0.000)
EA BIG
1.4553***, a
1.7265***
1.3119***
1,006
(0.000)
(0.000)
(0.002)
CIG BIG
3.2296***, a
2.8566**
2.9780***
300
(0.005)
(0.012)
(0.008)
FORECAST BIG
0.4555***, a
0.3682***
0.4367***
15,869
(0.000)
(0.000)
(0.000)
RECOMM BIG
0.3190***, a
0.5386***
0.2793**
3,721
(0.023)
(0.001)
(0.041)
Tab le 4.
Cum ulative A bnor m al R eturn s for S M A L L Firm around Extre m e R eturn s and
Corp or ate Events for B I G Firm . Notes: The table shows average cumulative abnormal returns for
the SMALL firm around extreme returns as well as around corporate events for the BIG firm. The columns
correspond to three windows of [-2, +2], [-5, +5], and [-10, +10] intervals (of 10-min) centered on the event.
The table presents the data for the 31 statistically significant pairs. ER BIG 99, 99.9, and 99.95, corresponds
to largest (extreme) returns corresponding to 99, 99.9, and 99.95 percentiles. ER JUMPS are identified using
the Lee and Mykland (2008) procedure. - and “+” corresponds to negative and positive extreme news
correspondingly. EA are Earnings Announcements, CIG are Corporate Issued Guidance, FORECAST is the
analyst forecast, and RECOMM is the recommendations issued by analysts. P-values are in parentheses.
There are two categories of significance in the table: asterisks (***,**,*) correspond to statistically significant
values different from zero at 1%, 5%, and 10% correspondingly. Small letters (a, b, and c) correspond to
the significant difference between the 31 significant pairs and the 254 pair control group at 1%, 5%, and
10% correspondingly.
EVENTS
RET SMALL
[-2, +2]
RET SMALL
[-5, +5]
RET SMALL
[-10, +10]
N
(EVENTS)
(1)
(2)
(3)
(4)
ER BIG 99(-)
-0.12%***, c
-0.14%***, b
-0.17%***, c
19,132
(0.000)
(0.000)
(0.000)
ER BIG 99(+)
0.13%***, b
0.14%***, c
0.11%***
19,140
(0.000)
(0.000)
(0.002)
ER BIG 99.9(-)
-0.37%***
-0.40%***, b
-0.48%***, a
1,959
(0.000)
(0.000)
(0.000)
ER BIG 99.9(+)
0.36%***, b
0.35%***
0.43%***, b
1,940
(0.000)
(0.004)
(0.004)
ER BIG 99.95(-)
-0.43%***
-0.58%***, a
-0.57%***, b
1,000
(0.000)
(0.000)
(0.000)
ER BIG 99.95(+)
0.61%***, a
0.54%***, c
0.51%**, c
970
(0.000)
(0.005)
(0.014)
ER JUMPS(-)
-0.12%***, c
-0.12%**
-0.12%*
8,247
(0.005)
(0.035)
(0.053)
ER JUMPS(+)
0.16%***, a
0.11%**
0.11%*
8,000
(0.000)
(0.030)
(0.080)
EA BIG(-)
0.18%
0.09%
0.15%
504
(0.490)
(0.739)
(0.583)
EA BIG(+)
-0.01%
-0.01%
0.18%
502
(0.968)
(0.970)
(0.431)
CIG BIG(-)
0.17%
0.35%
0.46%
135
(0.580)
(0.273)
(0.158)
CIG BIG(+)
0.24%
0.40%c
0.61%*, c
165
(0.170)
(0.146)
(0.097)
FORECAST BIG(-)
0.01%b
-0.03%
-0.07%
7,912
(0.697)
(0.385)
(0.121)
FORECAST BIG(+)
0.08%**
0.09%**
0.08%*
7,957
(0.015)
(0.031)
(0.070)
RECOMM BIG(-)
-0.06%
-0.15%*, a
-0.14%c
1,820
(0.419)
(0.056)
(0.118)
RECOMM BIG(+)
0.16%**
0.13%*
0.19%**
1,901
(0.015)
(0.099)
(0.027)
Tab le 5.
Descriptiv e Statistics for the 31 sign ificant pairs. Tu rn over an d Returns. Notes: The table shows the mean values of
descriptive characteristics for the entire sample, compared to significant and insignificant co-movers in turnover and returns. SIZE BIG
(SMALL) is the size of the bigger (smaller) firm in dollars. DIFF SIZE is the LOG difference. SAME EXCHANGE is a dummy variable
that equals one if both companies in the pair are listed on the same exchange. INSTITUTIONAL HOLDINGS BIG (SMALL) is the
mean value of the percentage of the shares outstanding held by institutional investors over the co-existing period. VOLATILITY BIG
(SMALL) is the standard deviation of the 10-min return for the BIG (SMALL) company. ANALYST COVERAGE BIG (SMALL) is
the mean value of the number of following analysts over the co-existing period. OVERLAP is the length of co-existence of the pair in
10-minute intervals. The levels of significant differences in between INSIGNIFICANT and SIGNIFICANT pairs are denoted by asterisks.
***, (**, *) indicate significance at 1% (5%, 10%) level.
TURNOVER (ALL PAIRS)
RETURNS (ALL PAIRS)
RETURNS (31 SIGN. PAIRS)
ENTIRE
SAMPLE
(254 PAIRS)
INSIGNIF.
TURNOVER
(223 PAIRS)
SIGNIF.
TURNOVER
(31 PAIRS)
INSIGNIF.
RETURN
(198 PAIRS)
SIGNIF.
RETURN
(56 PAIRS)
INSIGNIF.
RETURN
(13 PAIRS)
SIGNIF.
RETURN
(18 PAIRS)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
SIZE BIG
$6,871,717
$4,424,993
$24,472,338**
$5,977,779
$10,032,424
$34,975,480
$16,886,736
LOG(SIZE BIG)
13.65
13.42
15.28***
13.31
14.82***
14.88
15.57***
SIZE SMALL
$296,757
$149,942
$1,352,880***
$152,132
$808,112***
$395,563
$2,044,275**
LOG(SIZE SMALL)
10.39
10.18
11.92***
9.99
11.81***
10.75
12.76***
DIFF SIZE
3,483
3,460
3,648
4,364
368
7,405
935
SAME EXCHANGE
0.20
0.18
0.35***
0.15
0.39***
0.38
0.33
INSTITUTIONAL HOLDINGS BIG
0.41
0.40
0.51
0.38
0.51
0.41
0.57***
INSTITUTIONAL HOLDINGS SMALL
0.22
0.21
0.30***
0.19
0.32***
0.14
0.39
VOLATILITY BIG
0.007
0.007
0.006
0.008
0.006***
0.007
0.005
VOLATILITY SMALL
0.013
0.014
0.010
0.015
0.009***
0.010
0.009
ANALYST COVERAGE BIG
38.67
34.62
67.74***
34.72
52.61***
53.15
78.28
ANALYST COVERAGE SMALL
9.90
7.97
23.81***
7.47
18.50***
9.08
34.44*
OVERLAP
39,619
36,569
61,558
34,870
56,412
38,685
78,078
Tab le 6.
Tu r n o v e r C o r r e lat ions for Sam p le an d C on trol g r o u p . Notes: The table shows OLS regressions of partial (of the market) correlation coefficients between BIG and SMALL
firms for 254 pairs. The dependent variable in columns 1(2,3), and 4 is the partial correlation coefficient for total (retail, institutional) turnover. The dependent variable in columns 5
and 6 is the partial correlation coefficient for returns. Columns 1-3 and 5 use the original 254 pairs. Columns 4 and 6 use the control 254 pairs. SIZE BIG (SMALL) is the size of the
bigger (smaller) firm in dollars. SAME EXCHANGE is a dummy variable that equals one if both companies in the pair are listed on the same exchange. INSTITUTIONAL HOLDINGS
BIG (SMALL) is the mean value of the percentage of the shares outstanding held by institutional investors over the co-existing period. VOLATILITY BIG (SMALL) is the standard
deviation of the 10-min return for the BIG (SMALL) company. ANALYST COVERAGE BIG (SMALL) is the mean value of the number of following analysts over the co-existing period.
OVERLAP is the length of co-existence of the pair. TYPE 1-4 corresponds to the pairs identified according to four algorithms. P-values are in parentheses. All parameter values are
multiplied by 1,000 for convenience. ***, (**, *) indicate significance at 1% (5%, 10%) level.
Variables
Turn. Total
Turn. Retail
Turn. Inst.
Turn. Control
Return
Returns Control
(1)
(2)
(3)
(4)
(5)
(6)
Intercept
-1.79
1.08
-2.15
-3.02
-2.00
-13.75***
(0.500)
(0.695)
(0.464)
(0.176)
(0.688)
(0.001)
LOG(SIZE BIG)
0.18
0.05
0.28*
0.21
9.0
1.06***
(0.194)
(0.728)
(0.073)
(0.116)
(0.746)
(0.000)
LOG(SIZE SMALL)
0.23*
0.12
0.22*
0.15
-2.0
0.53**
(0.056)
(0.349)
(0.094)
(0.196)
(0.904)
(0.015)
SAME EXCHANGE
1.22**
1.18*
1.35**
0.24
4.0***
0.18
(0.045)
(0.061)
(0.045)
(0.572)
(0.004)
(0.821)
INSTITUTIONAL HOLDINGS BIG
-0.55
-0.07
-0.22
1.37
2.0
2.15
(0.478)
(0.937)
(0.817)
(0.113)
(0.130)
(0.184)
INSTITUTIONAL HOLDINGS SMALL
-2.62**
-0.81
-1.32
-1.75
4.0*
3.44
(0.014)
(0.504)
(0.309)
(0.170)
(0.065)
(0.150)
VOLATILITY BIG
11.63
26.49
-27.05
-34.91
-123.0**
-48.40
(0.676)
(0.409)
(0.429)
(0.283)
(0.024)
(0.427)
VOLATILITY SMALL
-17.25
-8.05
-20.82
-33.40
-32.0*
-149.56***
(0.197)
(0.566)
(0.158)
(0.197)
(0.092)
(0.002)
ANALYST COVERAGE BIG
0.01**
0.02**
0.001
0.001
-1.0
-0.04***
(0.047)
(0.018)
(0.707)
(0.550)
(0.401)
(0.001)
ANALYST COVERAGE SMALL
0.04***
0.06***
0.04***
0.03**
-2.0
-0.02
(0.002)
(0.000)
(0.006)
(0.029)
(0.516)
(0.438)
OVERLAP
0.00
0.00
0.00
0.00
0.45
0.00
(0.615)
(0.880)
(0.769)
(0.376)
(0.408)
(0.148)
TYPE1
-1.20
-3.34*
-2.60
2.0
(0.448)
(0.068)
(0.182)
(0.515)
TYPE2
-1.76
-4.13**
-3.64*
-1.0
(0.262)
(0.023)
(0.060)
(0.970)
TYPE3
-1.72
-3.33*
-2.97
2
(0.274)
(0.065)
(0.124)
(0.489)
TYPE4
-1.07
-2.78
-2.78
4.5
(0.530)
(0.173)
(0.196)
(0.890)
R-Squared
0.32
0.35
0.30
0.15
0.43
0.20
Tab le 7.
Estim ated Volum e and T r ansaction Costs du e to Confusion Trading for M ajor Volum e an d C o rp or ate E v e nts. Notes:
The table presents the estimated volume and transaction costs in the SMALL firm in each pair caused by erroneous trades due to confusion around EVs and corporate
events. The table presents the data for the 31 significant pairs. Panel A shows the total transaction costs. Panel B shows the total dollar and volume cost for
individual events. Panel C shows the dollar and volume cost for an average pair for each type of event. N is the total number of 10-minutes intervals used in
estimations. N EVENTS is the total number of events of each type used in estimations. Trading volume due to confusion (VCOST) is estimated using equation 9
as the trading volume in the SMALL firm associated with the volume in the BIG firm within the same 10-minute interval as the event. The effective transaction
costs are estimated using the Hasbrouck (2009) approach and equation 10. $COST is the dollar transaction costs for the SMALL firm due to confusion. $COST is
calculated in Panel A within the same 10-minute interval as the event. TOTAL V is the total trading volume for the SMALL company for the sample period. TOTAL
$ are the total transaction costs for the SMALL company for the sample period. VCOST/YR and $COST/YR are corresponding annualized values for trades due to
confusion. In Panels B and C, VCOST and $COST are accumulated over [-2, +2], [-5, +5], and [-10, +10] 10-min intervals around a significant event in the BIG
firm. EV 99, 99.9, and 99.95 correspond to extreme volume events in the BIG firm. EA is an earnings announcement, CIG is company issued guidance, FOR is a
forecast issued by an analyst, and RECOM is an analyst recommendation. The dollar cost values are in thousands of dollars and the volume cost is in millions of
individual shares.
Panel A. Total Transaction Costs
N
Trans Costs%
VCOST
VCOST/YR
TOTAL V
$COST
$COST/YR
TOTAL $
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
TOTAL
1,639,597
-
565
255.56
11,594
$66,482
$34,716
$1,371,034
AVERAGE
52,890
1.54%
18.22
8.24
374
$2,145
$1,119
$44,226
Panel B. Total Transaction Costs for Different Events
EVENTS
N
EVENTS
VCOST
[0,+1]
VCOST
[-2,+2]
VCOST
[-5,+5]
VCOST
[-10,+10]
$COST
[0,+1]
$COST
[-2,+2]
$COST
[-5,+5]
$COST
[-10,+10]
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
EV 99
19,020
61.5
169.2
293.4
460.0
$6,542
$18,999
$33,545
$52,988
EV 99.9
1,917
15.0
34.0
53.0
75.0
$1,540
$3,781
$6,022
$8,622
EV 99.95
967
10.0
21.0
31.0
43.0
$999
$2,291
$3,534
$4,915
EA
1,006
1.5
5.3
9.2
14.0
$186
$592
$1,002
$1,483
CIG
300
0.7
2.3
3.9
5.7
$85
$259
$436
$638
FOR
18,869
10.2
45.5
87.5
147.9
$1,220
$5,312
$10,128
$16,974
RECOM
3,721
3.4
13.4
24.3
39.7
$418
$1,592
$2,877
$4,646
TOTAL
45,800
102.3
290.7
502.3
785.3
$10,991
$32,826
$57,545
$90,266
AVERAGE
6,543
14.6
41.6
71.8
112.2
$1,570
$4,689
$8,221
$12,895
Tab le 7. (continued )
Panel C. Transaction Costs per Event
EVENTS
N
EVENTS
VCOST
[0,+1]
VCOST
[-2,+2]
VCOST
[-5,+5]
VCOST
[-10,+10]
$COST
[0,+1]
$COST
[-2,+2]
$COST
[-5,+5]
$COST
[-10,+10]
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
EV 99
19,020
0.0032
0.0089
0.0154
0.0242
$0.34
$1.00
$1.76
$2.79
EV 99.9
1,917
0.0081
0.0178
0.0276
0.0393
$0.80
$1.97
$3.14
$4.50
EV 99.95
967
0.0106
0.0216
0.0322
0.0443
$1.03
$2.37
$3.65
$5.08
EA
1,006
0.0015
0.0053
0.0091
0.0139
$0.18
$0.59
$1.00
$1.47
CIG
300
0.0024
0.0077
0.0129
0.0191
$0.29
$0.86
$1.46
$2.13
FOR
18,869
0.0005
0.0024
0.0046
0.0078
$0.06
$0.28
$0.54
$0.90
RECOM
3,721
0.0009
0.0036
0.0065
0.0107
0.11
$0.43
$0.77
$1.25
Tab le 8.
Characteris tics of Confusio n Trading by Retail In v esto r s . Notes: Presented are descriptive
statistics from the Barber and Odean (2000) discount brokerage dataset with potential mistaken
trades and reversals. Time to reversals is the time between the SMALL stock is bought/sold on
the event-days in the BIG company and time when the stock is sold/bought. MAR is the daily
market- and commissions-adjusted return. Equity is the total value of the equity account. Age is
the account holder’s age. Gender is a dummy variable, which equals one for males. Married is a
dummy variable, which equals one for married account holders. ***, (**, *) indicate significant
differences in means between the reversals sample and the general population at 1% (5%, 10%).
N
Mean
Min
Median
Max
St.Dev.
(1)
(2)
(3)
(4)
(5)
(6)
Number of mistaken trades
2,150
Number of reversals within 30
days
146
Reversals of BUYs
119
Reversals of SELLs
27
Reversals Sample
Time to reversal
146
11.59
0.00
8.50
30.00
9.06
Commission
146
$102.31
$0.00
$93.42
$401.48
$61.57
Commission (%)
146
1.62%
0.00%
1.26%
5.20%
1.24%
Daily gross MAR total
146
0.87%
-7.77%
-0.06%
20.83%
4.02%
Daily net MAR total
146
0.30%
-9.62%
-0.38%
18.67%
3.82%
Daily net MAR for [0,reversal]
99
0.52%
-6.20%
-0.38%
18.67%
3.73%
Daily net MAR for [+1,reversal]
47
-0.17%
-9.62%
-0.47%
18.67%
3.98%
Equity
52
$74,059
$0.00
$34,690
$334,630
$86,957
Age
52
38.92
18.00
35.00
75.00
16.40
Gender (1=male)
52
0.81***
0.00
1.00
1.00
0.40
Married
52
0.62
0.00
1.00
1.00
0.49
General Population
Equity
36,644
$60,150
-$2,975
$19,925
$10,108,562
$200,875
Age
36,644
41.07
18.00
35.00
75.00
14.91
Gender (1=male)
36,644
0.94
0.00
1.00
1.00
0.24
Married
36,644
0.67
0.00
1.00
1.00
0.47
Tab le 9.
Con fusion pe r sistence and ch an g ed tic k ers. Notes: The table shows the pairs identified by our algorithms (from our significant
31 pairs) for which one company changes its ticker (and often name) just before or after the time period in the sample. The column
“Pairs” shows which ticker in the original pair has changed, followed by the description of the name change (if any). “TYPE” variable
identifies the algorithm used to select the pair. “Time” variable shows whether the ticker change occurred before (PRE) or after (POST)
our original sample interval. Original as well as the new overlap periods are shown in columns four and five. Last column shows
estimates from equation 1 for the 10 pairs with changed tickers/names. P-values are in parentheses. ***, (**, *) indicate significance at
1% (5%, 10%) level.
TICKER PAIRS
TYPE
Time
Old Overlap
New Overlap
Comments
TURN BIG
(1)
(2)
(4)
(5)
(6)
(7)
1
ACG/(ACGY->SOSA)
Atcor Medical/(Acergy->Stolt Offshore)
3
PRE
04/10/2006 -
01/07/2011
03/07/2001 -
04/07/2006
Ticker and Name Change on
04/7/2006 after a sale.
-0.003
(0.942)
2
(AREL->ALTS)/ARLCF
(Alpharel -> Altris Soft)/Arel Comm.
2
POST
12/02/1994 -
10/25/1996
10/28/1996 -
3/11/1998
Ticker Symbol and Name Change
on 10/26/1996.
0.011**
(0.033)
3
(BEL)/(BELT->GBIX)
(Bell Atlantic)/(Bell Technology Group-> Globix)
3
POST
01/25/1996 -
5/29/1998
6/01/1998 -
6/30/2000
Ticker Symbol and Name Change
on 06/01/1998.
0.121
(0.854)
4
CHK/(CHKM->ACMP)
Chesapeake Energy/(Chesapeake Midstream Partners
->Access Midstream)
3
POST
07/29/2010 -
07/23/2012
07/24/2012 -
12/31/2013
Ticker Symbol and Name Change
on 07/24/2012.
-0.012*
(0.077)
5
(FCFS->PAWN)/FCF
(First Cash Fin.->First Cash)/First Commonwealth
3
PRE
01/22/1999 -
12/31/2015
06/10/1991 -
01/21/1999
Ticker Symbol and Name Change
on 01/22/1999.
0.029
(0.310)
6
(JPM->CMB)/JPMX
(JP Morgan Chase->Chase Mnhtn)/JPM Company
3
PRE
01/02/2001 -
05/23/2001
04/01/1996 -
12/29/2000
Merger with J P Morgan & Co
on 12/31/2000.
-0.001
(0.876)
7
MDC/(MDCAF->MDQ)
M D C Corp/M D C Communications
3
PRE
10/02/1998 -
1/29/1999
04/17/1995 -
10/01/1998
Ticker Symbol Change on
10/1/1998.
0.001
(0.893)
8
(SLE->HSH)/SARA
(Sara Lee->Hillshire Brands)/Saratoga Res.
1
POST
07/20/2011 -
06/28/2012
06/29/2012 -
12/31/2013
Ticker Symbol and Name Change
after a spin-off on 06/29/2012.
-0.000
(0.529)
9
UHT/(UHTS->UVE)
Univ. Health/(Univ. Heights->Univ. Insurance Holdings)
3
POST
12/16/1992 -
11/26/1997
04/30/2007 -
12/31/2013
Ticker Symbol and Name Change
after going public on 04/30/2007.
-0.000
(0.994)
10
USB/(USBC->ACAP)
US Bancorp/(USA Biomass->Amcor Capital)
3
PRE
08/31/1998 -
12/07/2000
12/24/1997 -
08/28/1998
Ticker Symbol and Name Change
on 08/31/1998.
-0.106
(0.231)
Appendix A. The listing of 254 candidate pairs with Tickers, Company Names, Size (in thousands of dollars), Industry SIC code, Start and End dates, and TYPE.
TICKER
BIG
TICKER
SMALL
START
END
NAME BIG
NAME SMALL
SIZE BIG
SIZE
SMALL
SIC
BIG
SIC
SMALL
TYPE
APAGF
APCO
1/29/1993
7/20/1999
Apco Argentina Inc Cayman Isl
Automobile Protection Corp Apco
539.36
153.89
13
75
1
BBBY
BED
7/2/1993
5/5/2006
Bed Bath & Beyond Inc
Bedford Property Investors Inc
8052.54
434.84
57
67
1
BYD
BOYD
5/10/1994
9/13/2004
Boyd Gaming Corp
Boyd Bros Transportation Inc
2206.54
24.83
79
42
1
CGRO
CROP
6/23/1994
3/30/1995
Crop Growers Corp
Crop Genetics Intl Corp
81.21
4.71
64
28
1
CYNI
CYAN
5/9/2013
12/31/2013
Cyan Inc
Cyanotech Corp
281.08
29.38
73
28
1
DEBS
DEB
6/16/2006
10/23/2007
Deb Shops Inc
Wisdomtree Trust
390.23
23.07
56
67
1
ECO
ECHO
1/29/1993
1/31/2003
Echo Bay Mines Ltd
Electronic Clearing House Inc
665.77
119.64
10
60
1
EMLTF
EMCO
4/26/1994
1/2/2001
Emco Ltd
Engineering Measurements Co
177.44
29.67
50
36
1
F
FORD
6/15/1995
12/31/2013
Ford Motor Co Del
Forward Industries Inc Ny
54919.69
12.81
37
31
1
GLUU
GLU
3/22/2007
12/31/2013
Glu Mobile Inc
Gabelli Global Util & Income Tr
318.96
68.32
73
67
1
HMGD
HMG
9/21/1993
2/28/1994
Hmg Digital Technologies Corp
H M G Property Investors Inc
44.05
10.94
36
67
1
ILXO
ILEX
2/21/1997
2/10/1998
Ilex Oncology Inc
International Leisure Entpr Inc
984.26
16.83
80
70
1
ISIP
ISIS
1/29/1993
5/9/2000
Isis Pharmaceuticals Inc
Independence Square Income Sec
6913.99
28.60
28
67
1
ISKO
ISCO
10/26/1993
6/23/1999
Isco Inc
Illinois Superconductor Corp
92.40
12.16
38
36
1
KREG
KOLL
10/6/1993
11/23/1994
Koll Real Estate Group Inc
Koll Management Services Inc
112.36
50.45
67
65
1
LRCX
LAM
1/29/1993
11/9/2000
Lam Resh Corp
Latin America Investment Fd Inc
12579.49
84.81
35
67
1
LWIN
LEAP
9/24/1998
2/8/2002
Leap Wireless Intl Inc
Leap Group Inc
18.79
10.69
48
73
1
MEA
MEAD
4/9/1997
1/29/2002
Mead Corp
Meade Instruments Corp
3257.49
5.87
51
38
1
NAFC
NASH
8/1/2013
11/18/2013
Nash Finch Company
Localshares Investment Trust
349.01
10.38
51
67
1
NSDB
NSD
8/5/1993
9/22/2000
Nsd Bancorp Inc
National Standard Co
117.31
5.07
60
33
1
ORX
ORYX
4/6/1994
2/26/1999
Oryx Energy Co
Oryx Technology Corp
1102.18
0.73
13
36
1
PLL
PALL
1/8/2010
12/31/2013
Pall Corp
E T F S Palladium Trust
13581.45
189.59
35
67
1
SLE
SARA
7/20/2011
6/28/2012
Sara Lee Corp
Saratoga Resources Inc
10978.64
5.57
20
13
1
TATT
TAT
12/8/2009
12/31/2013
Tat Technologies Ltd
Transatlantic Petroleum Ltd
65.18
57.00
34
13
1
TRKA
TRAK
12/5/1994
5/27/1999
Trak Auto Corp
Canterbury Park Holding Corp
50.97
47.39
55
79
1
VAPH
VASO
12/10/2003
3/31/2004
Vaso Active Pharmaceuticals Inc
Vasomedical Inc
44.01
10.54
28
38
1
VCO
VINA
8/10/2000
6/5/2003
Vina Concha Y Toro S A
Vina Technologies Inc
45.78
16.79
20
36
1
AREL
ARLCF
12/2/1994
10/25/1996
Alpharel Inc
Arel Comms & Software Ltd
87.40
19.58
35
73
2
ASTA
ASFI
11/14/1995
8/12/1997
A S T Research Inc
Asta Funding Inc
307.94
95.96
35
61
2
AXYS
AXPH
1/12/1998
11/16/2001
Axsys Technologies Inc
Axys Pharmaceuticals Inc
627.87
149.80
36
28
2
BJS
BJRI
8/16/2004
4/28/2010
B J Services Co
Bjs Restaurants Inc
7230.00
1095.62
13
58
2
BNC
BNCM
3/11/1998
1/12/2000
B F C Construction Corp
Bnc Mortgage Inc
69.01
51.19
16
61
2
BRE
BXMNF
8/19/1996
5/2/1997
Bankamerica Rlty Invs
Bre X Minerals Ltd
4715.48
500.00
67
10
2
COHR
CHRI
2/16/1996
3/4/1999
Coherent Inc
Cohr Inc
1574.82
40.61
38
73
2
DAN
DRF
11/21/1997
6/24/1999
Daniel Industries Inc
Dan River Inc Ga
416.45
16.76
35
22
2
DIGI
DGII
1/29/1993
9/4/1998
Digital Switch Corp
Digi International Inc
3170.98
287.90
36
35
2
DURA
DRRA
8/15/1996
11/9/2000
Dura Pharmaceuticals Inc
Dura Automotive Systems Inc
1402.82
6.35
28
37
2
DYNA
DGIX
1/29/1993
3/31/1993
Dynascan Corp
Dyna Group International Inc
20.24
3.28
36
79
2
EON
ELAB
5/23/2002
7/26/2005
E On A G
Eon Labs Inc
3351.13
2754.52
12
28
2
EVER
EVK
5/3/2012
12/31/2013
Everbank Financial Corp
Ever Glory International Grp Inc
1996.78
35.49
60
23
2
FLAG
FTHL
2/11/2000
6/12/2002
First Federal Savings Bank Ga
Flag Telecom Holdings Ltd
433.39
10.06
60
48
2
GLAS
GLAR
5/2/1996
1/12/1998
Glasgal Communications Inc
Glas Aire Industries Group Ltd
118.87
3.11
73
37
2
HMG
HMGC
10/4/1993
10/22/2001
H M G Property Investors Inc
Hmg Worldwide Corp
10.94
1.50
67
39
2
ICOS
IVIS
12/10/1997
11/15/2006
I C O S Corp
Icos Vision Systems Corp
2231.04
436.77
28
35
2
JADE
IGAF
10/5/1999
7/2/2001
L J International Inc
Jade Financial Corp
63.35
24.85
39
60
2
MTC
MTCEF
1/29/1993
9/20/1995
Monsanto Co
Mtc Electronic Techs Co Ltd
32705.33
23.57
28
36
2
NILE
NLTX
5/13/2008
5/11/2011
Blue Nile Inc
Nile Therapeutics Inc
428.22
24.64
59
73
2
ODDS
ODDE
1/29/1993
6/17/1993
Sport Of Kings Inc
Odds N Ends Inc
6.63
0.93
65
53
2
OLS
OLSA
12/19/1995
6/4/1999
Olsten Corp
Ols Asia Holdings Ltd
682.44
0.43
73
15
2
PICO
PPI
11/21/1996
2/26/1999
P I C O Holdings Inc
Pico Products Inc
237.54
1.32
63
50
2
REA
REOGF
11/4/1993
12/12/1997
American Real Estate Invt Corp
Rea Gold Corp
116.75
10.98
67
10
2
ROM
ROMT
10/13/1995
2/26/1999
Rio Algom Mines Ltd
Rom Tech Inc
1061.25
47.98
50
73
2
SCG
ONNN
4/28/2000
8/9/2000
Scana Corp
Scg Holding Corp On Semiconducto
8645.05
3433.12
49
36
2
SHO
SMFC
7/1/1994
12/10/1997
Starrett Housing Corp
Sho Me Financial Corp
79.41
76.45
16
67
2
TEE
TENXF
8/12/1993
5/21/1997
National Golf Properties Inc
Tee Comm Electronics Inc
156.92
12.96
67
36
2
UNIT
UNT
1/29/1993
5/23/2001
Unitrin Inc
Unit Corp
2577.11
615.05
63
13
2
UTEK
UTK
10/25/2000
3/15/2010
Ultratech Stepper Inc
Utek Corp
523.88
48.84
35
67
2
VARI
VARL
4/5/1999
7/6/2000
Varian Inc
Vari L Inc
1506.76
84.84
38
36
2
WIT
WITC
6/4/1999
9/1/1999
Witco Chemical Corp
Wit Capital Group Inc
922.27
710.57
28
62
2
YES
YSCO
1/29/1993
9/19/1996
Yankee Energy System Inc
Yes Clothing Co
471.84
6.16
49
23
2
ZION
ZN
1/3/2007
12/31/2013
Zions Utah Bancorporation
Zion Oil & Gas Inc
5577.25
70.71
60
29
2
AAF
AAFG
12/12/1996
7/16/1998
American Government Inc Port Inc
All American Food Group Inc
116.81
0.82
67
54
3
AAG
AAGIY
1/29/1993
5/21/1999
American Annuity Group Inc
Anglo American Gold Invt Ltd
745.29
48.84
36
10
3
ABI
ABII
7/9/1997
10/9/1997
American Bankers Ins Group Inc
American Business Info Inc
2327.10
614.20
63
73
3
ABI
ABIIA
10/10/1997
7/31/1998
American Bankers Ins Group Inc
American Business Info Inc
2327.10
249.74
63
73
3
ABI
ABIIB
10/10/1997
7/31/1998
American Bankers Ins Group Inc
American Business Info Inc
2327.10
243.58
63
73
3
ABM
ABMI
1/29/1993
10/12/1994
American Building Maintenance In
American Biomed Inc
1596.37
4.70
73
38
3
ACG
ACGY
4/10/2006
1/7/2011
A C M Government Income Fund Inc
Acergy S A
1863.14
626.93
67
13
3
AIP
AIPN
1/29/1993
11/6/2000