Fifteen Minutes of Fame? The Market Impact of Internet Stock Picks*
We examine 120 Nasdaq and over‐the‐counter “buy” recommendations by Internet sites from April 1999 to June 2001. The stock picks show substantial short‐ and long‐run price and liquidity gains, although no new information is revealed about them. We find that stocks with lower initial liquidity have proportionately greater liquidity gains on the pick day. Further, stocks with lower initial liquidity and higher pick‐day liquidity have higher pick‐day excess returns. These results support the idea that stocks have multiple liquidity equilibria and that the stock picks, by coordinating uninformed trading activity, push initially illiquid stocks to a higher liquidity equilibrium.
The low cost of setting up a Web site and the ability to quickly send information to many people, combined with the burgeoning equity culture during the market boom of the late 1990s, gave rise to a new breed of “stock pickers”: the so‐called momentum Web sites. On a prespecified day and time, the sites announced their pick—typically a buy recommendation for a stock. To “sell” the pick, the sites emphasized the stock’s low float, growth potential, and/or lack of visibility and claimed large percentage gains for prior picks. However, no new information was offered about the stocks themselves, other than references to publicly available company press releases. In fact, some recommended firms released statements denying that any material changes had occurred to their financial conditions.1 Before the pick, the sites attempted to coordinate synchronous buying by large numbers of investors. They informed subscribers via e‐mail and exhorted them to log on to the site’s home page around the pick time.2 They also attempted to coordinate with other stock picking sites.
In this article, we examine the market impact of 60 Nasdaq and 60 OTC Bulletin Board (OTCBB) picks by Internet Web sites on the liquidity and valuation of the stocks from April 1999 to June 2001, after which the sites mostly became moribund. We find substantial increases in liquidity and returns on the pick day. If markets are efficient, however, the stock picks should not have any lasting impact since no new information was revealed. Consistent with market efficiency, liquidity and returns partially revert to prepick levels in the month after the pick. Surprisingly, liquidity gains (such as increased turnover) remain higher for the picks 1 year after the event, compared with a sample matched on the book‐to‐market ratio, liquidity, and size in the pre‐event year. Market value and returns are also substantially higher for stock picks for a few months after the event relative to the matched sample.
We contribute to the literature by analyzing the liquidity effects from a no‐news event and the correlation between valuation and liquidity. In comparison, prior research focuses exclusively on the impact of no‐news events on returns.3 Klibanoff, Lamont, and Wizman (1998) find that prices of closed‐end country funds react much stronger to prominent (i.e., front‐page) news than to less salient news. Huberman and Regev (2001) show that prominent news of a cancer‐curing drug, although previously published, had a massive, long‐lasting impact on drug company stocks. Cooper, Dimitrov, and Rau (2001) find dramatic price increases following corporate name changes to Internet‐related dotcom names, independent of the firm’s level of involvement with the Internet. Rashes (2001) documents the comovement of stocks with similar ticker symbols.4
Another contribution of our article is that we propose and test an explanation for the liquidity and return gains following stock picks. Although the sites produce no new information about the stocks, they may increase liquidity by coordinating uninformed trading activity (as suggested by a decline in adverse selection costs following stock picks). Models of liquidity externality argue that such coordination may push stocks to higher liquidity, Pareto‐superior equilibria (Pagano 1989a, 1989b; Dow 2005).5 Consistent with these models, we find that, after controlling for fundamental and microstructure factors, stocks with lower initial liquidity (e.g., higher proportional bid‐ask spread) have proportionately larger liquidity gains (e.g., larger percent decreases in proportional spreads) on the pick day. Additional results support the characterization of multiple liquidity equilibria in Pagano (1989b) and Dow (2005). Finally, we find that stocks with lower initial liquidity and higher pick‐day liquidity have higher pick‐day excess returns, consistent with Amihud and Mendelson (1986), who argued that higher liquidity is associated with lower expected returns. In general, our results support the idea that publicity per se may boost stock returns due to liquidity externalities.
Our sample has some unique advantages over prior studies. It is highly unlikely that our events are signals of future firm value. Also, the stock picks are from a broad cross‐section of industries, and so results are not specific to Internet or technology stocks. Finally, since we know the time of recommendation, we analyze intraday announcement effects and real‐time market efficiency, similar to Busse and Green (2002), who study large stocks recommended by analysts on CNBC TV. Like us, they find a price run‐up prior to the event, a quick price reaction to the event followed by a reversal, and a jump in the number of (mostly buy) trades after the event. However, they do not examine liquidity measures other than the quoted depth, for which they find no statistically significant change. Also, they find no long‐term valuation effects for positive recommendations and some evidence that negative news effects persist for 15 days, whereas we find liquidity and valuation gains persisting for months after the event.
Similar to prior studies, however, our sample is special. The Nasdaq stock picks have a mean market value of less than $8 million and are less liquid than similar firms. They have negative excess returns leading up to the pick date, little media coverage and no analyst following. In related studies, Cooper et al. (2001) study small OTCBB firms, Huberman and Regev (2001) follow biotechnology firms, and Rashes (2001) studies examples of ticker symbol confusion. The sample specialness is, perhaps, unavoidable, since this literature focuses on specific instances of no‐news events.
The rest of the article is organized as follows. In Sections I and II, we describe the data and the empirical methodology, while Section III presents descriptive statistics. Sections IV and V deal with the impact of stock picks at intraday and daily frequencies, respectively. In Sections VI and VII, we explore the determinants of liquidity gains and excess returns on the pick. Section VIII discusses the long‐run performance of the picks, and Section IX concludes. Finally, the appendix contains a summary of results for the OTCBB stock picks.
I. Description of Data
We manually collected 127 stock picks of seven Internet Web sites from April 1999 to June 2001. The Web site names were obtained through an Internet search using the keywords “momentum” and “stock picking.”6 We subscribed to these sites and recorded the stock picks as they were sent to us by e‐mail. To the best of our knowledge, these stocks constitute all picks by the Web sites over this period. The sites did not pick stocks every week, and some sites went in and out of operation. All picks in our sample occurred during regular trading hours. We omit seven picks with confounding information (e.g., an earnings announcement) on the event date. Five Nasdaq and four OTCBB stocks were picked twice; each pick is treated as a separate event. Table 1 shows the distribution of stock picks by market and year. Of a total of 60 Nasdaq and 60 OTCBB picks, 31 Nasdaq and 18 OTCBB picks are in 1999, with only two picks in 2001.
A. Nasdaq Data
We obtain intraday data from the Nastraq database, maintained by the NASD. The data report the price, quantity, and time for each trade, the best prevailing bid/ask quotes, and market makers’ quoted size or depth and bid/ask prices. To purge errors, we delete observations when (1) the trade price or volume is missing or the trade is outside regular trading hours, (2) the price is less than the bid price or it is greater than the ask price, (3) quoted bid/ask prices are zero or negative or the quoted bid‐ask spread is negative.
Trades are matched to the most recent quotes. When the trade execution time is missing, we use the reported time. If the trade execution time is missing or the reported and execution times do not match, we require a lag of at least 2 seconds between the trade and the previous quote.
During the 60 days after the pick date, five stocks were delisted from Nasdaq. For four of these stocks, we use closing quote and volume data from the OTCBB for the postdelisting period. The remaining delisted stock did not trade on the OTCBB, and we set its buy‐and‐hold return (BHR) and bid‐ask spread in the postdelisting period equal to the average BHR and bid‐ask spread, respectively, of the other four delisted stocks for that period.
B. OTCBB Data
Daily OTCBB stock data, obtained from Bloomberg, include open and close prices, closing bid and ask quotes, and daily volume. Outliers in the OTCBB data are cross‐checked with 10K and 10Q filings to remove errors. The data filters used for Nasdaq stocks are also applied to OTCBB stocks. One OTCBB firm was liquidated 4 days after the pick date. Its final price is set equal to the per‐share liquidation proceeds, as it was determined in court.
OTCBB stocks trade over the counter. The OTCBB is a quotation medium for subscribing market makers. Unlike the Nasdaq market, it does not have listing requirements, automated trade execution, or comparable market maker obligations.7 While qualitative results for Nasdaq and OTCBB stock picks are similar, for reasons of brevity we do not report results for OTCBB stock picks but provide a summary in the appendix. Full results are available from the authors.
II. Empirical Methodology
We now describe our performance measures, procedures for selecting matched samples, and tests of significance. Since our focus is on liquidity, we report a number of different measures of it—immediacy, adverse selection, price impact, and depth.
A. Return and Risk Measures
For intraday intervals, we calculate the buy‐and‐hold return (BHR) as the log ratio of the last quote midpoint in the interval to the last quote midpoint in the previous interval. For daily intervals, we calculate the excess return ERET in an interval [a, b] as 
where 
is the closing midquote and 
is the closing Russell 2000 index for day 
. The measure of volatility is STDR, the standard deviation of intraday returns.8
B. Liquidity Measures
We have five measures of liquidity: the proportional quoted half‐spread PQBAS, the proportional Roll covariance estimator PRBAS, the proportional adverse selection cost PROLLAS, the price impact PRIMP, and the total depth TDEPTH.9 The first measure, PQBAS, captures, for a single trade, the cost of immediacy or price concession needed to make an immediate trade at the market maker’s quoted size. The second measure, PRBAS, is due to Roll (1984) and captures temporary price changes due to noninformational (e.g., order‐processing) costs, and so is a measure of “real frictions” (Stoll 2000).10 The third measure, PROLLAS, estimates the “adverse selection bias” in the Roll estimator (which ignores adverse selection costs; see Glosten 1987). Since the bid‐ask spread varies across trade sizes, it may be a misleading measure of trading cost, especially for large trades (O’Hara 1995). Thus, our fourth measure, PRIMP, is the price impact or the sensitivity of prices to trade size, as in Kyle (1985). Finally, market makers supply immediacy by posting bid and ask quotes that are good for a specific size of trade. Thus, TDEPTH, the total depth or quoted size, indicates the market maker’s willingness to supply liquidity at the posted quotes.
We calculate PQBAS as 
where At (Bt) is the inside ask (bid) quote and 
is the quote midpoint for trade t. A reduction in PQBAS indicates a more liquid market.
We estimate PRBAS in interval i as 
where ΔP is the change in the trade price, Mi is the last quote midpoint in the interval, κ is the kurtosis, and n is the number of price changes in the interval. The term in brackets adjusts for the downward bias in a small sample due to Jensen’s inequality. For comparability with half‐spread measures, we do not multiply by 2, as is typical. A reduction in PRBAS indicates a more liquid market.
Following Schultz (2000), the “adverse selection bias” in the Roll estimator in interval i is 
where EBAS is the mean effective half‐spread in interval i, δ is the percent of EBAS due to adverse selection: 
and 
, … , T is the number of trades in the interval. The trade indicator 
(−1) for a (sell) (buy) trade if the price is closer to the bid (ask). 
if the price is equal to the quote midpoint. Buys and sells are determined by comparing prices with the prevailing quotes at the trade time, as in Ellis, Michaely, and O’Hara (2000). We normalize ROLLAS by dividing by the last quote midpoint Mi in the interval i to obtain the proportional adverse selection bias: 
A reduction in PROLLAS indicates a more liquid market.11
To estimate price impact, we start with the price change in an interval i: 
where Mi is the closing quote midpoint in interval i, 
for daily intervals and 
for intraday intervals, and Rd is the daily return on the Russell 2000 index. Note that we use market‐adjusted price changes for daily, but not for intraday, intervals. We use the quote midpoint, instead of the price, to abstract from the bid‐ask bounce. Then, the price impact in interval i is12 
where PVIMBi is the percent volume imbalance in interval i, defined as 
Lower absolute values of PRIMP indicate more liquid markets.
Finally, we report TDEPTH, the sum of the bid and ask depths quoted by market makers at the inside bid and ask prices. Higher values of TDEPTH indicate more liquid markets.
C. Matching Procedure and Statistical Tests of Significance
Let v be the number of matching variables, xj the data for stock pick x and the jth matching variable, and yj the data for a candidate firm y and the jth matching variable, where j=1 to v. The Euclidean distance between the two firms x and y is: 
We select a matched firm y to minimize 
. Since variables with large variance tend to have more effect on 
than those with small variance, they are standardized before computation.
The matching occurs in two steps. First, we consider firms recommended in 1999. For Nasdaq stock picks, we use CRSP data to find all firms trading on Nasdaq in 1998 that were not in our sample of recommended stocks.13 We select matched firms, equal in number to those recommended in 1999, using values of the matched variables from Compustat for fiscal year‐end 1998. In the second step, we select firms matching those recommended in 2000 and 2001, using Compustat values for fiscal year‐end 1999. A matched firm that is identical to a firm selected in the first step is deleted, and the next closest unique match is selected.
To show statistical significance, we test whether the cross‐sectional mean or median is different from a control mean or median. These tests are used in tables 6 and 7 for comparing nonreturn statistics in the event and post‐event period against their values in the pre‐event period (the control period). For the means test, we use Dunnett’s (1955) many‐to‐one t‐statistic for comparing multiple test‐means to the same control mean.14 For the median test, we use a nonparametric test for location differences based on median scores that equal one for observations above the median and zero otherwise. A standardized Z‐statistic is then computed based on the actual, expected, and variance of the sum of scores. Since the number of stock picks is relatively small, we estimate by Monte Carlo simulation the exact (instead of asymptotic) p‐values for testing the null of no difference in medians.
We also test whether mean and median returns are different from zero. For mean returns, we use a Student’s t‐test. For median returns, we use a sign test based on the difference between the number of values above and below zero. Values equal to zero are omitted.
III. Descriptive Statistics
In this section, we present descriptive statistics in order to provide a better understanding of the Web sites, the types of investors who read them, and the characteristics of stock picks.
A. Characteristics of Web Sites
It is possible, though generally unlikely, that some Web sites may provide superior information about future firm value. For example, Metrick (1999) finds that, while the average investment newsletter provides no value, some have superior records. We provide indicators of a Web site’s longevity, visibility, and the track record of its picks. To benchmark a stock’s record, we calculate mean excess returns for daily intervals [−3, −1] and [0, 2], where 
is the pick day, and then subtract the mean excess return in the [−100, −6] interval. We also calculate the means of PEBAS and DVOL, the daily volume, for intervals [−3, −1] and [0, 2] and divide by the corresponding mean in the [−100, −6] interval.
Table 2 lists the name of each Web site and its characteristics. We find that only one site, Greatpicks.com, is still functioning.15 Possibly, the sites were shut down as small investors lost interest in stocks following the market crash. A Factiva and Lexis‐Nexis search reveals that the sites had few media mentions in the month before the picks. The business of a typical pick (as described in Bloomberg) is not concentrated in any industry. Relative to the benchmark, the mean DVOL is higher, while PEBAS is lower, during the 3‐day windows before and from the pick day, and these differences are mostly statistically significant. Mean excess returns are also higher than the benchmark, which may be due to prepick buying by Web site operators or information leakage, rather than superior forecasting ability of the Web sites. Overall, no site clearly stands out as providing superior value to its subscribers.
B. Characteristics of Web Site Subscribers
Who subscribes to the Web sites? If the sites are not informative, why do they trade the picks? The initial buyers could be responders to the coordination efforts of Web sites or investors with high search costs attracted by the price and volume run‐ups in the days preceding the pick day (Barber and Odean 2002). Later buying may be due to cascades,16 or the presence of feedback traders (DeLong et al. 1990). Trading itself justifies more trading, creating a positive externality and sustaining liquidity for a long period.
Small investors are likely to have higher search costs, since they cannot spread such costs over large trading volumes. Table 3 shows that the daily trade size, averaged over all trades on the pick day, is less than 1,000 shares for Nasdaq stock picks. The average daily dollar volume on the pick day is less than $2,000 for Nasdaq picks and less than $50 for OTCBB picks. Thus, preliminary evidence suggests that small investors constitute the vast majority on pick day.
To gain further insight into trader behavior, we examine the pick‐day trading pattern in the Nasdaq stock ABIX, picked at 9:45 a.m. on February 16, 2000. It has average trade size on pick day of 792, close to the median for all stocks. We show, in table 4, all large trades (i.e., those 3,000 shares or greater) in ABIX on the pick day. Out of 681 trades in the day, only 25 are large buy or sell trades. The average dollar trade size is less than $3,000. These results show that small investors dominate trading in ABIX on pick day.
What is the holding period of small investors? Our data do not allow a precise answer to this question, but the buy‐sell trends for ABIX in table 4 may provide some insights. These trends show alternating waves of buys and sells throughout the day. For the first 4 1/2 minutes after pick time, the buys prevail as the buy volume is about 62,000 shares, compared with just 11,000 shares of sell volume. Sell volume modestly exceeds buy volume in the next 2 minutes, but this trend is reversed in the next 13 minutes. Overall, buy volume exceeds sell volume by 74,000 shares in the first 19 1/2 minutes after pick time. Over the next 3 1/2 hours, however, the opposite is true as the sell volume exceeds buy volume by 34,000 shares. There is a final spurt of buying over the next 2 hours, followed by a deluge of selling in the last half‐hour. Over the day, there is excess buy volume of more than 35,000 shares. The evidence suggests that, while some initial buyers may have held on to their stocks, others sold out to “follow‐on” waves of small investors, possibly attracted by the initial burst of trading.
C. Characteristics of Stock Picks
Web site subscribers may be interested in certain industries. For example, stocks followed on message boards are mainly from technology and health care sectors (Wysocki 1999).17 We obtain industry descriptions from Standard Industrial Classification (SIC) codes and Bloomberg, since SIC codes may be less descriptive for new industries. SIC codes are available for 117 stock picks. We define each industry broadly to include firms involved in equipment manufacturing, services, or trade. The results (not reported) show that, at most, 22% of all picks are technology‐related companies. The remaining picks are from a broad range of industries, such as financial services and manufacturing. Using the Bloomberg definition, we devise a broader definition of technology firms. For example, if a firm’s primary business is to provide healthcare services through the Internet, we classify it as an Internet firm even if it is a healthcare firm according to the SIC code. The result is similar when using the alternative industry definition: just 25% of picks are technology firms.
Next, we report descriptive statistics for the stock picks and a matched sample. There are 59 recommended Nasdaq firms, since one firm was recommended twice in the same fiscal year, and they are matched to 59 Nasdaq firms based on market value (MV) and book‐to‐market value (BMV), as described in Section II.C. We do not match on liquidity since we hope to assess whether the sample firms have lower liquidity than the matched firms, as claimed by the Web sites. Later, when studying long‐run performance, we match on MV, BMV, and liquidity.
Table 5 reports statistics for Nasdaq stock picks and the matched sample. The Nasdaq picks have higher median proportional quoted half‐spread (PQBAS), and the difference is statistically significant, indicating lower liquidity relative to the matched firms. They also have lower mean MV, higher mean revenue‐to‐market value (RMV) and median earnings per share (EPS), but lower mean net income‐to‐market value (NMV). Compared with the matched firms, the picks also have lower shares outstanding (SOUT) and mean turnover (TOVER), although these differences are not statistically significant. Overall, the Nasdaq picks have lower liquidity and size compared with the matched firms but superior revenues.
Table 5 also reports NEWS, the total number of news items about the firm in the Bloomberg news archive for the 6 months prior to the pick month. The mean NEWS is 7 for the picks and about 8 for the matched firms, but this difference is not statistically significant. The range of NEWS is zero to 36 for the picks, suggesting that visibility varies substantially across stocks. By comparison, Gadarowksi (2001) finds that, for NYSE/AMEX nonfinancial firms, the average annual number of news stories is 22, with a range of zero to 524.
IV. The Intraday Market Impact of Stocks on Pick Day
What is the immediate market impact of stock picks? To what extent are the Web sites successful in coordinating large‐scale buying? To answer these questions, we document changes in the valuation and liquidity of stock picks on the recommendation day. We report both mean and median statistics to assess whether the results are due to a few outlier stocks.
Figure 1 plots mean values of the cumulated buy‐and‐hold return CUMRET, PQBAS (multiplied by 2), and VOLUME (divided by 1,000) per minute on the pick day. The first interval is [open, −3): the interval from market open to 3 minutes before the pick time. All other intervals are of 3 minutes duration. CUMRET is around 40% from market open to the event, decreases thereafter, and ends the day at about 25%. VOLUME per minute is less than 1 initially, exceeds 5 in the event interval, and drops sharply thereafter but still remains at a relatively high level at the end of the day. PQBAS is around 11% initially, falls to around 8% in the event interval, continues to decrease thereafter, and is around 5% at the end of the day.
Fig. 1.— Intraday market impact of Nasdaq stocks on pick day from market open to market close
Table 6 reports statistics for eight intraday intervals, with end‐points that indicate time in minutes from pick time (time 0). For example, the second and third intervals are [−3, 0]—the 3 minutes before the pick time—and [0, 3]—from pick time to 3 minutes after the pick time. In panel A of table 6, we report measures of activity, returns, and risk. The buy‐and‐hold return BHR is positive and statistically significant from the market open to 3 minutes after the pick time, statistically indistinguishable from zero from then on to 60 minutes before close, and negative and statistically significant in the [60, close] interval. The surge in prices prior to the pick time may be due to news leakage, buying by owners of the Web sites, errors in the pick time reported by the Web sites, or slow uploads by the Web sites.18 The standard deviation of returns per minute is STDR divided by the square root of the interval length. Its mean value is higher, relative to the market open, from 3 minutes before to 15 minutes after the event (the median is higher for 60 minutes), and this increase is statistically significant. The mean number of trades NTRADE per minute increases from less than 1 initially to 6 in the [−3, 0) interval and to 22 in the event‐interval. The mean increase in VOLUME and NTRADE is statistically significant up to 30 minutes after the event, while the median increase is statistically significant until market close. The mean percent volume imbalance PVIMB is between −0.35 and −0.07 from market open to 30 minutes after the event, showing large buy imbalances at this time.
Panel B of table 6 reports the intraday liquidity measures. PQBAS decline from the event interval to market close when they are about 2% lower than opening levels. While the proportional Roll covariance PRBAS fluctuates, the mean is about 1.5% lower at market close compared to opening levels. Consistent with increased activity by uninformed traders, the adverse selection measure, PROLLAS, declines monotonically throughout the day. The median price impact PRIMP is higher from the event interval to market close, and this increase is statistically significant. The median total depth TDEPTH is higher from 3 minutes before to 60 minutes after the pick, and this increase is also statistically significant.19
To summarize, prices and volume increase sharply on the pick day. There is a temporary increase in the buy imbalance and return volatility around pick time that is reversed by market close. In spite of increased activity and volatility, the bid‐ask spread is lower. More market makers step up to supply liquidity by increasing the quoted depth, particularly on the bid side.
V. The Market Impact of Stock Picks Before and After the Pick Day
In the previous section, we established that Nasdaq stocks experienced substantial gains in liquidity and returns on pick day. Is this a short‐lived phenomenon? Chan (2003), for example, shows those stocks with large price movements but no identifiable news show reversal in the next month. Also, how different are the prepick and postpick performances? To study these questions, we examine stock performance from 100 days before to 60 days after the pick day. We report both mean and median statistics to demonstrate robustness of the results.
Figure 2 plots the cumulated mean excess returns of Nasdaq stock picks from 100 days before to the prepick day (CUMRET1) and from the pick day to 60 days after the pick day (CUMRET2), mean VOLUME (divided by 20,000), and mean PQBAS (against the right y‐axis). Prices decrease till shortly before the pick day and so CUMRET1 is negative. CUMRET2 is almost 25% on the pick day, falls slightly below zero 10 days after the pick day, and fluctuates around zero from then on. VOLUME spikes on the event day, falls right after but remains higher than initial levels. PQBAS varies between 5% and 6% initially, drops to 3.3% on the pick day, and rises thereafter but remains well below 5% for the entire postpick period.
Fig. 2.— Market impact of Nasdaq stocks from 100 days before to 60 days after pick day
In table 7, we report statistics for six intervals. The end‐points of an interval are days relative to the event date, or day 0. For example, [−5, −1] is the period from 5 days to the day before the pick date while [0, 0] is the pick day. There are two samples. Sample 1 consists of all 60 Nasdaq stock picks. To compare the performance of pre‐ and post‐1999 stock picks, we create a second sample consisting of 29 stocks that were picked in the year 2000 or later. The market impact of later (i.e., post‐1999) picks may be different from that of earlier picks. For one, later picks may be of better or worse companies. Even if they are similar companies, investor reaction could be different. For example, if small investors have short horizons, they may conclude that short‐term return reversal of picks indicates poor long‐run performance. Finally, we examine whether postpick liquidity gains occur primarily on days with positive firm news.
A. Activity and Liquidity Before and After the Pick Day: All Nasdaq Stock Picks
Panel A of table 7 reports results for all 60 Nasdaq stocks. In the [−100, −6] interval, excess returns are about −30% but turn positive in the 5 days before the pick date. Cooper et al. (2001) also report a pre‐event rise in prices for their sample. Median excess returns are 24% on the pick day and −27% in the following 10 days. After 10 days, excess returns are not statistically different from zero. The median NTRADE is 5 during the [−100, −6] interval, jumps to 383 on the pick day, and drops to 10 after 20 days, still higher than initial levels, and this increase is statistically significant. The median STDR is lower after 60 days, and this decrease is statistically significant. PVIMB is initially positive, turns negative on the pick day, implying excess buy imbalance, and becomes positive again after the pick day.
Turning to liquidity, the various bid‐ask half‐spread measures remain below initial levels by about 0.4% to 0.7% even after 60 days, a statistically significant decline. PQBAS declines sharply on the pick day and then increases, but its level after 60 days remains below prepick levels. The median PRBAS increases on the pick day relative to initial levels, indicating increased order handling costs, but it falls thereafter. PROLLAS declines from initial levels, consistent with lower adverse selection costs due to increased uninformed trading. The median PRIMP is higher for 10 days after the event but returns to initial levels afterward. The median TDEPTH is about 300 shares more than initial levels 20 days after the pick day.20 Overall, the results show that the return gains from stock picks last about 10 days whereas liquidity gains and volatility reductions persist at least 60 days.
B. Additional Investigations
To compare the market impact of early and late picks, we report results in panel B of table 7 for 29 stocks picked after 1999 (sample 2). We find that pick day returns and liquidity gains are similar for pre‐ and post‐1999 picks. For example, pick‐day median returns are 23.15% versus 23.57% for all stocks, and pick ‐day median PQBAS decreases by 1.46% from initial levels compared to 1.53% for all stocks. However, longer‐term liquidity gains are lower for post‐1999 picks. For example, the median PQBAS in the [21, 60] interval is similar to initial levels for post‐1999 picks, whereas it is lower than initial levels by 0.78% for all stocks. While PRBAS and PROLLAS decrease for post‐1999 picks, the improvement relative to initial levels is about half that for all stocks. Thus, for post‐1999 stock picks, pick‐day return and liquidity gains remain substantial, but liquidity gains are lower after 60 days compared with earlier stock picks.
The improved liquidity in the postpick period may be due to positive corporate news. After searching the Bloomberg news archive for company news, there were 170 firm days with news out of a possible 3,600 firm days (i.e., 60 days for each of 60 firms). We find that, on these 170 firm days, the correlation between ERET and PQBAS is −0.09 but not statistically different from zero.21 After conditioning on 85 firm days with positive ERET, the correlation doubles to −0.19 and is statistically different from zero at the 10% level or less.22 These results indicate that days with news (especially positive news) are associated with improved liquidity. However, the correlation is moderate, and the number of news days is about 0.5% of the postpick sample, so it appears unlikely that the postpick liquidity gains are primarily related to positive news events.
To check for outliers, we omit six firms with negative pick‐day excess returns and six firms with the highest pick‐day excess returns. We also remove outliers on the basis of trading activity and liquidity variables. In all cases, the results are similar to those for the whole sample.
Results are similar for OTCBB stock picks (see the appendix). In particular, liquidity gains persist for 60 days, they remain statistically significant for post‐1999 picks, and they cannot be attributed to good news in the postpick period.
VI. Determinants of Liquidity Gains on the Pick Date
In this section, we seek to explain the pick‐day improvement in liquidity documented earlier. To this end, we estimate a cross‐sectional regression where the dependent variable is 
In words, CPQBAS is the percent change in a stock’s PQBAS on the pick date (day 0) relative to the average PQBAS in the 5‐day period prior to the pick date. Below, we relate each set of explanatory variables to our two main hypotheses: liquidity externality and visibility.
l. Liquidity externality. Due to a coordination problem, stocks may have multiple liquidity equilibria. Stock picks, by increasing uninformed trading, may push stocks from low to high liquidity equilibria. We hypothesize that stocks with low initial liquidity should have proportionately higher increases in liquidity on the pick day. For example, an increase of 1,000 shares on pick day is likely to be more beneficial for a stock if its initial trading volume is 10,000 shares rather than 100,000 shares. The measure of initial liquidity is PQBAS[−100, −6], the average PQBAS from 100 days to 6 days before the pick date. A negative coefficient indicates that stocks with higher initial bid‐ask spreads enjoy larger proportional decreases in spreads (i.e., increases in liquidity) on the pick date. In the next section, we experiment with alternative measures of liquidity.
2. Visibility. Web site owners claimed to have picked stocks based on poor prior visibility. Also, prior research finds a correlation between liquidity and visibility.23 Our visibility proxy is log(NEWS): the natural log of the number of news items in the 6 months before the event month. Since NEWS is zero for one stock, we add 10−4 to ensure that the log is defined. A positive coefficient implies that stocks with lower initial media exposure have higher pick‐day liquidity gains.
Liquidity gains may also be due to microstructure factors, such as increased trading volume on the pick day, or past price and volume changes, or better fundamentals. Accordingly, we use additional variables to control for these effects, as discussed below.
3. Microstructure factors. Since microstructure theory argues that liquidity decreases with risk and increases with trading activity, we conjecture that changes in liquidity are related to changes in volatility and volume, measured over the same period as the change in liquidity: 
and 
4. Past returns. Since prices are increasing in the 5 days leading up to the pick day (table 7), the event‐day excess returns could be due to positive returns in the recent past that attract buying interest by investors with large search costs.24 We use ERET[−5, −1]: the average excess return from 5 days before to the day before the pick date.
5. Fundamental factors. Prior research shows that fundamental‐to‐price ratios are associated with future returns. If pick‐day returns are correlated with liquidity (as we will show later), then the ratios may also determine pick‐day liquidity changes. Following Fama and French (1992), we use these variables, measured as of the fiscal year‐end prior to the event‐year:25 log(MV): the natural log of the stock’s market value; BMV: the book‐to‐market value; and EPS: the earnings to price ratio.
A. Results
Panel A of table 8 reports results when the dependent variable is CPQBAS. The t‐statistics are corrected for heteroskedasticity using the Newey‐West (1987) procedure. The first specification tests for the main hypotheses of liquidity externality and visibility without including any control variables. The coefficient on PQBAS[−100, −6] is negative and statistically significant, indicating that stocks with higher initial PQBAS have larger percent decreases in PQBAS on the pick date, consistent with the hypothesis of liquidity externality. The coefficient on log(NEWS) is also statistically significant and negative, indicating that stocks with higher media exposure initially have larger liquidity gains, perhaps because greater prior coverage provides more credibility to the recommendation. Once we control for CVOLATILITY and CVOLUME, the statistical significance of PQBAS[−100, −6] and log(NEWS) increases, and the adjusted 
jumps from 13% to 45%. An increase in volatility increases PQBAS while an increase in volume decreases it, and both changes are highly statistically significant. Past returns ERET[−5, −1] are not statistically significant, and its addition does not improve the adjusted 
. Of the fundamental factors, only log(MV) is statistically significant and positive, indicating that smaller stocks have greater percent reductions in PQBAS on the pick day.
B. Robustness Checks
We repeat the analysis with alternative measures of liquidity. When we replace PQBAS with PEBAS, the proportional effective half‐spread, the results (not reported) are similar. In particular, the hypotheses of liquidity externality and visibility are strongly supported.
In Dow (2005), illiquidity derives from asymmetric information and there are multiple equilibria with different levels of the bid‐ask spread. The quoted and effective bid‐ask spreads used above include the impact of adverse selection. However, we further isolate it by using CPROLLAS as the dependent variable. CPROLLAS is defined as the percent change in proportional adverse selection costs (PROLLAS) from 5 days before to the event day. Then, the initial liquidity variable is PROLLAS[−100, −6], the average PROLLAS from 100 days to 6 days before the pick date.
The results, which are in column 1 of table 8, panel B, show that the coefficient on PROLLAS[−100, −6] is −12.5, indicating that an increase of 1% in the initial adverse selection cost is associated with a 12.5% reduction in PROLLAS on the event day. This result strongly supports the hypothesis of liquidity externality and, further, is consistent with Dow (2005).
In Pagano (1989b), low‐liquidity equilibria are characterized by few traders and high price volatility. The proxy for the number of traders is the number of trades NTRADE. Thus, our dependent variables are log(CNTRADE) and CVOLATILITY, the (log of the) percent change in the number of trades and the percent change in volatility, respectively, from the [−5, −1] interval to the event day. In Pagano (1989b), multiple equilibria exist only if transaction costs (defined as lump sum participation fees) are high enough. Using PQBAS as a measure of transaction costs, we define the initial liquidity variables as 
and 
.
Thus, a low (high) value of the former (latter) is associated with a high‐transactions‐cost‐low‐liquidity equilibrium. When CVOLATILITY is the dependent variable, we only use stocks where 
(i.e., when volatility declines on the event day).
The evidence reported in columns 2 and 3 of table 8, panel B, supports Pagano (1989b). All initial liquidity variables have negative and statistically significant coefficients, indicating that stock picks with lower trading frequency and higher transaction costs initially have larger percent increases in trading frequency on the pick day and those with higher volatility and higher transaction costs initially have greater percent reductions in volatility on the event day.
Finally, we use CDEPTH, the percent change in the quoted depth from the [−5, −1] interval to the event day, as the dependent variable and DEPTH[100, −6] as the initial liquidity variable. The results, shown in the last column of table 8, panel B, indicate that the coefficient of initial depth is negative but not statistically significant. However, the coefficient of log(NEWS) is negative and statistically significant, as it is in two other cases (cols. 1 and 3 of table 8, panel B). Thus, stock picks with greater prior visibility have larger liquidity gains on pick day.
C. Summary of Results: Explaining Pick‐Day Liquidity Gains
Results for OTCBB stock picks (discussed in the appendix) show that stocks with higher volatility and higher transaction costs have greater percent reductions in event‐day volatility. Therefore, initially illiquid Nasdaq and OTCBB stocks have proportionately greater liquidity gains on event day, consistent with network effects in liquidity. We also find specific support for the liquidity externality models of Pagano (1989b) and Dow (2005). Finally, for Nasdaq stocks, greater prior media exposure is associated with larger pick‐day liquidity gains.
VII. Determinants of Excess Returns on the Pick Day
In this section, we examine the causes of return gains on the pick day. Our key hypotheses relate to liquidity, visibility, and past returns, while additional hypotheses relate to a “new economy effect” and market manipulation, as described below.
Liquidity. Amihud and Mendelson (1986) find a positive relation between quoted bid‐ask spreads and the risk‐adjusted expected return. Liquidity increases on pick day, implying that expected returns should decrease (i.e., current prices increase). Also, the initial level of liquidity may matter, since it is negatively related with liquidity gains on the pick date. Thus, we expect excess returns to be negatively related to PQBAS[0, 0], the average PQBAS on the pick date, and positively related to PQBAS[100, −6], the initial level of liquidity.
Visibility. As before, the visibility proxy is log(NEWS). Since stocks with higher initial visibility have greater liquidity gains, they may also have higher event‐day returns.
Past returns. As before, past price changes are captured by ERET[−5, −1], the excess returns in the 5 days prior to the event day.
Initial market impact. Web sites emphasized the low float and high price impact of stock picks. Such stocks also facilitate a “pump and dump” strategy of buying the stock, organizing a buying cascade via the Web sites and then selling out shortly after the pick is made public. We include:
SOUT[−100, −6]: shares outstanding from 100 days to 6 days before the pick day. | |||||
New economy. The valuation effects may be an outcome of investor fascination with Internet stocks, a so‐called new economy effect. For example, Ofek and Richardson (2003) argue that investors in Internet stocks were relatively overoptimistic. We use the Internet dummy variable:
INTERNET: equal to one for firms in an Internet‐related business and zero otherwise. | |||||
For day t, Rit is the buy‐and‐hold return of stock i, Rmt is the Russell 2000 index return, and I[−5,0] is a dummy variable equal to one in the interval [−5, 0] and zero otherwise. Then 
, the sum of the OLS estimates of a1 and a2 in (11). The mean 
; 
is positive and statistically significant for most stocks, so systematic risk increases during the event period.
The results are in table 9. The initial specification tests the hypotheses of liquidity, visibility, and past returns. The coefficient of past returns ERET[−5, −1] is negative and statistically significant, indicating that stocks with lower excess returns in the 5 days prior to the pick date enjoy higher excess returns on the pick day. This is consistent with Lehman (1990), who finds stock price reversals at weekly intervals. As hypothesized, the coefficient of PQBAS[0, 0] is negative and statistically significant while the coefficient of PQBAS[−100, −6] is positive and statistically significant. Thus, stocks with low initial liquidity and high pick‐day liquidity have the larger pick‐day excess returns. The log(NEWS) is positive and statistically significant at the 10% level, but the result is not robust since the significance disappears in later specifications. Beta boosts the adjusted 
from 18% to 41% and its coefficient is positive and statistically significant, indicating that stocks with higher announcement period systematic risk have higher pick‐day excess returns. SOUT[−100, −6] and the INTERNET dummy do not have statistically significant coefficients. Finally, of fundamental factors, only the coefficient of log(MV) is positive and statistically significant.
We tested for cascades by including PBUY, the percent of total trades that are purchases in the 3 minutes just prior to pick time, as an additional explanatory variable in the regression. While PBUY has the right sign (i.e., positive), its coefficient is not statistically significant.
In summary, Nasdaq stock picks with lower initial liquidity and higher pick‐day liquidity have larger excess returns on the event day. For OTCBB stock picks (discussed in the appendix), only log(NEWS) is statistically significant and its coefficient is negative, indicating that stock picks with higher initial media exposure have lower pick‐day excess returns.
VIII. Long‐Horizon Performance of Stock Picks
Earlier, we examined the postpick performance of stocks for 60 days and found that the average postpick return gains are reversed within 10 days of the pick day. However, analysis of individual stocks (not reported) reveals that some picks maintain return gains beyond 10 days, while other picks that did poorly (well) in the first 10 days tended to do better (worse) later. To get a clearer idea about the longevity of pick‐day gains, and to better control for differences in firm characteristics, we examine the monthly performance of stock picks relative to matched samples for a period of 1 year from the pick month. We also consider their postpick quarterly operating results to assess whether the picks were simply “bad” firms whose stock prices were manipulated. Since the picks have lower liquidity compared to stocks with similar MV and BMV in the pre‐event fiscal year (table 5), we compare their performance to a sample of stocks matched on MV, BMV, and liquidity.
Survivorship bias is an important issue in long‐run analysis, especially for small firms, as stock picks and matched firms are acquired, delisted, or liquidated over the course of a year. We initially select 43 Nasdaq firms that were not acquired, delisted, or liquidated and had Nastraq (trade and quote) data for 1 year from the pick month. They are matched to 43 Nasdaq firms, also with 1 year of Nastraq data, by omitting firms with missing data and then repeating the match. To address survivorship bias, we then redo the analysis with the full sample of 60 Nasdaq picks and a matched sample of 60 Nasdaq firms, after using OTCBB data (wherever available) for the postdelisting period. For returns, turnover and market value data, there are only 14 missing months out of 720 for the Nasdaq picks26 and no missing month for the matched firms. For bid/ask data, there are 75 missing months for the Nasdaq picks, and 24 missing months for the matched firms. Accounting data were unavailable for five firms.
For performance indicator X, the period‐to‐period change 
, where Xm is the closing value of X in period m. For each stock, the average change for horizon k is 
where 
is the event period. Thus, 
is the average percent change in performance of a stock pick over horizon k, relative to the matched firm.27 We report the cross‐sectional mean and median differences in 
between the two samples for different horizons k, with a positive number indicating a higher value for the stock picks.
A. Matching Based on MV, BMV, and the Bid‐Ask Spread
Table 10 reports results for the Nasdaq picks and a sample matched on MV, BMV, and PQBAS. Panel A reports differences in average monthly changes for RETURN, calculated relative to the closing price of the previous month, and for TOVER, PQBAS, SOUT, and MV. Consider first the mean statistics for the 43‐firm sample. Relative to the pre‐event month, the mean TOVER in the event month (
) is higher for stock picks by more than 400%, and this increase is statistically significant. After 1 year (
), the mean TOVER is higher for the picks by 35% per month. The mean SOUT is higher for the picks in 7 out of 12 months, with a difference of about 3.5% per month after 1 year, and this increase is also statistically significant. Differences in the mean MV are not statistically different from zero in the first 6 months, although they are higher for the picks after that, with a difference of about 7.7% per month after 1 year. Mean differences for PQBAS and returns are generally not statistically significant. Thus, the average Nasdaq stock pick outperforms the average matched firm for up to a year after the pick month, especially with respect to turnover and shares outstanding.
Turning to the median statistics, we find that stock picks have higher median increases in TOVER at most horizons, and the increases are statistically significant; the difference is 22% per month after a year. They also have higher median increases in MV, RETURN, and SOUT at most horizons. For example, at 
, the picks have higher median MV, RETURN, and SOUT of 3% per month, 1.8% per month, and 0.04% per month, respectively. The median PQBAS is lower for the picks after 4 months by 8.5% per month, and the decrease is statistically significant. Overall, both mean and median numbers indicate that the average Nasdaq stock pick has higher turnover and shares outstanding up to 1 year after the event compared to the matched sample.
Next, consider the full sample of 60 Nasdaq picks. Figure 3 shows mean cumulated returns, MV, and TOVER (plotted against the right y‐axis) relative to matched samples. Stock picks have better performance in all cases. Panel B of table 10 shows that, while the mean difference in performance is generally not statistically significant except for TOVER, the median performance of picks is superior, especially for 7 months from the pick month. During this 8‐month period, the picks have higher median TOVER in every month, higher median SOUT in 7 months, higher median RETURN and MV in 4 months, and lower PQBAS in 2 months. Although the evidence for superior returns and MV is weaker in the last 4 months, the median MV is higher for picks after 1 year by almost 5% per month. These results remain qualitatively intact when we put returns, turnover, and market value for the picks equal to zero in the 14 missing months. Hence, even after accounting for survivorship bias, Nasdaq picks continue to have higher TOVER and SOUT compared to similar firms, especially for the median stock in the first 8 months after the pick. There is also evidence (albeit weaker in the full sample) of higher MV and returns for stock picks for a few months after the event day.
Fig. 3.— Market impact of Nasdaq stocks from event month to 1 year after
To assess the operating results of picks, table 11 reports differences in quarterly changes in revenue‐to‐market value RMV, BMV, and net income NI. There are no statistically significant differences between pick and nonpick firms in operating results. Hence, there is no detectable ability of Web sites to forecast firms with “good” fundamentals. Finally, table 12 reports that the mean or median change in NEWS during 6 months after the event month, relative to the 6 months before, is similar for stock picks and matched firms.
B. Matching Based on MV, BMV, and Shares Outstanding
Since Web site owners explicitly chose low‐float stocks, we also compare the performance for 43 Nasdaq picks with a sample matched on MV, BMV, and shares outstanding SOUT. The results (not reported) are broadly similar to those using PQBAS. The mean and median TOVER and the median SOUT are higher for Nasdaq picks at most horizons. The mean MV of stock picks is higher after a year by 5% per month while the median RETURN is higher after 9 months by 2% per month, and these increases are statistically significant. Revenue, net income, and media coverage are not statistically different for stock picks and matched firms.
C. Summary of Results for Long‐Horizon Analysis
Nasdaq picks have greater gains in turnover and investor base compared to the matched sample, especially during the first 8 months of the post‐event year. The result is robust to survivorship bias and to the liquidity variable used in the matching procedure. OTCBB stock picks (discussed in the appendix) have larger gains in turnover even after 1 year. Nasdaq and OTCBB stock picks also have higher market value and returns for a few months after the event. Operating results of Nasdaq picks in the post‐event year are similar to those of matched firms.
IX. Conclusion
In this article, we examine stock picks by Internet Web sites and find that they generate large return and liquidity gains on the event day. More surprising, liquidity remains higher 1 year after the event, compared with a sample matched on size, book to market, and liquidity in the pre‐event year. We find that initial liquidity is the key determinant of changes in liquidity and excess returns on the pick day. After controlling for firm characteristics, changes in volume and volatility, and past returns, stocks with lower initial liquidity have proportionately greater liquidity gains on the pick day. Further, stocks with lower initial liquidity and higher pick‐day liquidity have higher pick‐day excess returns. These results support the idea that stock picks may act as a coordination device to push initially illiquid stocks to a higher liquidity equilibrium.
While implicit coordination is not illegal, the SEC frowns on explicit market manipulation. It has warned of “pump and dump” sites that defraud investors with false statements and whose owners profit from share positions in the touted stocks.28 Indeed, the small size, lack of visibility, and transparency of stock picks may facilitate market manipulation. In our sample, we do not find compelling evidence for or against market manipulation. The initial float, price level, or price impact of the stocks (features that make manipulation easier) are unrelated to event‐day returns. Returns and trading activity jump in the week before the event, which may indicate “pumping” by Web site owners but could also be caused by other factors (such as information leakage). Finally, operating results of stock picks in the post‐event year are no worse than that of similar companies. Whatever the intentions of Web site owners, however, investors appear to benefit from increased liquidity of the stock picks.
Firms could be harmed if the “momentum” sites in our study, which typically do not engage in fundamental research, add “noise” to the stock price and thus create transitory volatility. However, we find that, while volatility increases around the time of stock picks, in the year after the pick volatility is not higher compared with similar firms.
Publicity generated by Web sites may prove beneficial to small companies. Compared with more traditional channels such as investment newsletters, the Internet can increase transparency by cheaply and simultaneously disseminating information to many people (Leinweber and Madhavan 2001), facilitating fair disclosure.29 Our study shows that such synchronous information disclosure has liquidity externalities above and beyond its specific news content, especially for small stocks. Good news may create lasting liquidity gains and facilitate the creation of new liquidity pools (Madhavan 2000), but the danger is that bad news may push small stocks into a persistent low‐trade equilibrium.
The halcyon days of stock‐picking Web sites coincided with the market boom of the last years of the 1990s. Could we see a resurgence of this phenomenon any time soon? Since the cost of setting up these Web sites is low, investor appetite for rapid and implausible return gains may “trigger” another wave of momentum sites, even during periods when stock prices are not booming. However, any resurgence may only be temporary because, given the difficulty of weeding out manipulative sites from those merely providing publicity, such sites are unlikely to gain legitimacy in the mainstream investing community.
Appendix Summary of Results for OTCBB Stock Picks
We briefly summarize the key results for OTCBB stock picks. Complete results are available from the authors. If the methodology or variable definition is different from that for the Nasdaq analysis, this is noted in the text. Otherwise, they should be assumed to be the same.
Figure A1 plots mean levels of CUMRET1, CUMRET2, VOLUME (divided by 20,000), and PQBAS for OTCBB stock picks from 100 days before to 60 days after the pick day. CUMRET1 and CUMRET2 are defined as in figure 2. PQBAS is plotted on the right y‐axis. Prices decrease till shortly before the pick day, surge just before and on the pick day, and then fall. CUMRET2 falls slightly below zero 13 days after the pick day and, unlike for the Nasdaq picks, is strongly negative at the end of 60 days. VOLUME jumps on the event day and falls just after but remains higher than initial levels. PQBAS drops from 16% initially to around 13% on the pick day and then drifts up by another 1% or so after 60 days.
Fig. A1.— Market impact of OTC stocks from 100 days before to 60 days after pick day
As with Nasdaq stocks, we follow OTCBB stock picks from 100 days before to 60 days after the stock pick for the two samples described in the text. Formal mean and median tests confirm that liquidity gains for the OTCBB picks persist up to 60 days after the event. Further, the gains do not diminish for post‐1999 picks or picks without positive news. Unlike the Nasdaq picks, however, excess returns are strongly negative from 11 days to 60 days after the stock picks, and volatility does not decrease in the postpick period.
To explain liquidity gains, the change in volatility, CVOLATILITY, is regressed on initial volatility times the spread, 
, log(NEWS), and the control variables used for analyzing Nasdaq picks. To measure volatility, we use APRCH, the absolute value of the daily price change: 
where Close is the closing price and Open is the opening price of the day.
Like the Nasdaq picks, stocks with higher initial volatility and transaction costs have greater reductions in volatility on the event day. However, when CPQBAS is the dependent variable, the initial spread PQBAS[−100, −6] is not statistically significant, contrary to results for the Nasdaq picks. The evidence regarding prior visibility is weak and not robust. Specifically, log(NEWS) is negative and statistically significant at the 10% level when CVOLATILITY is the dependent variable, whereas it is positive and statistically insignificant when CPQBAS is the dependent variable.
To explain return gains, excess returns are regressed on liquidity, visibility, and past return variables, and control factors used for analyzing Nasdaq picks, such as the announcement beta. The mean BETA increases from 2.61 initially to 2.76 during the announcement period (which is defined to be from 5 days before to the pick date), and this increase is statistically significant. The results show that only log(NEWS) is statistically significant. Its coefficient is negative, indicating that OTCBB stock picks with higher initial media exposure have lower pick‐day excess returns.
The matching procedure is the same as with Nasdaq firms, except that we use Compustat instead of CRSP to obtain the initial sample of OTCBB firms, from which the matched sample is subsequently drawn. A comparison of OTCBB stock picks with an MV‐BMV‐matched sample, for the pre‐event fiscal year, reveals that the picks have lower shares outstanding SOUT and inferior earnings. They also have higher median PQBAS, but the difference is not statistically significant.
Given that OTCBB stock picks have lower shares outstanding than similar firms, to examine long‐run performance, we match the picks to a sample based on MV, BMV, and SOUT. Since few OTCBB firms have bid‐ask spread data for 1 year after the event, we do not match on PQBAS. We follow 46 stock picks and matched firms with complete price data for 1 year from the pick month. The results show that the median TOVER is higher for the stock picks in all months. Median RETURN and MV are higher for the picks up to 3 months after the event, after which the differences are not statistically significant. The difference in SOUT is also not statistically significant. We do not compare operating results for OTCBB firms due to lack of data. The mean NEWS is higher for the stock picks after the event, but the median NEWS is lower. In summary, OTCBB stock picks have higher turnover than matched firms and higher returns and market value for 3 months after the event.
Reference
- Admati, Anat R., and Paul Pfleiderer. 1988. A theory of intraday patterns: Volume and price variability. Review of Financial Studies 1:3–40.
- Amihud, Yakov, and Haim Mendelson. 1986. Asset pricing and the bid‐ask spread. Journal of Financial Economics 17:223–49.
- Banerjee, Abhijit. 1992. A simple model of herd behavior. Quarterly Journal of Economics 107:797–818.
- Barber, Brad M., and Terrance Odean. 2002. All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors. Working paper, University of California, Davis.
- Biais, Bruno, Lawrence Glosten, and Chester S. Spatt. 2002. The microstructure of stock markets. Discussion Paper no. 3288, Center for Economic Policy Research.
- Busse, Jeffrey A., and T. Clifton Green. 2002. Market efficiency in real time. Journal of Financial Economics 65:415–37.
- Camerer, Colin, and Keith Weigelt. 1991. Information mirages in experimental asset markets. Journal of Business 64:463–93.
- Chan, Wesley S. 2003. Stock price reaction to news and no‐news: Drift and reversal after headlines. Journal of Financial Economics 70:223–60.
- Cooper, Michael J., Orlin Dimitrov, and P. Raghavendra Rau. 2001. A rose.com by any other name. Journal of Finance 56:2371–88.
- Cutler, David M., James M. Poterba, and Lawrence H. Summers. 1989. What moves stock prices? Journal of Portfolio Management 15:4–12.
- Daniel, Kent D., David Hirshleifer, and Avanidhar Subrahmanyam. 1998. Investor psychology and security market under‐and‐over reactions. Journal of Finance 53:1839–85.
- DeLong, J. Bradford, Andrei Shleifer, Lawrence H. Summers, and Robert J. Waldmann. 1990. Noise trader risk in financial markets. Journal of Political Economy 98:703–38.
- Dow, James. 2005. Is liquidity self‐fulfilling? Journal of Business 77:895–908.
- Dunnett, Charles W. 1955. A multiple comparisons procedure for comparing several treatments with a control. Journal of the American Statistical Association 50:1096–1121.
- Ellis, Katrina, Roni Michaely, and Maureen O’Hara. 2000. The accuracy of trade classification rules: Evidence from Nasdaq. Journal of Financial and Quantitative Analysis 35:529–51.
- Falkenstein, Eric G. 1996. Preferences for stock characteristics as revealed by mutual fund portfolio holdings. Journal of Finance 51:111–35.
- Fama, Eugene F., and Kenneth R. French. 1992. The cross‐section of expected stock returns. Journal of Finance 47:427–64.
- Fleming, Michael. 2003. Measuring treasury market liquidity. Economic Policy Review 9 (September): 83–108.
- Gadarowski, Christopher. 2002. Financial press coverage and expected stock returns. Working paper, Cornell University.
- Glosten, Lawrence. 1987. Components of the bid‐ask spread and the statistical properties of transactions prices. Journal of Finance 42:1293–1307.
- Huberman, Gur, and Tomer Regev. 2001. Contagious speculation and a cure for cancer: A nonevent that made stock prices soar. Journal of Finance 56:387–96.
- Klibanoff, Peter, Owen Lamont, and Thierry A. Wizman. 1998. Investor reaction to salient news in closed‐end country funds. Journal of Finance 53:673–99.
- Kyle, Albert S. 1985. Continuous auctions and insider trading. Econometrica 53:1315–35.
- Lehman, Bruce. 1990. Fads, martingales and market efficiency. Quarterly Journal of Economics 60:1–28.
- Leinweber, David J., and Ananth Madhavan. 2001. Three hundred years of stock market manipulation. Journal of Investing 10:1–10.
- Madhavan, Ananth. 2000. In search of liquidity in the Internet era. Working paper, Investment Technology Group Inc.
- Madhavan, Ananth, Matthew Richardson, and Mark Roomans. 1997. Why do security prices change? A transaction‐level analysis of NYSE stocks. Review of Financial Studies 10:1035–64.
- Metrick, Andrew. 1999. Performance evaluation with transactions data: The stock selection of investment newsletters. Journal of Finance 54:1743–75.
- Newey, Whitney K., and Kenneth D. West. 1987. A simple, positive semi‐definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55:703–8.
- Ofek, Eli, and Matthew Richardson. 2003. Dotcom mania: The rise and fall of Internet stock prices. Journal of Finance 58:1113–38.
- O’Hara, Maureen. 1995. Market microstructure theory. Cambridge, MA: Blackwell.
- Pagano, Marco. 1989a. Endogenous market thinness and stock price volatility. Review of Economic Studies 56, no. 2:269–87.
- ———. 1989b. Trading volume and asset liquidity. Quarterly Journal of Economics 104, no. 2:255–74.
- Rashes, Michael S. 2001. Massively confused investors making conspicuously ignorant choices (MCI‐MCIC). Journal of Finance 56:1911–27.
- Roll, Richard. 1984. A simple implicit measure of the effective bid‐ask spread in an efficient market. Journal of Finance 39:1127–39.
- Romer, David. 1993. Rational asset‐price movements without news. American Economic Review 83:1112–30.
- Schultz, Paul. 2000. Regulatory and legal pressures and the costs of Nasdaq trading. Review of Financial Studies 13:917–57.
- Stoll, Hans R. 2000. Frictions. Journal of Finance 55:1479–1514.
- Welch, Ivo. 1992. Sequential sales, learning and cascades. Journal of Finance 47:695–732.
- Wysocki, Peter D. 1999. Cheap talk on the Web: The determinants of postings on stock message boards. Working paper, University of Michigan Business School.
-
* We thank a referee and also editor Albert Madansky for insightful comments on earlier drafts. We also thank Jonathan Berk, Larry Glosten, Charlie Himmelberg, Prem Jain, Charles Jones, Jim Mahoney, Lubos Pastor, Robert Schwartz, Rene Stulz, Ingrid Werner, and seminar participants at the American Finance Association meeting in 2003, Rutgers University, and the Federal Reserve Bank of New York. We acknowledge Michael Emmet and Priya Gandhi for excellent research assistance. The views stated here are ours and do not necessarily reflect the views of the Federal Reserve Bank of New York or the Federal Reserve System. We are responsible for all errors. Contact the corresponding author, Asani Sarkar, at asani.sarkar@ny.frb.org.
-
1. For example, after its stock was posted, Derma Sciences Inc. issued a press release stating that “the company is not aware of any recent corporate developments that would serve as a basis for substantial increases in its common stock’s trading volume or price” (press release, Derma Sciences Inc., November 15, 1999).
-
2. Subscribers provide their e‐mail addresses to the momentum Web sites and receive reminders about forthcoming stock picks and notification of the selections. The sites typically do not charge for the subscription.
-
3. Rashes (2001) briefly compares the bid‐ask spread on high‐volume and normal‐volume days.
-
4. More generally, research shows that stock‐price reactions to events can be disproportionate to its direct news content. For example, in Romer (1993), rational reassessments of fundamentals occur without the arrival of outside news. In Daniel, Hirshleifer, and Subrahmanyam (1998), investors overweight private signals and discount pubic signals due to behavioral biases. In experimental economics, “information mirages” (i.e., overreaction to uninformative trades) occur (Camerer and Weigelt 1991). Empirically, Cutler, Poterba, and Summers (1989) conclude that economic fundamentals or news cannot fully explain extreme market movements.
-
5. In related work, Admati and Pfleiderer (1988) show that there exists a unique equilibrium with bunching by uninformed traders resulting in liquid and illiquid periods of trading. Also, on‐the‐run Treasury notes trade at a yield discount to off‐the‐run Treasury notes (Fleming 2003), although they are close substitutes, perhaps because of investors’ self‐fulfilling expectations that the notes will not be traded once they go off the run. (We thank Jonathan Berk for this example.) Biais, Glosten, and Spatt (2002) and Dow (2005) discuss models of liquidity externality.
-
6. It is likely that the search engines failed to find some stock‐picking sites that advertised mainly through Internet chat room postings and word of mouth.
-
7. Visit http://www.otcbb.com/aboutOTCBB/comparison.stm for details.
-
8. Alternative benchmark indices, such as the Nasdaq index, and alternative volatility measures, such as the standard deviation of buy (or sell) prices and the sum of absolute returns, give qualitatively similar results.
-
9. We also estimated the following liquidity measures: PEBAS, the proportional effective half‐spread defined as the effective half‐spread EBAS divided by M, where EBAS is the absolute difference between P and M; PBDEPTH, the bid depth as a percent of TDEPTH; and MMAKERS, the number of market makers. Results using these measures are similar to those reported and available from the authors.
-
10. Schultz (2000) argues that the Roll spread, estimated with intraday data, is an accurate measure of trading cost for Nasdaq stocks in his sample.
-
11. We also estimate the adverse selection component of the spread, as in Madhavan, Richardson, and Roomans (1997). Except for the event day, these estimates generally prove less reliable than PROLLAS since we sometimes obtain negative estimates of adverse selection costs, and so we only report results for the latter.
-
12. This measure is similar to that in Stoll (2000), who regresses ΔPt on a constant, PVIMBt, and PVIMBt−1 for daily intervals. However, the estimated coefficient on PVIMBt−1 is insignificant in more than 95% of his regressions. Our price impact measure is applicable to both intraday and daily intervals.
-
13. We use CRSP rather than Compustat to obtain the initial sample of Nasdaq‐traded firms because Compustat records only the current exchange listing of a stock. In particular, firms that traded on Nasdaq in the prior fiscal year but subsequently moved to OTC appear on Compustat as OTC firms.
-
14. The test statistic controls for (1) the fact that each of the test means is being compared to the same control and (2) the overall type 1 error rate for all comparisons.
-
15. Another site, stockmarketpicks.com, sold its domain name and is now a different business.
-
16. It is rational for cascade investors to ignore their own information and buy a stock if enough investors before them also bought it (Banerjee 1992; Welch 1992).
-
17. Unlike stock picks, however, they have high trading volume, positive past returns, low book‐to‐market ratios, and high analyst following and were used by short sellers to spread negative information (Wysocki 1999).
-
18. In particular, since the time required to upload information to the Web page is 1–3 minutes, the Web sites may have started the uplink process slightly before the reported pick time.
-
19. In unreported results, we find support for these conclusions using a number of other liquidity measures. PEBAS is about 2% lower at market close relative to the event interval, similar to PQBAS. Estimates of the adverse selection component of the spread are lower from 3 minutes after pick time till market close, relative to the prior period. The median number of market makers, MMAKERS, per minute increases sharply from 3 minutes before to 60 minutes after the pick, consistent with the behavior of TDEPTH. The ratio of bid to total depth, PBDEPTH, is also lower during this period, consistent with market makers selling from inventory to customers.
-
20. Other measures also show that liquidity gains persist beyond the event day. PEBAS declines sharply on the pick day and then increases for 20 days (but it remains below initial levels) before leveling off. MMAKERS increases from 8 in the [−100, −6] interval to 11 on the pick day before falling to initial levels. PBDEPTH declines on pick day as market makers meet increased liquidity demands, but it reverts to original levels thereafter.
-
21. We thank a referee for suggesting this analysis.
-
22. Examples of news leading to large positive returns are an award of a new patent, intention to develop a new Web site, acquisitions, or a positive investment opinion on the stock.
-
23. Falkenstein (1996) and Chan (2003) find that the number of news stories is correlated with firm size, price, and turnover, while Gadarowski (2002) finds that it is correlated with analyst coverage.
-
24. We also examine the role of volume run‐ups by including EVOL[−5, −1], the ratio of the average volume in the [−5, −1] interval to the average volume in the [−100, −6] interval, but its coefficient is not significant in the regression.
-
25. The results do not change if we use the most recent quarter values of these variables. Further, since the sample firms sometimes delay reporting their 10Q results for several months, investors may not have access to the most recent quarter information on the pick day.
-
26. One month was missing due to acquisition and 6 months due to liquidation. We do not know why the remaining 7 months are missing.
-
27. With accounting values, the mean difference between pick and nonpick firms may be a large negative number. This may happen, for instance, if some pick firms have little or no monthly change in revenues, whereas the matched firms experience a large positive percent change in revenues in the same period (either due to improved performance or because revenues were low in the previous period).
-
28. See “Pump&Dump.com: Tips for Avoiding Stock Scams on the Internet,” Securities and Exchange Commission, September 2000.
-
29. In addition, investors can assess the quality of the advice better as it is unaffected by confounding news released during the time delays associated with more traditional channels.