Economic subjects | Investments, Stock exchange » Masulis-Shivakumar - Intraday Market Response to Equity Offering Announcements, A NYSE AMEX NASDAQ Comparison

Source: http://www.doksinet Intraday Market Response to Equity Offering Announcements: A NYSE/AMEX-NASDAQ Comparison Ronald W. Masulis* Owen Graduate School of Management Vanderbilt University Nashville, TN 37203 Lakshmanan Shivakumar * London Business School London, NW1 4SA United Kingdom March 11, 1999 * We are grateful to Tarun Chordia, Bill Christie, Dean Furbush, Roger Huang, Debra Jeter, Craig Lewis, Tim McCormick, Paul Schultz, C. Sinha, Hans Stoll and finance workshop participants at Tulane and Vanderbilt Universities for useful discussions. Financial support from the Financial Markets Research Center at Vanderbilt University is gratefully acknowledged. Source: http://www.doksinet ABSTRACT This study uses transactions data to compare the speed of price adjustments to seasoned equity offering announcements by NYSE/AMEX and NASDAQ stocks. We find that NASDAQ stocks react faster to equity offering announcements than NYSE/AMEX stocks over the first 15 minutes following the

news release. The faster NASDAQ response is surprising given that NASDAQ stocks have on average a smaller offering size, lower equity capitalization and less frequent trading activity than NYSE/AMEX stocks. Further analysis suggests that the faster price reaction of NASDAQ stocks is due to several differences in market structure across the two types of markets. We find evidence that all the following conditions contribute to more rapid NASDAQ stock price adjustment: greater risk-taking by NASDAQ dealers, more rapid electronic order execution on NASDAQ, a more potent information trading threat (SOES bandits) on NASDAQ, stale limit orders on the NYSE/AMEX and a less efficient price discovery mechanism at the open of the NYSE/AMEX. Source: http://www.doksinet 1. Introduction This study compares the efficiency with which stock prices incorporate new information under alternative market structures. Using transactions data, we compare the relative speed of price reactions to a major

corporate announcement by National Association of Security Dealers Automated Quote System (NASDAQ) listed stocks against New York Stock Exchange (NYSE) and American Stock Exchange (AMEX) listed stocks. These exchanges have distinctly different market structures, which could affect the price adjustment process. For example, NASDAQ employs an electronic listing of competing dealer quotes and the NYSE and AMEX employs a specialist system with a centralized limit order book. These and other structural differences make for an informative experiment as to whether the speed of price discovery differs across these markets and if it differs, what market features are the likely causes. The importance of market structure has recently elicited much interest in the finance literature. However, most empirical research examining the impact of market structure on stock price behavior focuses on stock return volatility (e.g, Amihud and Mendelson (1987), Stoll and Whaley (1990), Masulis and Ng (1995)).

In this study, we use several years of transactions data to examine and contrast the price, volume and bid-ask spread reactions of stocks listed on the NASDAQ with those listed on the NYSE/AMEX to a major corporate event. The corporate event we study is an initial announcement of a seasoned equity offering (SEO) by an U.S issuer using a firm commitment underwriting contract. 1 By comparing the speed with which prices reflect new information across distinctly different market structures, we further our understanding of how trading mechanisms can aid or impede price discovery in capital markets. To preview our results, we uncover a surprising finding, that the price adjustment process following a SEO announcement is at least 15 minutes faster for NASDAQ listed stocks than for NYSE/AMEX listed stocks. This evidence is striking given that NASDAQ stocks have smaller equity capitalization and on average are less frequently traded. We examine a number of potential explanations for this

result. We find that differences in speeds of price adjustment across exchanges can not be attributed to differences in (1) the magnitudes of two-day announcement returns, (2) the frequencies of trading halts, (3) the frequencies of overnight and daytime announcements, or (4) equity capitalization or bid-ask spreads. In short, the result that NASDAQ stock prices are updated more rapidly than NYSE/AMEX prices, does not appear to be a statistical artifact of the data. It suggests that stock characteristics are relatively less important than market microstructure characteristics in explaining the differing speeds with which stock 1 See Lease, Masulis and Page (1991) for an examination of SEO offering date returns and the impacts of market microstructure changes associated with this event. 1 Source: http://www.doksinet prices incorporate the economic effects of new information. On the other hand, we find evidence that a faster electronic execution system on NASDAQ, the threat of SOES

bandits, the greater risk-bearing of NASDAQ dealers and stale limit orders on the NYSE/AMEX, all appear to contribute to faster price reactions by NASDAQ stocks to SEO announcements. For overnight announcements, we find that NYSE/AMEX price reactions are slowed by a relatively less efficient opening pricing mechanism. The remainder of the paper is organized as follows. In section 2 we review institutional details, discuss our methodology and present descriptive statistics of our data. The next section presents our initial statistical results and explores possible data biases. Section 4 examines the economic hypotheses for the observed differences in price adjustment speeds to SEO announcements between NASDAQ and NYSE/AMEX stocks using univariate analysis. In section 5, we shift to a multivariate framework to more rigorously evaluate competing hypotheses. We summarize our findings and draw conclusions in the last section 2. Institutional Details, Methodology and Descriptive

Statistics 2.1 Differences in organizational structure on NASDAQ and the NYSE/AMEX The NYSE and AMEX are order-driven continuous auction markets, the linchpins of which are market makers called specialists. Specialists facilitate continuous trading by posting quotes for their own account or by reflecting the best quotes on their limit order book, which represent a centralized depository for limit orders to buy or sell stocks at specified prices or better. Limit orders play a major role in providing immediacy and liquidity on the NYSE/AMEX as seen by the fact that over 80% (88%) of the volume on NYSE (AMEX) arise from trades in which the specialists do not participate for their own account. 2 In contrast to the NYSE/AMEX markets, the NASDAQ market is based on a competing dealer system in which each dealer continually posts firm bid and ask quotes on an electronic screen. Further, there is no central limit order book on NASDAQ, although limit orders may be left with individual

broker-dealers. However, unlike the NYSE/AMEX, limit orders on NASDAQ do not drive the posted quotes since dealers are not required to consider limit orders in setting their quotes. 3 Also, dealer competition is diminished by rules allowing directed order flow to less competitive dealers who agree to meet the best quotes. 2 1992 NYSE and AMEX fact books respectively. Stoll (1985) estimates that in approximately half of trades in which specialists participate, they act as brokers in executing limit orders. 3 After our observation period, NASDAQ revised their rules to require limit orders held by individual dealers be taken into account in setting their quotes. However, there is no centralized book of limit orders 2 Source: http://www.doksinet Other important institutional differences exist between the NYSE/AMEX and NASDAQ exchange systems, which may affect the speed of price adjustment. Specialists on the NYSE/AMEX are not allowed to trade ahead of limit orders at the same

prices and their quotes often reflect the limit order book rather than commitments by individual specialists. This reduces specialists’ incentives to immediately adjust quotes to new information, since execution of the posted quotes often has no impact on their inventory positions or wealth. Furthermore, limit orders can not be updated instantaneously, nor can they be conditioned on public information such as the stock’s last transaction price. As a result, limit prices are temporarily stale immediately after public announcements. This slow updating of limit orders can delay revisions in the best bid and ask quotes. Moreover, hitting stale limit orders following announcements with negative price impacts may not be easy for traders in the NYSE/AMEX markets due to the uptick rule, which prevent traders from short selling on a down tick or a zero tick following a down tick. More generally, arbitrage of stale limit orders is discouraged by bid-ask spreads and the price impacts of

market orders, which together can exceed the potential profits from arbitrage when information effects are modest and/or secondary markets are not highly liquid. NASDAQ dealers must post firm bid and ask quotes for at least 1000 shares and can not rely on the limit orders of other investors. Neither dealers nor investors are constrained from short selling by an uptick rule. 4 Thus, if dealers do not immediately adjust their quotes to new information, they are vulnerable to other traders selectively hitting their stale quotes, causing them trading losses. Hence, NASDAQ dealers have strong financial incentives to revise their quotes immediately following public announcements, even in the absence of trades. These arguments suggest NASDAQ quotes should react faster to public announcements than NYSE/AMEX quotes. The speed of market reaction to offering announcements may also differ across exchanges, due to cross-exchange differences in the speed of trade execution systems. Incoming market

orders on the NYSE/AMEX are manually executed on the exchange floor to expose them to potentially new opposite market orders from specialists and floor brokers that can better the currently posted quotes. 5 In contrast, during the period examined in this study, the best price quotes of NASDAQ stocks were electronically disseminated. For orders of a 1000 shares or less, electronic execution on the SOES system was available. Thus, NASDAQ investors could often execute trades more expeditiously, but with little opportunity for price enhancement. These 4 5 As of September 6, 1994, NASDAQ adopted a rule similar to the NYSE uptick rule. This delay can be reduced by using the Super DOT electronic execution system on the NYSE. 3 Source: http://www.doksinet differences in trading mechanisms may facilitate faster execution of sell orders on NASDAQ than on the NYSE/AMEX. For firms announcing SEOs overnight, the price setting mechanism at the open can also affect the speed of market response.

Opening prices on NYSE/AMEX are set in a process resembling a call auction. More specifically, each specialist examines overnight buy and sell orders and the stock’s limit order book before choosing an opening price that crosses supply and demand. Any order imbalances at the open can be offset by the specialist trading for his/her own account or by delaying the opening of trading to allow offsetting orders to arrive. On NASDAQ the prices are set according to the usual competing dealer system, which is used through out the trading day. Prior to the official open, competing dealers tend to begin posting quotes at varying times over the prior two hours. Stoll and Whaley (1990) demonstrate that the greater market power of the specialist at the open increases the noise in NYSE/AMEX opening prices, which may slow the price adjustment process for firms making overnight SEO announcements in these markets. One or more NASDAQ dealers are generally the underwriters of a firm’s SEO. This

situation implies that one or more NASDAQ dealers can have superior information concerning impending SEO announcements if inadequate fire walls exist between the trading desk and the underwriting desk of these dealers. This information advantage is greater for daytime SEOs when other investors have no grace period to investigate the pricing implications of news prior to the commencement of trading. This potential dealer information advantage can cause the prices of these stocks to adjust more quickly to SEO announcements than otherwise. In contrast, most NYSE/AMEX specialist firms do not underwrite SEOs. Hence, there are several institutional reasons that potentially cause differences in the price adjustment process between NASDAQ and NYSE/AMEX listed stocks. Given differences in organizational structure across exchanges, such as: the existence of a limit order book, trading rules, financial risks borne by dealers, separation or integration of broker-dealer and underwriting functions,

speed of trade execution and the trading mechanism used at market open, we investigate the null hypothesis that these institutional characteristics have negligible influence on the speed of price adjustments to new information. 2.2 Measurement of intraday stock price reactions to corporate news releases We test the null hypothesis of no differences in the average speed of price adjustment across exchange samples by analyzing stock returns in fifteen-minute trading intervals around 4 Source: http://www.doksinet SEO announcements. Our intraday analysis focuses on the 12 fifteen-minute intervals (3 hours) before a SEO announcement, the fifteen-minute interval containing the announcement and the 12 fifteen-minute intervals (3 hours) after the SEO announcement. Since the NYSE, AMEX and NASDAQ are generally open from 9:30 a.m to 4:00 pm EST, there are 26 fifteen-minute trading intervals per trading day. We analyze stock returns around both daytime and overnight SEO announcements. For

daytime announcements, “event interval 0” is defined as the fifteen-minute trading interval containing the SEO announcement. In contrast, for overnight announcements, “event interval 0” is defined as the first fifteen-minute trading interval following the SEO announcement. All other event intervals are identified relative to interval 0. Stock returns are based on the midpoints of the best bid and ask quotes at the end of each 15 minute trading interval. The first trading interval of each day is defined as beginning at 4:00 p.m of previous trading day and ending at 9:45 am of the current day. This interval captures the return over the first fifteen-minutes of trading including opening trades, but does not distinguish between the overnight return and the return over the first 15 minutes of the trading day. Interval returns are based on the last inside bid-ask average or transaction price in the interval. For overnight returns, the last quote or price in the first 15 minute

trading interval of the following day is used. 6 Since stock characteristics such as trading activity and bid-ask spreads can affect the speed of price adjustment, we also examine the SEO announcement induced changes in these characteristics across our two exchange samples. For each trading day, we define the timeweighted spread as the average of the quoted spreads weighted by the proportion of the trading day each spread was outstanding. 7 Abnormal spreads and number of trades for an event interval i are measured as follows: Ji ∑ (QSPRD Abnormal spread i = ij − ASPRD i )Tij j=1 Abnormal number of trades i = DSPRD K i - ATRD i DTRD 6 (1) (2) If there is no quote or transaction price in the first 15 minute interval, then the first price or quote in the first half of the next interval is used, when it is available. If that is also unavailable, then the last price or quote for the first interval where a price or quote is available is used and the overnight interval is

lengthened to include all periods up to the end of that interval. Alternatively, treating such overnight returns as missing does not alter our basic results. NASDAQ stocks almost always have quotes or prices available in the first interval following the official open. This is not the case for NYSE/AMEX stocks 7 Since trades can occur within quoted spreads, particularly on NYSE/AMEX, the quoted spreads may not represent the actual spreads faced by investors. Hence, we also examine whether effective spreads yield qualitatively similar results to those based on quoted spreads. Since they do, we do not report them 5 Source: http://www.doksinet where QSPRD ij is the quoted bid-ask spread for the jth quote in interval i (j=1.J), T ij is the proportion of interval i for which the jth quote is outstanding and K i is the actual number of trades in event interval i. ASPRD i , and ATRD i are respectively the interval i time-weighted average quoted spread, and average number of trades over the

benchmark period. The benchmark period consists of events days –30 to -5, where event day 0 is the SEO announcement date. Finally, DSPRD and DTRD are the benchmark period daily average spread and number of trades, respectively. 8 For each event interval, we test the null hypothesis that mean stock returns (quoted spreads or number of trades) are insignificantly different from zero using a bootstrap resampling technique. This technique is preferred to the parametric tests since the probability distributions of intraday returns, spreads and trades are unknown. The resampling procedure used here is similar to the one employed by Barclay and Litzenberger (1988) and controls for the composition of firms and the announcement’s time of day. This procedure compares individual stock means for the same 15 minute trading interval in the event date and the benchmark period. For each trading interval, an empirical distribution is obtained by randomly sampling the corresponding 15 minute trading

interval in the benchmark period. This comparison of means yields the significance level for this statistical test. 2.3 Data sources and sample characteristics The sample consists of initial announcements of common stock offerings made by NYSE, AMEX and NASDAQ listed stocks during the period January 1990 to December 1992. We obtain details of the completed primary and/or secondary equity offerings from Securities Data Corporation’s ‘New Issues’ database. We exclude non-underwritten offerings, rights offerings, standby offerings, shelf offerings, non-U.S issues, offers by non-US issuers, pure secondary offerings and simultaneous offerings of common stock and other securities. After these exclusions, the sample consists of 458 offerings of NASDAQ listed stocks and 408 offerings of NYSE/AMEX listed stocks. We obtain the initial public announcements of stock offerings by searching the Dow Jones Text and the LEXIS on-line services. 9 We exclude offerings when another major firm news

release, such as earnings, dividends, splits, investment or financing decisions, is made between the close of day -1 through the close of day +1. We also exclude offers where the initial 8 We alternatively scaled abnormal spread by ASPRDi and the abnormal number of trades by ATRDi. The results are qualitatively identical. We also examined the abnormal number of shares traded and found event time patterns similar to the abnormal number of trades. 9 More specifically, we searched the Press Release News Wire, the Business Wire and the Dow Jones News Service to obtain the earliest announcement time. 6 Source: http://www.doksinet announcements occur within one month of a change in stock listing from NASDAQ to the NYSE/AMEX to avoid any contamination from exchange listing effects around SEO announcements. 10 Data on intraday stock prices, quotes and trades are obtained from the Institute for Studies of Security Markets (ISSM) database. 11 Since the ISSM data for stocks on NASDAQ are

available only for the period January 1990 to December 1992, we limited our sample of SEO announcements for both NASDAQ and NYSE/AMEX listed stocks to this period. We require our sample to have transaction data available on ISSM for event days -1 to +1 and for at least 10 days in the benchmark period (i.e, in the 25 trading day period between event days -30 and -5) After all the above exclusions, our final sample consists of 320 NASDAQ stocks and 253 NYSE/AMEX stocks with SEO announcements. For each event interval in the benchmark and event periods, we treat returns, and changes in spreads and trades as missing up to the first interval for which quotes or trades are available. 2.4 Descriptive statistics for the SEO samples Figure 1 presents the intraday frequency distribution of SEO announcements in each halfhour interval between 7 a.m and 7 pm On both NASDAQ and NYSE/AMEX markets, the frequency of SEO announcements across the business day exhibits a bimodal pattern. Most firms

announce SEOs between 9 a.m and 11 am or between 4 pm and 6 pm (after the market close), with a greater likelihood of a morning announcement. On NASDAQ, very few offering announcements occur prior to 8 a.m, while NYSE/AMEX firms generally tend to wait until 9 a.m before making SEO announcements These findings suggest that firms purposefully select the timing of their announcements relative to when active trading in the stock occurs. Table 1 presents summary statistics for the SEO sample. Panel A presents summary statistics for firms classified by exchange listing. From this panel, we observe that the prior yearend equity capitalization of the NYSE/AMEX sample is substantially larger than the NASDAQ sample. On average, stock offers increase shares outstanding by 229% on NASDAQ, but by only 18.3% on the NYSE/AMEX Also, typical offer prices for NYSE/AMEX stocks are significantly higher than for NASDAQ stocks. Not surprisingly, gross proceeds average $35 million for NASDAQ issuers,

which is significantly smaller than the average $79 million for 10 Christie and Huang (1994) show that the bid-ask spreads for firms that switch from NASDAQ to NYSE/AMEX decrease dramatically after they leave NASDAQ. 11 We exclude trades that are reported out-of-sequence since the precise times of these trades are unknown. We also exclude non-BBO eligible quotes since these quotes do not represent firm commitments on the part of the market maker and are non-tradable. In addition, for 8 NASDAQ and 7 NYSE/AMEX events, ISSM data is not available due to mismatches of ticker symbols and CUSIP numbers between our SEO database and the ISSM database. 7 Source: http://www.doksinet NYSE/AMEX issuers. In contrast, the average dollar underwriter spread for NASDAQ offers is statistically indistinguishable from that of NYSE/AMEX offers at conventional significance levels. This indicates that NASDAQ SEOs are riskier on average, since they require similar underwriter compensation for a much

smaller offer size. 3.0 Univariate Analysis of the Differences in Price Reaction Speeds Across Exchanges 3.1 Sample differences in equity capitalization, bid-ask spreads and trading activity Several prior studies of daily and weekly stock returns show that there is a lead-lag relation in stocks of different equity capitalization levels. This evidence is consistent with larger firms adjusting faster to market-wide information than smaller firms. Lo and Mackinlay (1990) find the weekly returns of larger firms tend to lead those of smaller firms. Chan (1994) interprets this as evidence that larger firms react faster to economy-wide information. Further, Chordia and Swaminathan (1999) show that, even after controlling for firm size, a stock price’s speed of adjustment to new information is positively related to its trading activity level. 12 This evidence supports the proposition that stocks with higher equity capitalization experience faster price adjustment to new information,

including initial announcements of SEOs. Also, greater media coverage of large equity capitalization firms suggests that investors in these stocks generally have better access to timely information. Panel A of Table 1 shows that the NYSE/AMEX stocks in our sample have a higher average equity capitalization than the NASDAQ sample, which given prior evidence suggests NYSE/AMEX stocks should exhibit faster speeds of price adjustment to SEO announcements, ceteris paribus. However, there is no reason to believe that other relevant characteristics are identical across our two exchange samples. Thus, we need to carefully examine this critical assumption before proceeding. If stock liquidity is substantially higher on one exchange, then everything else the same, we would expect stocks on that exchange to have a faster average price adjustment speed. Table 1, Panel B presents summary statistics for time-weighted daily averages for several stock liquidity measures (quoted bid-ask spreads,

effective spreads and number of trades) for our two exchange samples in the benchmark period. We find that average quoted spreads and effective spreads are lower for the NYSE/AMEX sample than for the NASDAQ sample, which is consistent with the general findings of Christie and Schultz (1994) and Huang and Stoll (1996). Further, we find that the daily average number of trades is lower for NASDAQ stocks than for NYSE/AMEX stocks. These differences in number of trades are more substantive given that NASDAQ dealers report a 8 Source: http://www.doksinet trade when executing a sell order and again when executing the offsetting buy order, while on the NYSE/AMEX most buy and sell orders are directly crossed and reported as a single trade. Given the lower spreads and greater trading activity of the NYSE/AMEX sample, these stocks should react faster to new information than the NASDAQ sample. 3.2 The size of a stock’s announcement return and its effect on price adjustment speed Before we

evaluate the speed of price reactions across exchanges, we need to examine whether mean two-day SEO announcement effects for the two samples are comparable. This is an important issue since a stock’s price adjustment speed can be a positive function of the announcement effect’s size. Studies by Jennings and Stark (1985) and Woodruff and Senchak (1988) conclude that stock price reaction speeds to firm announcements are positively correlated with the magnitudes of these price reactions. This suggests that differences in the average magnitudes of market reactions to SEO announcements across our two exchange samples can cause differences in their price adjustment speeds as well. Mech (1993) also reports evidence that larger announcement effects are more rapidly incorporated into stock returns. To assess how the magnitude of a news event impacts a market’s reaction speed, we assume that the conventional two-day event window (event days 0 and 1) is sufficient to capture the market’s

full price reaction to an SEO announcement. Table 1, Panel C presents two day announcement returns based on trade prices for our NASDAQ and NYSE/AMEX samples. Market adjusted stock returns are also reported where the market adjustment is calculated by subtracting the contemporaneous two day return on an equally weighted index of NYSE/AMEX or NASDAQ stocks. We observe that the announcement period mean return (market adjusted return) for NASDAQ stocks is a significantly negative –1.75% (-195%) In comparison, we find a larger and significantly negative mean return (market adjusted return) of -2.70% (-278%) for NYSE/AMEX stocks. The slower price reaction on the NYSE/AMEX is somewhat surprising, since market microstructure theory and our evidence on stock characteristics predicts the opposite. 3.3 Measuring speeds of price adjustment to corporate news We initially measure price adjustment speeds from returns based on the midpoints of the best bid and ask quotes at 15 minute trading

intervals around the SEO announcement. The result of this horse race across the two exchanges is presented in Table 2. For NASDAQ stocks, there are significant negative mean returns in intervals –1, 0 and +1, with the largest return being in interval 0. For NYSE/AMEX stocks, there are significant negative returns in intervals 0 through 12 In section 5.2, we examine the importance of dollar spreads in affecting stock price adjustment speeds 9 Source: http://www.doksinet 3, with intervals 0 and 1 having similar large negative returns. Statistical significance holds whether we use boot-strap probabilities or a nonparametric sign test. 13 Looking at cumulative returns starting in interval 0, we see that the NASDAQ sample is more negative out to interval +12 where the two means are nearly indistinguishable. We also see from the median returns that the larger NASDAQ mean return in interval 0 is not driven by a small number of outliers. To further investigate the evidence that NYSE/AMEX

stock prices on average react more slowly to firm announcements, we examine how individual stock price changes in interval 1 are related to changes in interval 0. In Table 3, we report transition probabilities where individual stock returns are categorized as negative, zero (with or without trading) or positive. Panel A reports NASDAQ evidence and Panel B covers the NYSE/AMEX. We observe several interesting patterns in the cell frequencies. First, if the interval 0 return is negative, then the conditional probability of a negative interval 1 return is nearly twice as large for the NYSE/AMEX than for NASDAQ. Second, following no trade or quote revision in interval 0, the conditional probability a negative interval 1 return is nearly twice as large for the NYSE/AMEX sample. Following a positive interval 0 return, the conditional probability of observing a nonpositive interval 1 return is more than twice as likely for the NYSE/AMEX sample This is further evidence that NYSE/AMEX stock

prices take longer to capitalize negative information and are more likely to initially exhibit a positive return, to only later reflecting the negative import of the SEO announcement. These patterns are consistent with a slower price reaction on the NYSE/AMEX, even though the frequency of no trading in both event intervals is greater for the NASDAQ sample. 3.4 Relative Size and Frequency of Overnight and Daytime Announcements The above analysis does not distinguish between announcements made during the daytime trading period and the overnight non-trading period. However one possible explanation for smaller NYSE/AMEX mean returns in the initial 15 minute intervals following SEO announcements is that the NASDAQ sample has relatively more overnight SEO announcements relative to daytime announcements. Since market participants have more time to digest overnight announcements, we expect the market to take less time to capitalize overnight SEO announcements into stock prices. SEO

announcements in non-trading hours represent 35% of the 13 We replicate this experiment with transaction prices and find very similar results. 10 Source: http://www.doksinet NASDAQ sample and 43% of the NYSE/AMEX sample. 14 Since SEO announcements made during non-trading overnight hours are relatively less frequent for NASDAQ listed stocks, the above hypothesis predicts that on average NYSE/AMEX stock prices should exhibit more rapid price adjustment to SEO announcements. This is inconsistent with the empirical evidence As discussed earlier, we expect news to be more rapidly incorporate into prices when there are larger price impacts. Thus, if daytime (overnight) SEO announcements of NASDAQ stocks have significantly larger mean two-day returns compared to NYSE/AMEX stocks, then the average price adjustment speed of NASDAQ stocks to SEO daytime (overnight) announcements would be expected to exceed that of NYSE/AMEX stocks. Comparing mean two-day returns to daytime and overnight

announcements within each exchange sample, we find the mean returns for the daytime and overnight announcements to be very similar in magnitude. 15 Given that the NYSE/AMEX sample has larger two-day mean returns in both announcement samples, we expect NYSE/AMEX stock prices to react faster to both overnight and daytime SEO announcements. This prediction is reinforced by the fact that the differences in offer characteristics across exchanges are similar for both SEO overnight and daytime announcement samples. After separating the SEOs into daytime and overnight announcements, we re-examine price adjustment speed evidence in Table 4. This evidence is again based on 15 minute returns calculated from bid-ask midpoints. Panel A presents daytime SEO announcement means returns which are significantly negative in intervals 0 and +1 for NASDΑQ stocks, with the interval 0 return representing 87% of the two day mean return. 16 The mean returns in event intervals 0 to +2 are significantly

negative for NYSE/AMEX stocks. The interval 0 mean (median) return is 101% (-27%), which is substantially smaller than the corresponding mean (median) return for the NASDAQ sample of –1.49% The interval +1 mean returns are significantly negative for both exchange samples, but the NYSE/AMEX sample exhibits a -.70% mean return, while the NASDAQ sample has only a -.52% mean return Over intervals +2 to +5, the mean returns of the 14 We treat announcements occurring after 9:30 a.m but before the actual opening of trading as overnight announcements, which causes the percentage of overnight announcements to be higher than otherwise. These events are concentrated in the NYSE/AMEX sample. 15 The mean two day returns for daytime and overnight SEO announcements are –1.71% and –180% respectively for NASDAQ stocks and –2.77% and –261% respectively for NYSE/AMEX stocks 16 Mean returns in intervals -2 and –3 are significantly negative for NASDAQ, though much smaller in magnitude than

interval 0 or 1. The mean returns for NYSE/AMEX stocks in intervals -1 and -2 are also negative. This evidence suggests inside information leaks or broadcast delays of corporate press releases for both samples. 11 Source: http://www.doksinet NYSE/AMEX stocks are significantly negative, while those for the NASDAQ stocks are approximately zero. 17 Panel B of Table 4 presents the 15 minute returns for overnight SEO announcements. We find for NASDAQ listed firms, that the interval 0 return has a large negative mean (-1.60%) and median (-1.17%), while the interval +1 return has a positive mean but a zero median Also, the mean returns for intervals +2 to +5 are all negative, with intervals +4 and +5 being statistically significant. This suggests some delay in the speed of adjustment of NASDAQ prices to fully capitalize overnight SEO announcements. For overnight SEO announcements by NYSE/AMEX firms, the interval 0 mean return is -.92% Further, for intervals +1 through +6, NYSE/AMEX mean

returns are all negative and significantly different from zero in half the sample, suggesting lagged price adjustments. Our results show that NASDAQ stocks react faster to SEO announcements, regardless of whether the announcement occurs overnight or in the daytime. 4.0 Alternative Hypotheses for Differences in Price Reaction Speeds across Markets 4.1 Overnight announcement effects when markets use different opening mechanisms One potential explanation for the differences across exchanges in average price adjustment speeds to overnight SEO announcements, is that the NYSE/AMEX employs a less efficient opening pricing mechanism. Consistent with this hypothesis, Francis, Pagach and Stephan (1991) find that overnight earnings announcements by NYSE listed stocks are not fully reflected in their opening prices. Amihud-Mendelson (1987) and Stoll-Whaley (1990) report that opening prices on the NYSE exhibit much greater volatility than closing prices. In contrast, Chan-Christie-Schultz

(1995) report that price discovery at the open on NASDAQ appears to occur within the first 5 minutes. These studies suggest that the opening price effect could induce the NASDAQ sample to appear to generally capitalize overnight SEO information faster, when in fact the only problem is the use of a less efficient opening pricing mechanism on the NYSE/AMEX. To further examine efficiency of NASDAQ day 0 opening prices, we calculate the overnight return from the day –1 close to the first quote on day 0, and the return from the first quote to the official open on day 0, as well as the daytime return on day 0. Examining Table 5, we find the market reaction to NYSE/AMEX overnight announcements occurs primarily during the trading period of day 0. This is in marked contrast to the market reaction to NASDAQ overnight 17 Mean stock returns for the 2 hours prior to event interval 0 are very similar across the NYSE/AMEX and NASDAQ samples. This suggests that pre-announcement leakage effects can

not explain the differences in the speed of price adjustment in the two samples. 12 Source: http://www.doksinet announcements, where a large portion of the price reaction is concentrated between the first quote and the official open of day 0, plus an insignificant negative return over the trading period of day 0. This suggests that a partial explanation for a faster price adjustment speed on NASDAQ is its opening pricing mechanism. However, this can not be the full explanation, since NYSE/AMEX daytime announcements also exhibit slower price adjustments, as seen in Table 4, Panel A. 4.2 Differences in trading halts around SEO announcements If the proportion of SEO announcements associated with trading halts is dissimilar across the two samples, trading halts could measurably affect the price adjustment speeds across exchanges. This is a serious concern given that the NYSE/AMEX has significantly more liberal rules for calling trading halts. Specifically, NYSE/AMEX rules allow a

specialist to request a trading halt when an unusually large order flow imbalance occurs or when there is an impending or an actual news announcement that is likely to have a substantial stock price impact. 18 In contrast, trading halts on NASDAQ are allowed only for news dissemination and pending news announcements and not for order imbalances. This biases our results against the NYSE/AMEX We investigate this issue by first examining the percentage of trading halts on event day 0 in the two exchange settings. NYSE/AMEX trading halts occur for 21% of daytime SEO announcements and 3% of overnight SEO announcements. 19 In contrast, NASDAQ trading halts occur in only 2.4% of daytime SEO announcements and 18% of overnight announcements 20 On average, SEO related trading halts last just over one hour. To eliminate the effects of trading halts, we redefine interval 0 to be the first 15 minute trading period following a trading halt in which a SEO is announced. 21 We then re-examine the

average 15 minute returns of the NASDAQ and NYSE/AMEX daytime announcement samples. Not surprisingly, there is little change in the NASDAQ results. The NYSE/AMEX mean return in interval 0 is mildly more negative. Thus, this adjustment increases the apparent speed of NYSE/AMEX stock price reactions to SEO news. However, compared to the NASDAQ sample, the NYSE/AMEX sample 18 During a trading halt on the NYSE/AMEX, the specialist engages in price exploration by issuing indicator quotes, which are lower and upper bounds for the probable reopening price. Following the halt, trading reopens with a call auction market. 19 Most trading halts associated with overnight announcements are actually delayed openings. 20 While the ISSM codes indicates that no trading halts occur around SEO announcements by NASDAQ stocks, on further examination, we find 7 cases where stocks exhibit sequences of bid and ask quotes with zero values around SEO announcements. Conversations with NASDAQ officials

indicate that these cases are almost surely trading halts and we treat them as such in our analysis. 21 If a SEO announcement is made before or after a trading halt on day 0, we excluded it from our sample. Also, excluding all trading halts does not qualitatively alter our results. When we separate stocks having trading halts on the SEO announcement date, we find no significant differences in average two day announcement returns for these stocks compared to the remainder of the sample. Neither did we find any 13 Source: http://www.doksinet continues to exhibit a slower price reaction speed, indicating that trading halts are not the primary cause of this effect. 22 4.3 Trading activity patterns around SEO announcements Differences in the relative speeds of price adjustment across exchanges can reflect differing trader response speeds across exchanges. Exposing orders to floor traders for price improvement and the existence of an uptick rule on NYSE/AMEX can slow down the relative

response of NYSE/AMEX stock prices to corporate news releases, especially negative news. Requiring orders to be sent to the specialist post for manual execution, introduces a potential lag in order execution and thus, in price adjustments. Furthermore, the uptick rule on NYSE/AMEX can prevent information traders from easily taking advantage of overpriced stale limit orders (to buy) following a SEO announcement by shorting the stock to hit the stale limit orders. In contrast, traders on NASDAQ can use SOES for rapid execution of trades of 1000 shares or less. To examine the empirical importance of this issue, we analyze the abnormal number of trades in the 25 fifteen-minute intervals around SEO announcements. The patterns across event time in the abnormal number of trades are found in Figure 2. We observe that the mean abnormal number of trades in interval 0 is greater for NASDAQ stocks than for NYSE/AMEX stocks for both daytime and overnight SEO announcements. Further, in the

subsequent intervals NASDAQ stock trading activity falls relatively faster than NYSE/AMEX stock trading. This evidence is consistent with traders on NASDAQ responding faster to SEO announcements relative to traders on NYSE/AMEX. The disparity in price adjustment speeds appears to be positively correlated with the level of trading activity. In addition, this evidence is consistent with both manual execution of trades and the uptick rule having a negative effect on NYSE/AMEX price adjustment speeds and initial trading activity levels following negative news releases. It is also consistent with the stale limit order hypothesis discussed in section 4.6 4.4 Effects of SOES bandits and other information traders In this section, we examine the effects of order execution systems and rules defining market maker quote responsibilities to explain the disparity in reaction speeds across exchanges. First, with the mandatory use of the small order electronic trading system (SOES), NASDAQ dealers are

more vulnerable to losses from information traders when they are slow to revise their quotes. This creates a strong incentive for NASDAQ market makers to react more quickly to new information, which is reinforced by the fact that all quotes on NASDAQ are firm and are made by significant differences in equity capitalization, offering size, underwriter spreads or average trading activity during the benchmark periods. 14 Source: http://www.doksinet competing dealers. These dealers do not benefit from investors actively placing limit orders, who share the adverse selection risk of trading with informed investors. Furthermore, information traders can execute larger trades, more rapidly on NASDAQ using SOES than on the NYSE/AMEX. Under this hypothesis, we predict an increased frequency of 1000 share orders on NASDAQ immediately after SEO announcements with associated large price impacts. According to Harris-Schultz (1997) information traders on NASDAQ rely on SOES for their trade

execution. Under NASDAQ rules, a dealer must accept five 1000 share trades over SOES before there is an automatic pause to allow the dealer to revise his or her quotes. 23 To test the importance of SOES trading, we compare the frequency of abnormal 1000 share trades around the SEO announcements for both NASDAQ and NYSE/AMEX stocks. In the case of NYSE/AMEX stocks, this variable is a proxy for information traders who do not necessarily use electronic execution. Table 6 presents evidence on the frequency of abnormal 1000 share trades in 15 minute intervals around SEO announcements for our two exchange samples, separated into daytime and overnight announcements. We observe significant increases in the frequency of abnormal 1000 share trades for the NASDAQ sample beginning in interval 0. For daytime announcements, there is a significant increase in these abnormal trades for four 15 minute intervals following the SEO announcement, with the largest being interval 0 and then slowly falling

for the next three intervals. For overnight announcements, the pattern is similar but the size is smaller and only intervals 0 and 1 are significant. This is consistent with overnight announcements being more quickly capitalized into prices. However, the overall pattern appears to bear out the importance of SOES bandits and/or information traders on NASDAQ. 24 4.5 Evidence of superior access to SEO information by some NASDAQ dealers Since SEO underwriters are often dealers in the same NASDAQ stocks, they could realize an information advantage concerning forthcoming SEO announcements, enabling them to adjust their quotes more quickly to SEO news releases. Under this hypothesis, NASDAQ dealers who are also underwriters of the SEO are predicted to adjust their quotes to SEO announcements 22 The daytime announcement interval 0 mean returns for the NASDAQ and NYSE/AMEX samples are – 1.58% and –106% respectively These results are available from the authors 23 Trades can be

electronically executed up to and including 1000 shares. After SEO announcements, traders with superior abilities to interpret public information can immediately execute trades at the maximum allowable size to maximize their expected profit. 24 There is also a much smaller positive series of abnormal 1000 share trades for daytime announcements by NYSE/AMEX stocks for intervals 0 through 3 which are only significant in interval 1. This suggests some information based trading as well. 15 Source: http://www.doksinet more quickly than other dealers on NASDAQ or specialists on the NYSE/AMEX. However, when there are many competing dealers on NASDAQ, this effect is likely to have little impact on the stock’s inside quotes, whereas when the number of competing dealers is small, the effect is likely to be much more discernible. To test this hypothesis, we examine whether or not the price reaction speeds among NASDAQ stocks is negatively related to the number of competing dealers. An

alternative hypothesis is that dealers have no discernible information advantage given the firewall that exists between a broker-dealer’s trading and underwriting desks. Instead, a larger number of dealers is likely to proxy for higher average trading volume. Again, other studies find that higher trading volume enhances the speed of price responses to new information. This yields the prediction that price reaction speeds are positively related to the number of dealers. Evidence on how the number of competing NASDAQ dealers impacts a stock’s price adjustment speed is presented in Table 7. In this table, we present interval 0, 0 plus 1 and 0 through 3 returns segmented into quintiles based on a stock’s number of competing dealers. We find for daytime announcements that stocks with the fewest dealers (quintiles 1 and 2) have interval 0 mean returns that are smaller than other quintiles. This evidence does not support the hypothesis that dealers who underwrite SEOs have superior

information. The evidence is consistent with the number of dealers being a proxy for a stock’s trading activity. However, a more robust test of this hypothesis would control for trading activity levels and then evaluate the marginal impact of the number of dealers on the price reaction speed. We examine this issue further in section 5. 4.6 Effects of stale limit orders on the NYSE/AMEX Once a SEO is announced, with its typical negative price reaction, limit order investors have incentives to cancel their old orders to buy (bids) stock. Likewise, information traders have incentives to hit old limit orders to buy if they own the stock or can short it quickly and cheaply. No such incentives exist for limit orders to sell (ask). Until new limit orders arrive, the asymmetric response of limit order investors and information traders implies a widening in the bid-ask spread. However, the SEO may also trigger a temporary rise in adverse selection, which can inhibit the arrival of new limit

orders and discourage specialists from stepping in front of limit orders with more aggressive quotes. 25 If new limit orders are slow to arrive immediately 25 Kim and Verrecchia (1994) show that the presence of traders with superior abilities to interpret public information such as corporate news releases generates adverse selection for liquidity providers immediately after public announcements. Liquidity providers faced with increased adverse selection following SEO announcements have incentives to increase their bid-ask spreads and/or decrease depth of their quotes to 16 Source: http://www.doksinet after SEO announcements, then the depth of limit orders to buy is likely to drop, as the old limit orders are hit or canceled. Thus, the stale limit order hypothesis predicts that: (1) stock prices exhibit a slow decline as the highest limit orders to buy are sequentially hit or replaced, (2) bid depth gradually falls as informed traders hit stale limit orders and new limit orders

are slow to arrive, (3) the number of trades occurring at the bid rises relative to other trades and (4) bid-ask spreads widen. While the implications of the stale limit order hypothesis are not unique, we see several strands of supportive evidence in Table 3. First, the frequency of negative returns for the same stocks in both intervals 0 and 1 is twice as large for the NYSE/AMEX sample. Second, the frequency of an initial positive return followed by a negative return or no trade in interval 1 is more than double for the NYSE/AMEX sample. This evidence suggests that NYSE/AMEX stock prices do not immediately and unbiasedly reflect the import of SEO announcements, at least for the first 15 minutes following the informations release. Figure 3 presents the mean abnormal spreads (as defined in section 3) in the 25 fifteenminute intervals around SEO announcements classified by time of day and by exchange. This figure shows that abnormal spreads following SEO announcements are significantly

greater for the NYSE/AMEX than for NASDAQ. 26 For overnight SEO announcements, there is a small positive abnormal spread for NASDAQ stocks, but the mean abnormal spread for NYSE/AMEX stocks is significantly positive in event intervals 0 to +7, based on boot strap procedures. This finding is consistent with the existence of stale limit orders and/or an inefficient opening mechanism on NYSE/AMEX and suggests that it takes on average about two hours following an overnight SEO announcement for the limit order book to be fully updated. Likewise for NYSE/AMEX daytime announcements, the abnormal spreads are significantly positive for intervals 0 to +3, which suggests that it takes about an hour after a daytime SEO announcement for the limit order book to be fully updated. In sharp contrast, the mean abnormal spread is negligible for NASDAQ stocks. Table 8 details evidence on NYSE/AMEX limit order depth at the best bid and ask quotes around SEO announcements. We find the average depth of the

bid (or best limit order) falls significantly in intervals 0 through 8 following overnight SEO announcements and in intervals 0 limit their exposure to losses and thus, minimize the negative impact on their expected profits. CopelandGalai (1983) and Glosten-Milgrom (1985) develop formal models of this effect 26 We also examine the location of the trade prices relative to the bid and ask quotes. We find evidence that both samples exhibit trades closer to the bid, after daytime and overnight SEO announcements. This is evidence that dealers are setting their bid and ask quotes asymmetrically around the true price. This pattern is consistent with dealers not wanting to increase their inventory position in a stock immediately after a SEO announcement, possibly because of adverse selection concerns. 17 Source: http://www.doksinet through 3 for daytime SEO announcements. At the same time, the average depth at the ask (or best limit order to sell) falls after overnight SEO announcements

for intervals 0 through 12, while after daytime SEO announcements it remains unchanged. This evidence is consistent with stale limit orders to buy being canceled or hit by market orders, thereby lowering the quoted depth at bid. 27 Another prediction of the stale limit order hypothesis is an increase in trades of NYSE/AMEX stocks at the bid (buy) for several 15 minute intervals after an SEO announcement. We focus on abnormal trades at the bid, measured this by number of trades at the bid in a 15 minute interval minus average number of trades at the bid in the benchmark period for the same 15 minute interval, which we then scaled by this same benchmark average. We see in Table 9 that for daytime announcements on the NYSE/AMEX, trades at the bid rise significantly over intervals 0 through 5. This supports our earlier evidence that stale limit orders are a factor in the slow price response of NYSE/AMEX stocks following daytime SEO announcements. In contrast, NASDAQ stocks show a much

smaller rise in trades at the bid following daytime SEO announcements, and only after a delay of 45 minutes. Not surprisingly, following overnight SEO announcements, when there is more time to place new limit orders and cancel old limit orders, neither sample exhibits any significant patterns. In summary, the evidence in Tables 8 and 9 and Figure 3 is consistent with the stale limit order hypothesis being a partial explanation for the slower price reaction speeds of NYSE/AMEX listed stocks. To more rigorously test the stale limit order hypothesis, we use a multivariate approach in the next section to determine the power of abnormal depth, spread and percentage of trades at bid in explaining cross-sectional differences in price reaction speeds. If our previous findings are due to stale limit orders, then we expect the associated proxy variables to be related to the speed of price reaction for NYSE/AMEX stocks, but not for NASDAQ stocks. 5.0 Multivariate Tests for Differences in Speeds

of Price Adjustments across Markets In section 4, we uncovered evidence supportive of a number of hypotheses related to a market’s trading environment, which help explain differences across exchanges in price reaction speeds to firm-specific news. Using a multivariate framework, we seek to evaluate the marginal explanatory power of these alternative hypotheses. We estimate a cross-sectional regression model of individual stock returns over one or more 15 minute intervals following an SEO announcement. We examine the descriptive power of our trading environment hypotheses using 27 Since no reliable depth data is available for the NASDAQ market, we are unable to replicate this analysis for the NASDAQ sample. 18 Source: http://www.doksinet as explanatory variables: abnormal 1000 share trades (MSHRTRD), abnormal change in spread (ABNSPRD), and average share turnover (TRNOVER), each multiplied by an exchange listing indicator and abnormal NYSE/AMEX bid depth (BIDDEPTH). We also

include in our model both offer size (OFFSIZE), measured by the percentage change in shares, and multiplied by an exchange indicator and the number of competing NASDAQ market makers (NMKT). 28 To the extent that differences in price reaction speeds across exchanges remain unexplained, we also include an NYSE/AMEX listing indicator in our model. The intercept represents the portion of the mean return that is independent of these explanatory variables. Our basic statistical model for explaining SEO announcement returns, R 0i , over one or more 15 minute trading intervals is: R 0i = α 0 + α 1 NYAM + α 2 OFFSIZE*NYAM + α 3 OFFSIZE NASD + α 4 ABNSPRD*NASD + α 5 ABNSPRD NYAM + α 6 TRNOVERNASD + α 7 TRNOVER*NYAM + α 8 MSHRTRDNASD + α 9 MSHRTRDNYAM + α 10 BIDDEPTH*NYAM + α 11 NMKR + ε I (3) where all the variables are defined in Table 10. We estimate the model separately for overnight and daytime SEO announcements. In the regression estimates that follow, we use stock returns in

the 15 minute event interval 0, and cumulative returns over intervals 0 plus 1 and 0 through 3 as measures of price adjustment speeds across markets. Stocks that react faster to announcements should generally have more negative price reactions in interval 0. Moreover, if differences in interval 0 returns are due to differences in long-run price reactions to SEO announcements across the two samples, these differences are likely to persist for several subsequent trading intervals. Alternatively, if these differences are due to disparate price reaction speeds across exchanges, then they are likely to weaken as the length of the return interval increases. We test this hypothesis by examining whether the strength of the explanatory variables increases or declines as the dependent variable is cumulated over two or four adjacent trading intervals, rather than being just the interval 0 return. The predictions of the various market structure hypotheses for our model stock price adjustment

speeds across exchanges follow. We expect a negative coefficient for share turnover based on the findings of Chordia and Swaminathan (1999) that stocks with greater trading activity have faster price adjustments to news. We expect a negative coefficient on the proportion of 28 We did not control for differences in two day announcement returns due to endogeneity concerns. We also did not use other offer descriptors due to their likely multicolinearity with existing explanatory variables. 19 Source: http://www.doksinet 1000 share trades on NASDAQ, since SOES bandits and other information traders strongly prefer this trade size and their trades should increase the quote reaction speed of dealers. The impact of the number of NASDAQ dealers on price adjustment speed is ambiguous. It is positive if a specific dealer-underwriter is exploiting an information advantage, since this information advantage is less likely to affect a stock’s inside quotes as the number of dealers rises. It is

negative if the number of dealers proxies for trading activity and the level of information produced about the stock, increasing these variables should increase a stock’s price reaction speed. Offer size is meant to control for a possible adverse selection effect Larger offers (as a percent of outstanding shares) are predicted to be associated with greater stock price reaction, which implies a negative coefficient. 29 The extant literature finds that offer size (measured by percentage change in shares outstanding) is negatively correlated with two day announcement returns, we also expect a negative coefficient on this variable for our much shorter announcement period. The abnormal change in spread and the abnormal depth at the bid can proxy for the effects of either stale limit orders for NYSE/AMEX stocks or increased adverse selection borne by market makers in both exchange samples. 30 If stale limit orders to buy are picked off or canceled after negative news, then spreads of

NYSE/AMEX stocks should widen and bid depth should decrease until new limit orders arrive. If this stale limit order hypothesis is valid, then we expect a negative (positive) coefficient on abnormal spreads (abnormal depth) only for NYSE/AMEX stocks. Under this hypothesis, the coefficient should decrease as returns interval is lengthened since there is more time for new limit orders to arrive. However, if abnormal spreads or abnormal depth is capturing a temporary increase in adverse selection faced by liquidity providers in both markets, then we expect (1) the coefficient not to shrink as the return interval is increased and (2) the coefficient to be significant for both exchange samples. Stocks facing greater adverse selection are predicted to have more negative SEO announcement reactions. These same stocks are likely to exhibit larger positive (negative) spread (depth) reactions to the SEO information. This creates a negative (positive) relation between stock announcement returns

and abnormal spreads (abnormal depth). 5.1 Estimates of a multivariate model of stock price adjustment speed 29 See Krasker (1986) for an extension of the Myers-Majluf model to stock offers of varying size. We also examine several additional explanatory variables. These variables had statistically insignificant coefficients. These other explanatory variables are: the proportion of prices at the bid, equity capitalization and a trading halt indicator. The insignificant coefficient on equity capitalization likely to be due to its high correlation with share turnover and offer size. 30 20 Source: http://www.doksinet The first segment of Table 10 reports stock returns over event interval 0. The latter segments are based on regressions where the dependent variable is the cumulative return over intervals 0 and 1; and intervals 0 through 3 respectively. To keep the time interval of the explanatory variables consistent with that of the dependent variable, mean percentage spreads are

averaged across the same event intervals represented by the announcement returns. We expect most explanatory variables will lose significance and exhibit smaller coefficients as we move to longer return intervals, because this exceeds the length of time the price adjustment speed in one exchange outpaces the other. White standard errors are used to avoid biased standard errors due to heteroscedasticity, which is a common problem in cross sectional data. Examining the regressions shown in the first segment of Table 10, we find that a number of explanatory variables have statistically significant coefficients and their signs are qualitatively consistent with the earlier hypotheses. We find evidence that cross-sectional differences in interval 0 stock returns are partly explained by BIDDEPTH, ABNSPRD and OFFSIZE for NYSE/AMEX stocks and by MSHRTRD and TRNOVER for NASDAQ stocks. In the latter segments of Table 10, which represent longer return intervals (i.e 0 and 1 or 0 through 3), the

magnitude and significance of the regression coefficients decreases, with the exception of ABNSPRD and OFFSIZE in the NYSE/AMEX sample. These findings indicate that differences in market characteristics (BIDDEPTH for NYSE/AMEX stocks and MSHRTRD and TRNOVER for NASDAQ stocks) help explain the different price adjustment speeds in the two exchange settings. More specifically, there is also evidence that a stock’s price adjustment speed is negatively affected by stale limit orders (BIDDEPTH*NYAM is significantly positive), and positively affected by both SOES bandit trading (MSHRTRD*NASD is significantly negative) as well as the stock’s normal trading level (TRNOVER*NASD is significant negative). The consistent significance and non-diminishing size of the coefficients on ABNSPRD and OFFSIZE, as the return interval is lengthened, is evidence consistent with these two variables capturing an adverse selection effect. 31 In addition, ABNSPRD is significant for overnight announcements

in both exchange samples, which also supports the adverse selection prediction. In comparing results for overnight and daytime SEO announcements, we see that while the coefficients of each explanatory variable differs, they are qualitatively consistent. Generally speaking, stale limit orders, SOES bandits and the frequency of trading activity have more significant impacts for daytime SEO announcements. This is to be expected since overnight 31 When we include the abnormal number of trades as an additional regressor, we find that its coefficient is statistically significant and that the differences in returns across exchanges is reduced. This is added evidence supporting the hypothesis that NASDAQ traders react faster to information releases. 21 Source: http://www.doksinet announcements allow greater reaction time for investors and dealers before trading resumes. The one exception to this interpretation is that the number of dealers is only significant for overnight announcements.

The positive coefficient on the number of dealers supports the hypothesis that faster price adjustment on NASDAQ is, in part, due to a dealer’s information advantage from his/her dual role as an SEO underwriter. 32 5.2 Comparison of 15 minute returns across matched exchange samples To further evaluate the robustness of the estimates, we re-estimate our statistical models using SEO announcements by industrial issuers only. We also reduce our larger NASDAQ sample by matching each NYSE/AMEX event with a NASDAQ event having a two-day announcement return most closely matching it. 33 In this way, we create matched samples of similar two-day announcement returns across the two markets. The objective of this procedure is to ensure that differences in mean two day announcement returns and the proportions of utilities and financial issuers in the two samples do not impact our conclusions. 34 The results of matching two-day announcement returns are presented in Table 11 for the same three

segments (representing different dependent variables) as presented in Table 10. We find that the qualitative evidence reported in Table 10 is robust to these modifications of the dataset. Specifically, we find for daytime announcements that both the abnormal 1000 share trades and turnover have significantly negative coefficients for NASDAQ stocks as predicted. We also observe that offer size and abnormal spread have significantly negative coefficients for the NYSE/AMEX sample. Interestingly, the number of NASDAQ dealers is marginally significant and negative, indicating that dealer-underwriters do not have a significant information advantage concerning SEO announcements. When we estimate the regression model of daytime SEO announcements using cumulative returns over intervals 0 and 1, the larger announcement effect of the NASDAQ sample (measured by the coefficient on the NASDAQ indicator variable) is reduced. It becomes 32 This evidence is based on SEOs that include industrial,

utility and financial issuers, though we find similar results when the sample is restricted to industrial issuers. In Section 52, we examine the sensitivity of our results to excluding utilities and financials from the sample. 33 The mean two day return for our sample of SEO announcements by industrial, utility and financial issuers is lower than the market reaction reported in the literature for industrials, but is higher than that reported for financials and utilities (see Masulis and Korwar (1986), Asquith and Mullins (1986) and Polonchek, Slovin and Sushka (1989)). Eliminating utilities and financials makes our two day return more negative. 34 Over 14% of NYSE/AMEX overnight announcements are by utilities, while only 1.8% of NASDAQ overnight announcements are by utilities. This difference is more dramatic for the daytime announcement samples, with 23.8% of the NYSE/AMEX firms being classified as utilities, while only 48% of NASDAQ firms are utilities. The percentage of financial

firms across the daytime and overnight samples and across the two exchange samples are quite comparable and vary between 6.3% and 99% 22 Source: http://www.doksinet statistically insignificant as shown in column one of the second segment of Table 11. Similar patterns are observed for overnight SEO announcements. When the dependent variable is measured over four 15 minute intervals, the results become more attenuated as expected. In evaluating this evidence, it is important to recognize that our procedures are biased against finding a faster reaction for NASDAQ stocks. Nearly all instances of trading halts and trading delays occur on the NYSE/AMEX and when announcements occur in the morning after the official open but before the actual start of trading, we treat them as having occurred overnight. This causes the NYSE/AMEX interval 0 returns to appear to capture more of the overall two-day return than it actually does. Likewise, we define interval 0 to be the first 15 minute period

following a trading halt that includes an SEO announcement, even if this procedure postpones the designation of interval 0 for a number of 15 minute intervals following the interval when the SEO announcement actually occurs. 35 In summary, the regression evidence in Tables 10 and 11 points to several exchange mechanisms as likely causes for the faster price reaction speed to firm-specific news on NASDAQ. Likely causes for the reaction speed differences are: the existence of the small order electronic execution system (SOES) for relatively large trades (up to 1000 shares), significant levels of information trading on NASDAQ, greater dealer risk-bearing on NASDAQ, the use of a superior opening mechanism on NASDAQ and the persistence of stale limit orders on the NYSE/AMEX. However, it is important to recognize that NASDAQ’s price reaction speed lead over the NYSE/AMEX exists only for the first 15 minute interval following an SEO announcement. 6. Summary and Conclusions This study

compares the speed of price adjustment for stocks on the NYSE/AMEX with the speed on NASDAQ to seasoned equity offering announcements using transactions data. Our analysis uncovers several interesting findings. Most importantly, we find that NASDAQ stock prices react faster to equity offering announcements than do NYSE/AMEX stocks. This is surprising given that NYSE/AMEX stocks tend to have lower spreads and greater trading activity. 35 As a further test of whether differences in dollar bid-ask spreads can explain our results, we separately partition our NYSE/AMEX and NASDAQ samples into stocks with dollar spreads above and below their respective median dollar spread levels. We then re-estimate a simple regression model with four indicator variables, one for stocks with high spreads and one for stocks with low spreads in each of the two exchange samples. The results show that for low spreads, NASDAQ stock prices exhibit even more rapid adjustment compared to NYSE/AMEX stock prices.

For the high spread stocks, the difference in adjustment speeds across exchange samples is not statistically significant, even though on average NASDAQ stocks exhibit much more rapid price adjustments compared to NYSE/AMEX stocks. 23 Source: http://www.doksinet Further analysis suggests that the faster reaction of NASDAQ stock prices is due to a greater degree of adverse selection borne by NASDAQ dealers combined with significant information trading by investors using the SOES order system, the existence of stale limit orders on the NYSE/AMEX combined with the up-tick rule and a less efficient NYSE/AMEX opening price setting mechanism. It is noteworthy that all the causes we uncover for NASDAQ’s superior price adjustment speed are due to organizational differences across these exchanges. Finally, we uncover evidence that suggests profitable trading opportunities exist following NYSE/AMEX seasoned equity offering announcements that occur during non-trading hours. Specifically, we

find that on average these stocks fall in price over the next trading day No such apparent profit opportunity is found for NASDAQ listed stocks. The most likely causes of the persistence of this apparent profit opportunity on the NYSE/AMEX are the use of an inefficient opening price setting mechanism and the effects of stale limit orders. 24 Source: http://www.doksinet REFERENCES Amihud, Y., and H Mendelson, 1987, “Trading mechanisms and stock returns: An empirical investigation”, Journal of Finance, 42, 533-553. Asquith, P. and DW Mullins, Jr, 1986, “Equity issues and offering dilution”, Journal of Financial Economics, 15, 61-89. Barclay, M.J and RHLitzenberger, 1988, “Announcement effects of new equity issues and the use of intraday price data”, Journal of Financial Economics, 21, 71-99. Chordia, T. and B Swaminathan, 1999, “Trading volume and cross-autocorrelations in Stock Returns”, forthcoming, Journal of Finance. Chan, K. 1993, “Imperfect information and

cross-autocorrelation among stock prices”, Journal of Finance, 48, 1211-1230. Chan, K.C, W Christie and P Schultz, 1995, “Market structure and the intraday pattern of bid-ask spreads for NASDAQ securities”, Journal of Business, 68, 35-60. Christie, W.G and P Schultz, 1994, “Why do NASDAQ market makers avoid odd-eight quotes?”, Journal of Finance, 49, 1813-1840. Christie, W.G and JHarris and PSchultz, 1994, “Why do NASDAQ market makers stop avoiding odd-eight quotes?”, Journal of Finance, 49, 1841-1860. Christie, W. G and RD Huang, 1994, “Market structures and liquidity: A transactions data study of exchange listings”, Journal of Financial Intermediation, 3, 300-326. Copeland, T. and D Galai, 1983, “Information effects of the bid-ask spread”, Journal of Finance, 38, 1457-1469. Francis, J., D Pagach and J Stephan, 1992, “The stock market response to earnings announcements during trading versus non-trading periods”, Journal of Accounting Research, 30(2), 165-184.

Glosten, L. and P Milgrom, 1985, “Bid, ask and transaction prices in a specialist market with heterogeneously informed traders”, Journal of Financial Economics, 14, 71-100. Green, J.T and SG Watts, 1996, “Price discovery on the NYSE and the NASDAQ: The case of overnight and daytime news releases”, Financial Management, 25(1), 19-42. Harris, Jeffrey H. and Paul H Schultz, 1997, “The importance of firm quotes and rapid executions: Evidence from the January 1994 SOES rule change”, Journal of Financial Economics, 45, 135-166. Huang, R.D and HR Stoll, 1996, “Dealer versus auction markets: A paired comparison of execution costs on NASDAQ and the NYSE”, Journal of Financial Economics, 41, 313-357. 25 Source: http://www.doksinet Jennings, R. and L Starks, 1985, “Information content and the speed of the stock price adjustment”, Journal of Accounting Research, Spring, 336-350. Kim, O. and R Verrecchia, 1994, “Market liquidity and volume around earnings announcements”,

Journal of Accounting and Economics, 17, 41-67. Krasker, W.S, 1986, “Stock price movements in response to stock issues under asymmetric information”, Journal of Finance, 41, 93-106. Lease, R.C, RW Masulis and J Page, 1991, “An investigation of market microstructure impacts on event study returns”, Journal of Finance, 46(4), 1523-1536 Lee, C. MC, and MJ Ready, 1991, “Inferring trade direction from intraday data”, Journal of Finance, 46, 733-746. Lo, A. and C Mackinlay, 1990, “When are contrarian profits due to stock market overreaction?”, Review of Financial Studies, 3, 175-205. Masulis, R.W and A Korwar, 1986, “Seasoned equity offerings: An empirical investigation”, Journal of Financial Economics, 91-118 Masulis, R.W and V Ng, 1995, “Overnight and daytime stock-return dynamics on the London Stock Exchanges: The impacts of ‘Big Bang” and the 1987 stock-market crash”, Journal of Business Economics and Statistics, 13(4), 365-378. Mech, T., “Portfolio return

autocorrelation”, Journal of Financial Economics, 34, 307344 Polonchek, J., MB Slovin and ME Sushka, 1989, “Valuation effects of commercial bank securities offerings: A test of the information hypothesis”, Journal of Banking and Finance, 13, 443-461. Stoll, H., 1985, “The stock exchange specialists system: An economic analysis”, Monograph Series in Finance and Economics, New York University Stoll, H. and R Whaley, 1990, “Stock market structure and volatility”, Review of Financial Studies, 3, No. 1, 37-71 26 Source: http://www.doksinet TABLE 1 Panel A Descriptive statistics of NASDAQ and NYSE/AMEX firms announcing seasoned equity offering in 1990-1992. The t-statistics for the differences in means are based on the assumption of unequal variances across exchanges. EXCHANGE Pre-offer market value of equity ($ million) NASDAQ NYSE/AMEX Mean Mean Median Median (Standard (Standard deviation) deviation) 237.24 1596.36 131.22 381.39 (318.35) (4698.60) t-statistic for

difference in means 5.04 22.90 19.75 (14.06) 18.26 13.93 (14.30) -4.00 18.65 16.25 (9.72) 23.30 21.13 (15.36) 5.30 35.26 28.05 (27.18) 79.03 50.80 (105.31) 8.17 Gross underwriter spread ($ per share) 0.97 0.90 (0.41) 0.94 0.91 (0.40) -0.57 Underwriter spread as % of offering price 5.56 5.50 (1.13) 4.59 4.52 (1.32) -10.95 Change in shares outstanding (%) Offer price ($) Offer amount ($ million) 27 Source: http://www.doksinet TABLE 1 (contd) Panel B Daily average share volume, quoted spreads, effective spreads and number of trades over the benchmark periods for NASDAQ and NYSE/AMEX stocks making seasoned equity offering announcements over 1990-1992. The benchmark period is defined as event days –30 through –5 The t-statistics for the differences in mean are based on White heteroscedasticity robust standard errors. EXCHANGE Dollar quoted spread (cents) NASDAQ NYSE/AMEX Mean Mean Median Median (Standard (Standard deviation) deviation) 53.18 27.89 50.02 28.54

30.67 9.29 t-statistic for difference in means -13.98 Dollar effective spread (cents) 43.31 40.10 24.85 23.07 16.87 57.47 -5.23 Quoted percent spread (%) 3.34 2.70 2.58 1.56 1.34 0.95 -11.45 816.30 466.52 1190.73 969.66 381.50 1451.54 1.36 50.12 27.66 64.68 59.22 33.95 73.61 1.55 Share volume (100’s of shares) Number of trades 28 Source: http://www.doksinet TABLE 1 (contd) Panel C Two-day seasoned equity offering announcement returns for NYSE/AMEX and NASDAQ listed stocks are presented. Returns are based on closing transaction prices. Both raw returns and one factor market adjusted returns are presented The t statistics are based on White’s heteroscedasticity robust standard errors. Market adjusted stock returns are derived by subtracting the contemporaneous market return, measured by the equally weighted index of all stocks on the same exchange. TWO DAY ANNOUNCEMENT RAW RETURNS NASDAQ NYSE/AMEX t-statistic for difference in means Mean stock return (%)

t-statistic Median return (%) Sign test (p-value) % of Obs. = zero % of Obs. < zero No. of Observations -1.75 -5.51 -2.30 0.00 -2.70 -10.03 -2.00 0.00 9.2 63.3 316 10.1 55.4 253 -2.29 TWO DAY ANNOUNCEMENT MARKET ADJUSTED RETURNS NASDAQ Mean stock return (%) t-statistic Median return (%) Sign test (p-value) % of Obs. = zero % of Obs. < zero No. of Observations NYSE/AMEX -1.95 -6.25 -2.45 -2.78 -10.72 -2.00 0.00 0.6 71.2 316 0.00 1.6 62.7 253 t-statistic for difference in means -2.05 The sign test assesses whether the proportion of non-zero values that are positive is significantly different from 0.5 29 Source: http://www.doksinet TABLE 2 Common stock returns are reported over fifteen-minute trading intervals around seasoned equity offering announcements, separated into NASDAQ and NYSE/AMEX samples. The returns are based on midpoint of bid and ask price at the end of each 15 minute trading interval The p-values are based on bootstrap resampling under the null

hypothesis that the mean return in each interval is not different from zero. NASDAQ NYSE - AMEX Event Mean p-value Cumu- Median % of obs % of obs Sign Mean p-value Cumu- Median % of % of return lative return <0 =0 test* return lative return obs <0 obs =0 (%) return (%) (%) return (%) .06 .08 .00 5 88 .51 .03 .50 .00 15 65 -12 .04 .25 .00 6 88 1.00 -.03 .30 .00 19 63 -11 -.02 .28 .00 6 87 .64 -.01 .44 .00 16 66 -10 -.02 .34 .00 5 92 .42 -.04 .12 .00 17 69 -9 .03 .70 .00 7 87 1.00 -.02 .48 .00 14 68 -8 -.06 .00 .00 10 85 .03 -.04 .31 .00 17 68 -7 -.04 .02 .00 7 88 .64 .03 .60 .00 16 69 -6 .00 .52 .00 6 85 .24 .07 .09 .00 18 63 -5 -.01 .38 .00 8 86 .55 -.05 .13 .00 21 62 -4 -.07 .01 .00 9 84 .48 .01 .90 .00 17 61 -3 -.11 .00 .00 11 82 .11 -.03 .23 .00 27 56 -2 -.12 .06 .00 14 79 .00 -.01 .58 .00 22 56 -1 -1.53 .00 -1.53 -.94 63 29 .00 -.98 .00 -.98 -.43 58 29 0 -.30 .00 -1.83 .00 29 53 .01 -.53 .00 -1.51 .00 43 40 1 -.04 .12 -1.87 .00 15 68 .69 -.26 .00 -1.77 .00 33 46 2 -.06 .06

-1.93 .00 13 74 1.00 -.09 .06 -1.86 .00 28 51 3 -.08 .00 -2.01 .00 14 77 .05 -.18 .00 -2.04 .00 30 50 4 -.06 .03 -2.07 .00 10 81 .80 -.12 .00 -2.16 .00 28 51 5 .09 .00 -1.98 .00 8 81 .16 .01 .92 -2.15 .00 25 54 6 -.06 .00 -2.04 .00 8 86 .65 .00 .38 -2.15 .00 25 51 7 -.03 .19 -2.07 .00 9 83 1.00 -.04 .10 -2.19 .00 19 61 8 -.03 .09 -2.10 .00 10 82 .43 -.01 .56 -2.20 .00 23 57 9 -.03 .22 -2.13 .00 7 88 .52 .00 .62 -2.20 .00 25 56 10 .02 .80 -2.11 .00 5 86 .17 .00 .82 -2.20 .00 26 55 11 .01 .94 -2.10 .00 8 84 .78 .01 .81 -2.19 .00 18 62 12 * p-values from tests of whether the proportion of non-zero values that are positive is significantly different from 0.5 30 Sign test* .24 1.00 .75 .43 .43 .51 .91 .84 .36 .36 .04 .92 .00 .00 .02 .12 .02 .15 .58 .72 .92 .70 .22 .19 .68 Source: http://www.doksinet TABLE 3 Stock returns are classified by sign as negative, zero (with no trading or with trading) or positive. Conditional probabilities for signed returns in interval 1, given signed return

in interval 0, where returns are based on the midpoints of the best bid and ask quotes. Panel A: NASDAQ Panel B: NYSE/AMEX Return in Interval 1 Return in Interval 0 =0 =0 (No (With trades) trading) 9.8 20.3 15.3 31.5 >0 15.5 24.1 TOTAL Row freq. (No. of obs) 64.2 (203) <0 18.7 29.1 =0 (No trades) 2.9 22.5 7.9 62.5 1.9 15.0 0.0 0.0 12.7 (40) =0 (at least one trade) >0 6.3 41.7 3.8 25.0 4.8 31.3 0.3 2.1 15.2 (48) 2.5 32.0 1.0 12.0 2.2 28.0 2.22 28.0 7.9 (25) TOTAL: Col Freq (No. of obs) 30.4 (96) 22.5 (71) 29.1 (92) 18.0 (57) 100.0 (316) <0 <0 31.5 50.0 =0 (No trades) 8.6 13.7 =0 (No trades) 4.7 40.7 5.6 48.2 0.9 7.4 0.4 3.7 11.6 (27) =0 (at least one trade) >0 5.6 48.2 1.7 14.8 2.2 18.5 2.2 18.5 11.6 (27) 6.5 46.9 3.9 28.1 0.9 6.3 2.6 18.8 13.8 (32) 48.3 TOTAL: Col Freq (No. of obs) (112) 19.8 (46) 13.4 (31) 18.5 (43) 100.0 (232) Cell Percentage Row percentage Return in Interval 0 <0 Cell Percentage Row

percentage Return in Interval 1 =0 (With trading) 9.5 15.1 >0 13.4 21.2 TOTAL Row freq. (No. of obs) 62.9 (146) Source: http://www.doksinet TABLE 4 Common stock returns are reported for fifteen-minute trading intervals around seasoned equity offering announcements occurring over 1990-1992. The sample is separated by exchange listing (NASDAQ or NYSE/AMEX) and time of announcement (daytime or overnight). The returns are based on midpoints of the best bid and ask quotes. The p-values are based on bootstrap resampling under the null hypothesis of a zero mean return Panel A: DAYTIME ANNOUNCEMENTS Mean return (%) -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 .06 .03 -.04 .00 .06 -.07 -.01 .00 .01 -.07 -.16 -.14 -1.49 -.52 -.01 -.06 .01 .00 .08 -.05 -.04 -.04 .00 .02 .01 Mean return p-value .10 .65 .20 .74 .20 .01 .30 .36 .75 .02 .00 .09 .00 .00 .51 .12 .72 .77 .05 .02 .21 .16 .80 .92 .82 Cumulative return NASDAQ Median % of return obs. <0 (%) -1.49

-2.01 -2.02 -2.08 -2.07 -2.07 -1.99 -2.04 -2.08 -2.12 -2.12 -2.10 -2.09 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 -.84 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 5 6 8 3 6 11 5 6 8 10 14 15 62 37 14 12 13 11 6 9 9 10 7 6 9 % of obs. =0 89 88 85 93 86 85 88 86 85 83 79 77 31 45 66 73 75 78 84 84 85 81 88 85 83 Sign test* .54 .85 1.00 1.00 .36 .05 .42 .71 .72 .62 .05 .03 .00 .00 .15 .69 .68 1.00 .16 .86 .38 .75 .85 .47 1.00 NYSE - AMEX Mean Mean Cumu- Median % of % of return return lative return obs. <0 obs =0 (%) p-value return (%) .04 -.03 -.03 -.01 .00 -.02 -.04 .09 -.03 .02 -.05 -.02 -1.01 -.70 -.25 -.09 -.21 -.16 .09 .02 -.05 .00 -.02 .04 .01 .62 .32 .28 .56 .96 .48 .26 .04 .35 .92 .25 .40 .00 .00 .00 .10 .00 .00 .08 .80 .17 .69 .50 .49 .88 -1.01 -1.71 -1.96 -2.05 -2.26 -2.42 -2.33 -2.31 -2.36 -2.36 -2.38 -2.34 -2.33 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 -.27 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 16 21 18 16 13 16 20 20 19 17 27 25 55 49

31 27 32 28 25 25 18 24 26 27 18 65 60 66 67 66 68 67 58 62 61 55 52 36 38 50 54 48 51 49 49 60 55 52 49 65 Sign test* .58 .79 .78 1.00 .15 1.00 .14 .80 1.00 .35 .14 .81 .00 .00 .07 .21 .05 .34 .91 1.00 .60 .80 .47 .73 .89 Source: http://www.doksinet TABLE 4 (contd.) Panel B: OVERNIGHT ANNOUNCEMENTS Event -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 * NASDAQ Mean Mean Cumu- Median % of % of return return lative return obs. <0 obs =0 (%) p-value return (%) .04 .43 .00 5 88 .07 .22 .00 6 89 .02 .64 .00 3 91 -.05 .22 .00 7 91 -.04 .25 .00 8 88 -.03 .18 .00 9 86 -.09 .05 .00 11 86 .01 .93 .00 5 84 -.04 .18 .00 7 87 -.06 .13 .00 7 87 -.03 .17 .00 5 88 -.08 .04 .00 12 84 -1.60 .00 -1.60 -1.17 64 25 .13 .08 -1.47 .00 14 70 -.11 .16 -1.58 .00 18 71 -.06 .28 -1.64 .00 14 76 -.26 .00 -1.90 .00 17 80 -.18 .01 -2.08 .00 8 85 .11 .03 -1.97 .00 12 73 -.07 .05 -2.04 .00 6 91 -.01 .42 -2.05 .00 8 79 -.03 .31 -2.08 .00 10 83 -.07 .04 -2.15 .00 7 88 .03 .63 -2.12 .00 4

88 .01 .68 -2.11 .00 6 86 Sign test* 1.00 .77 .34 .11 .27 .45 .04 .24 .79 .79 1.00 .10 .00 .73 .22 .56 .00 .80 .71 .75 .40 .48 .58 .27 .61 NYSE - AMEX Mean Mean Cumu- Median % of % of return return lative return obs. <0 obs =0 (%) p-value return (%) .02 .64 .00 14 66 -.02 .60 .00 15 67 .02 .75 .00 14 66 -.08 .08 .00 18 72 -.04 .27 .00 15 72 -.06 .45 .00 19 68 .12 .04 .00 10 72 .05 .51 .00 15 71 -.07 .12 .00 23 63 .00 .72 .00 18 62 .00 .72 .00 26 56 .02 1.00 .00 17 61 -.92 .00 -.92 -.51 62 20 -.29 .00 -1.21 .00 35 43 -.27 .00 -1.48 .00 36 41 -.09 .34 -1.57 .00 30 47 -.15 .03 -1.72 .00 28 53 -.07 .17 -1.79 .00 28 50 -.09 .13 -1.88 .00 24 61 -.01 .32 -1.89 .00 26 54 -.03 .43 -1.92 .00 20 62 -.02 .70 -1.94 .00 21 60 .02 .89 -1.92 .00 23 61 -.06 .10 -1.98 .00 24 62 .01 .93 -1.97 .00 17 59 P-values from tests of whether the proportion of no-zero values that are positive is significantly different from 0.5???? Sign test* .32 .87 .32 .15 .72 .31 .15 1.00 .21 .88 .19 .54 .00 .13 .16 .43

.26 .34 .22 .48 .76 .88 .36 .12 .37 Source: http://www.doksinet TABLE 5 For overnight announcements, open-to-close and close-to-open stock returns for event days 0 and 1 are reported. The sample is classified by exchange listing (NASDAQ or NYSE/AMEX). The returns are based on the midpoints of the best bid and ask quotes at the open and close of trading For NASDAQ firms, the table also presents the returns based on the first quote on day 0 (First(0)), which is generally posted prior to the official opening quote for the day. The t-statistics are based on White’s heteroscedasticity robust standard errors. OVERNIGHT ANNOUNCEMENTS Return From: To: Mean (%) t-statistic Median (%) Sign test (p-value) No. of Obs > zero No. of Obs not equal to zero NASDAQ Close(-1) First(0) Open(0) Close(0) Open(+1) First(0) Open(0) Close(0) Open(+1) Close(+1) -0.17 -0.97 -0.56 -0.02 0.38 -1.83 -6.92 -1.54 -0.30 1.22 0.00 -0.40 0.00 0.00 0.00 0.09 0.00 0.24 1.00 0.06 3 3 38 19 49 13 59 88 38 80

Close(-1) Open(0) -0.64 -2.82 -0.44 0.00 25 87 NYSE-AMEX Open(0) Close(0) Open(+1) Close(0) Open(+1) Close(+1) -1.27 0.05 -0.45 -4.36 0.27 -2.00 -0.87 0.00 -0.34 0.00 1.00 0.02 31 45 36 101 89 96 Source: http://www.doksinet TABLE 6 This table presents the abnormal frequency of 1000 share trades for 15 minute trading intervals around seasoned equity offering announcements over 1990-1992. The sample is separated by exchange listing (NASDAQ or NYSE/AMEX) and time of announcement (daytime or overnight). The p-values are based on bootstrap resampling under the null hypothesis that the mean abnormal frequency in each interval is equal to zero. DAYTIME NASDAQ Interval Mean OVERNIGHT NYSE/AMEX p-value Mean p-value NASDAQ Mean NYSE/AMEX p-value Mean p-value -12 0.12 0.59 0.36 0.36 0.16 0.57 0.13 0.70 -11 0.50 0.00 -0.02 0.94 -0.21 0.36 -0.24 0.73 -10 -0.07 0.73 0.02 0.88 -0.18 0.56 -0.31 0.46 -9 -0.23 0.27 0.89 0.05 -0.59 0.01 -0.43 0.24 -8

0.30 0.12 0.60 0.06 -0.33 0.25 -0.28 0.55 -7 0.28 0.14 -0.19 0.46 0.29 0.26 0.46 0.25 -6 0.31 0.18 -0.17 0.66 -0.39 0.16 -0.41 0.16 -5 0.23 0.32 -0.25 0.53 0.12 0.59 -0.28 0.56 -4 0.71 0.00 -0.14 0.85 -0.08 0.78 0.34 0.24 -3 0.39 0.04 0.24 0.47 -0.09 0.76 0.03 0.84 -2 0.54 0.00 -0.17 0.63 0.28 0.24 -0.29 0.34 -1 0.38 0.07 0.48 0.22 0.31 0.16 0.76 0.03 0 2.04 0.00 0.56 0.07 0.78 0.00 0.41 0.16 1 1.25 0.00 0.93 0.02 0.45 0.02 0.36 0.18 2 0.64 0.00 0.49 0.13 0.40 0.10 0.00 0.94 3 0.89 0.00 0.50 0.12 0.40 0.11 0.37 0.30 4 0.18 0.31 0.08 0.69 0.40 0.20 0.56 0.20 5 0.09 0.59 0.08 0.89 -0.01 0.96 0.17 0.57 6 0.21 0.36 0.84 0.04 0.57 0.01 0.75 0.03 7 0.29 0.10 0.37 0.26 0.13 0.51 0.39 0.34 8 0.34 0.07 0.58 0.10 0.45 0.08 -0.10 0.87 9 0.25 0.32 0.30 0.37 -0.04 0.98 -0.05 0.93 10 0.12 0.46 0.31 0.42 0.26 0.36 0.02 0.74 11 0.25

0.21 0.09 0.62 0.32 0.16 0.79 0.04 12 0.28 0.18 0.10 0.74 -0.17 0.42 0.20 0.43 Abnormal frequency of 1000 share trades is measured by the frequency of 1000 share trade in event interval i minus the average frequency of 1000 share trades in interval i over the benchmark period all standardized by the daily average frequency of 1000 share trades in the benchmark period. Source: http://www.doksinet TABLE 7 Common stock returns are reported for 15 minute intervals (event intervals 0, 0 - 1 and 0 - 3) following seasoned equity offering announcements made by NASDAQ listed firms over 1990-1992. The stocks are classified into quintiles based on number of market makers in the stock at the time of the announcement. Daytime announcements Quintile 1 2 3 4 5 Return No. of for: dealers Mean Median % obs = 0 % obs > 0 N 6.5 7.0 Mean Median % obs = 0 % obs > 0 N 10.1 10.0 Mean Median % obs = 0 % obs > 0 N 13.3 13.0 Mean Median % obs = 0 % obs > 0 N 17.2

17.0 Mean Median % obs = 0 % obs > 0 N 25.4 23.5 Overnight announcements Interval Intervals Intervals 0 0 and 1 0 to 3 -0.84 0.00 50.0 5.3 -1.51 0.00 42.1 10.5 -1.73 -0.65 36.8 7.9 38 6.2 7.0 Interval Intervals Intervals 0 0 and 1 0 to 3 -1.03 0.00 52.9 0.0 -1.15 -1.05 47.1 0.0 -1.24 -1.05 47.1 0.0 -1.88 -1.56 34.8 0.0 -1.83 -1.56 34.8 0.0 -2.02 -1.48 21.7 8.7 -2.12 -2.15 16.7 11.1 -1.69 -2.15 11.1 16.7 -2.07 -1.80 5.6 27.8 -2.10 -2.19 0.0 26.7 -1.78 -1.87 0.0 26.7 -1.75 -1.90 0.0 33.3 -1.33 -0.79 19.2 19.2 -1.11 -0.79 19.5 26.9 -1.25 -1.01 3.8 23.1 17 -1.45 -0.94 29.7 5.4 -1.97 -1.75 16.2 8.1 -1.84 -2.04 5.4 18.9 37 10.0 10.0 23 -1.71 -1.41 18.4 10.2 -2.26 -2.00 8.2 12.2 -2.19 -2.56 2.0 16.3 49 13.2 13.0 18 -1.77 -1.28 33.3 2.8 -2.86 -2.66 11.1 5.6 -2.91 -2.62 5.6 16.7 36 36 No. of dealers 17.2 17.0 15 -2.30 -2.44 11.1 8.3 -2.30 -1.93 2.8 8.3 -2.12 -2.04 0.0 11.1 26.4 26.5 26 Source: http://www.doksinet TABLE 8 Abnormal quoted depth

of the best bid and ask quotes are reported at 15 minute trading intervals surrounding seasoned equity offering announcements of NYSE/AMEX stocks made in 1990-1992. Event -12 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 DAYTIME Depth at ask Depth at bid Mean p-value Mean p-value .05 .59 .03 .76 -.07 .59 .07 .46 -.11 .29 .18 .19 .00 .98 .27 .05 .06 .62 .33 .02 .08 .54 .17 .16 .11 .36 .26 .08 -.10 .37 .09 .41 .01 .92 .09 .44 -.04 .84 .40 .01 -.06 .60 .30 .01 -.07 .56 .27 .02 -.09 .36 -.23 .02 -.01 .94 -.19 .05 .13 .26 -.23 .03 .15 .18 -.23 .02 .11 .29 -.13 .24 -.04 .66 -.10 .38 -.05 .61 -.04 .76 -.14 .16 -.07 .66 -.10 .45 -.15 .15 -.09 .44 -.09 .40 -.02 .95 -.07 .54 -.01 .98 .02 .86 -.01 .98 .01 .95 OVERNIGHT Depth at ask Depth at bid Mean p-value Mean p-value -.15 .21 .00 1.00 -.23 .04 -.10 .47 -.14 .30 -.03 .97 -.02 .94 .06 .50 -.02 .76 .12 .33 -.12 .37 .05 .66 .00 .93 .07 .64 -.02 .88 .01 .83 -.08 .69 -.09 .58 -.13 .41 -.16 .23 .09 .50 -.08 .63 .14 .29 -.05 .74

-.26 .00 -.38 .00 -.30 .00 -.38 .00 -.27 .00 -.43 .00 -.23 .02 -.39 .00 -.19 .04 -.41 .00 -.37 .00 -.33 .00 -.41 .00 -.28 .02 -.44 .00 -.21 .07 -.44 .00 -.29 .01 -.44 .00 -.16 .20 -.32 .01 -.20 .12 -.30 .02 -.18 .15 -.23 .05 -.12 .34 Abnormal depth for either bid or ask is measured as Abnormal depth i = DEPi - ADEPi DDEP where DEP i is the time-weighted depth at either bid or ask in interval i, ADEP i is the benchmark period average for quoted depth in the same fifteen-minute interval corresponding to interval i and DDEP is the daily average of timeweighted depth in the benchmark period. The p-values are based on bootstrap resampling under the null hypothesis that the mean abnormal bid/ask depth in each interval is equal to zero. Source: http://www.doksinet TABLE 9 Percentage of trades occurring exactly at the bid, relative to all trades occurring in 15 minute trading intervals surrounding seasoned equity offering announcements over 1990-1992, classified by exchange listing

(NASDAQ and NYSE/AMEX) and time of announcement (daytime and overnight). DAYTIME NASDAQ Interval Mean OVERNIGHT NYSE/AMEX p-value Mean p-value NASDAQ Mean NYSE/AMEX p-value Mean p-value -12 0.06 0.52 0.58 0.00 0.03 0.90 -0.23 0.35 -11 0.35 0.00 0.96 0.00 0.11 0.55 0.39 0.06 -10 0.19 0.17 0.37 0.01 0.30 0.11 0.71 0.00 -9 0.32 0.03 0.68 0.00 0.02 0.87 0.29 0.15 -8 0.19 0.10 0.19 0.26 -0.19 0.24 0.12 0.54 -7 0.17 0.22 0.12 0.46 0.11 0.51 0.31 0.08 -6 0.26 0.06 0.07 0.62 0.01 1.00 -0.42 0.01 -5 -0.08 0.47 0.16 0.27 -0.28 0.10 -0.06 0.68 -4 0.14 0.31 0.10 0.43 0.09 0.43 -0.07 0.82 -3 0.10 0.43 0.12 0.46 -0.03 0.90 0.10 0.56 -2 0.02 0.85 0.90 0.00 -0.08 0.54 0.27 0.10 -1 0.09 0.49 -0.07 0.68 -0.27 0.02 -0.09 0.74 0 0.11 0.30 0.59 0.00 -0.30 0.02 -0.25 0.21 1 0.02 0.99 0.47 0.00 0.05 0.69 -0.14 0.47 2 0.17 0.13 0.55 0.00 0.02 0.76 -0.10 0.64 3

0.24 0.07 0.28 0.04 0.08 0.73 0.22 0.19 4 0.29 0.01 0.32 0.04 0.01 0.88 0.17 0.27 5 0.30 0.00 0.74 0.00 0.29 0.06 0.02 0.88 6 0.23 0.09 0.21 0.22 -0.01 0.96 0.00 0.98 7 -0.05 0.71 0.16 0.32 -0.04 0.80 -0.05 0.78 8 0.25 0.06 0.43 0.00 -0.06 0.72 0.15 0.42 9 0.27 0.03 -0.14 0.37 0.08 0.57 -0.02 0.83 10 0.09 0.34 0.54 0.00 0.11 0.48 0.22 0.14 11 0.14 0.21 0.28 0.08 -0.06 0.78 -0.03 0.96 12 0.19 0.16 0.16 0.36 -0.02 1.00 0.30 0.12 Abnormal quotes at bid = (% of trades at bid - average % of trades at bid for corresponding interval in benchmark period)/(average % of trades at bid in corresponding interval in benchmark period) Source: http://www.doksinet TABLE 10 OLS regression estimates of a model explaining common stock returns over one or more 15 minute trading intervals following seasoned equity offering announcements over the 1990-1992 period. The sample is separated by time of announcement

(daytime or overnight) Panel A: Daytime announcements Dependent variable: Interval 0 return INTERCEPT NYAM OFFSIZE*NASD OFFSIZE*NYAM 1 -0.064 (-2.37) 0.060 (1.78) 0.012 (0.64) -0.028 (-3.10) ABNSPRD*NASD+ ABNSPRD*NYAM+ TRNOVER*NASD+ TRNOVER*NYAM+ MSHRTRD*NASD+ MSHRTRD*NYAM+ BIDDEPTH*NYAM+ -0.497 (-2.03) 0.045 (0.23) -0.102 (-1.84) -0.029 (-0.51) 0.230 (2.24) 2 -0.048 (-2.57) 0.044 (1.59) -0.001 (-0.04) -0.027 (-3.36) 0.007 (0.01) -0.974 (-5.01) -0.341 (-1.94) 0.018 (0.09) -0.180 (-3.12) -0.060 (-1.11) NMKR+ Adj-R-square No. of Observations 0.32 193 0.44 274 3 -0.051 (-1.66) 0.048 (1.29) 0.005 (0.27) -0.028 (-3.10) -0.436 (-1.61) 0.045 (0.23) -0.122 (-1.95) -0.029 (-0.51) 0.230 (2.24) -0.028 (-1.14) 0.33 189 Cumulative return for intervals 0 and 1 1 2 3 -0.046 -0.028 -0.049 (-1.73) (-1.50) (-1.64) 0.014 -0.015 0.018 (0.37) (-0.46) (0.43) -0.022 -0.020 -0.022 (-0.92) (-1.40) (-0.86) -0.036 -0.038 -0.036 (-2.87) (-2.93) (-2.87) 0.006 (0.01) -1.102 (-2.70) -0.337 -0.122 -0.362

(-1.37) (-0.70) (-1.38) -0.175 -0.291 -0.175 (-0.67) (-1.17) (-0.67) -0.113 -0.118 -0.118 (-1.83) (-2.31) (-1.90) -0.014 -0.027 -0.014 (-0.42) (-0.69) (-0.42) 0.198 0.198 (2.67) (2.67) 0.009 (0.32) 0.41 0.47 0.40 212 297 208 Cumulative return for intervals 0 to 3 1 3 4 -0.052 -0.019 -0.065 (-1.08) (-0.65) (-1.13) -0.009 -0.050 0.004 (-0.15) (-1.23) (0.07) -0.009 -0.021 -0.005 (-0.23) (-0.85) (-0.10) -0.054 -0.061 -0.054 (-3.08) (-3.33) (-3.08) -1.438 (-1.21) -1.685 (-2.76) -0.325 -0.027 -0.405 (-0.76) (-0.10) (-0.86) -0.462 -0.568 -0.462 (-1.73) (-2.18) (-1.73) 0.046 -0.113 0.084 (0.42) (-1.61) (0.73) 0.000 0.008 0.000 (0.01) (0.26) (0.01) 0.130 0.130 (2.55) (2.55) 0.023 (0.46) 0.36 0.42 0.35 170 216 167 Source: http://www.doksinet TABLE 10 (contd) Panel B: Overnight announcements Dependent variable: INTERCEPT NYAM OFFSIZE*NASD OFFSIZE*NYAM Interval 0 return 1 -0.036 (-0.88) 0.012 (0.26) -0.033 (-1.36) -0.030 (-1.99) ABNSPRD*NASD+ ABNSPRD*NYAM+ TRNOVER*NASD+ TRNOVER*NYAM+

MSHRTRD*NASD+ MSHRTRD*NYAM+ BIDDEPTH*NYAM+ -0.281 (-0.73) -0.164 (-0.88) -0.138 (-0.94) 0.002 (0.05) 0.519 (2.54) 2 -0.029 (-1.27) -0.001 (-0.05) -0.024 (-1.40) -0.025 (-1.52) -1.389 (-1.78) -1.139 (-2.37) -0.178 (-0.85) -0.214 (-1.13) -0.139 (-1.12) -0.018 (-0.45) NMKR+ Adj-R-square No. of Observations + (coefficient*100) 0.41 123 0.41 164 3 -0.129 (-2.97) 0.105 (2.19) 0.000 (0.00) -0.030 (-1.99) -0.950 (-2.49) -0.164 (-0.88) -0.033 (-0.26) 0.002 (0.05) 0.519 (2.54) 0.109 (3.41) 0.43 119 Cumulative return for intervals 0 and 1 1 2 3 -0.025 -0.003 -0.117 (-0.85) (-0.17) (-3.04) -0.005 -0.033 0.087 (-0.14) (-1.13) (2.02) 0.008 0.001 0.036 (0.22) (0.05) (0.90) -0.017 -0.012 -0.017 (-2.35) (-1.36) (-2.35) -0.867 (-1.09) -1.930 (-2.69) -0.091 0.115 -0.806 (-0.31) (0.57) (-2.24) -0.193 -0.243 -0.193 (-1.08) (-1.24) (-1.08) -0.198 -0.148 -0.108 (-2.21) (-1.98) (-1.35) -0.050 -0.046 -0.050 (-0.70) (-0.60) (-0.70) 0.375 0.375 (3.38) (3.38) 0.084 (2.17) 0.34 0.33 0.34 129 173 124

Cumulative return for intervals 0 to 3 1 3 4 -0.016 -0.010 -0.100 (-0.50) (-0.43) (-2.23) -0.058 -0.078 0.026 (-1.35) (-1.93) (0.49) 0.011 0.006 0.037 (0.27) (0.23) (0.86) 0.007 0.017 0.007 (0.42) (0.89) (0.42) -1.281 (-1.18) -1.713 (-1.72) 0.039 0.085 -0.623 (0.12) (0.41) (-1.52) -0.547 -0.645 -0.547 (-2.03) (-2.12) (-2.03) 0.023 -0.040 0.067 (0.22) (-0.42) (0.62) -0.062 -0.056 -0.062 (-1.10) (-0.94) (-1.10) 0.338 0.338 (2.89) (2.89) 0.072 (1.53) 0.31 0.29 0.31 135 182 130 Source: http://www.doksinet Table 10 (contd) The t-statistics are based on White’s heteroscedasticity robust standard errors and are presented within parenthesis. Variable Definitions: NYAM: 1 for NYSE/AMEX stocks, 0 otherwise. NASD: 1 for NASDAQ stocks, 0 otherwise. TRNOVER: logarithm of the daily average number of shares traded in the benchmark period divided by the pre-offering shares outstanding, measured at the close of the quarter immediately before the offering. OFFSIZE: ratio of the number of shares

offered to the pre-offering shares outstanding, measured at the close of the quarter immediately before the offering. MSHRTRD: Abnormal frequency of 1000 share trades in event interval i, measured by the frequency of 1000 share trades in event interval i (number of 1000 share trades relative to all trades in event interval i) minus the frequency of 1000 share trades in interval i of the benchmark period, relative to the daily average frequency of 1000 share trades in the benchmark period. ABNSPRD: Abnormal percentage bid-ask spread is measured by the time weighted average percentage bid-ask spread in interval i minus the average time weighted percentage bid-ask spread in interval i for the benchmark period, all divided by the daily average time weighted percentage spread in the benchmark period. BIDDEPTH: Abnormal depth at the bid is measured by the time weighted average depth at the bid in the ith interval minus the time weighted average depth at the bid in the ith interval of the

benchmark period, all divided by the daily average time weighted depth at the bid in the benchmark period. NMKR: Average number of NASDAQ dealers in the stock over the benchmark period. The benchmark period is defined as day -30 to day -5, where day 0 is defined as the SEO announcement day, if the announcement occurs before 4pm. If the announcement occurs after 4pm, then day 0 is defined as the following trading day. Time weights are calculated as the proportion of the 15 minute trading interval i for which the jth quote is outstanding. Source: http://www.doksinet TABLE 11 OLS regression estimates of a model explaining common stock returns of industrial firms over one or more 15 minute trading intervals (intervals 0, 0 – 1, 0 – 3) following seasoned equity offering announcements over the 1990-1992 period. Each NYSE/AMEX stock is matched with a NASDAQ stock having the closest two-day announcement return. Announcements are separated by time of day the news is released (daytime or

overnight) Panel A: Daytime announcements Dependent variable: Interval 0 return INTERCEPT NYAM OFFSIZE*NASD OFFSIZE*NYAM 1 -0.060 (-1.82) 0.074 (1.70) -0.022 (-1.25) -0.023 (-2.22) ABNSPRD*NASD+ ABNSPRD*NYAM+ TRNOVER*NASD+ TRNOVER*NYAM+ MSHRTRD*NASD+ MSHRTRD*NYAM+ BIDDEPTH*NYAM+ -0.521 (-1.75) 0.243 (0.90) -0.216 (-1.42) 0.015 (0.20) 0.293 (1.56) 2 -0.058 (-2.27) 0.068 (1.85) -0.016 (-1.12) -0.020 (-2.23) -0.065 (-0.08) -1.009 (-5.92) -0.456 (-1.94) 0.195 (0.76) -0.261 (-1.96) -0.028 (-0.41) NMKR+ Adj-R-square No. of Observations 0.43 110 0.53 154 3 -0.029 (-0.75) 0.042 (0.89) -0.040 (-2.38) -0.023 (-2.22) -0.376 (-1.12) 0.243 (0.90) -0.226 (-1.61) 0.015 (0.20) 0.293 (1.56) -0.071 (-1.62) 0.44 109 Cumulative return for intervals 0 and 1 1 2 3 -0.037 -0.034 -0.045 (-1.01) (-1.35) (-0.91) 0.032 0.026 0.041 (0.60) (0.57) (0.65) -0.056 -0.045 -0.050 (-1.95) (-2.10) (-1.36) -0.026 -0.027 -0.026 (-1.65) (-1.58) (-1.65) 0.118 (0.11) -1.094 (-2.59) -0.229 -0.200 -0.260 (-0.69)

(-0.85) (-0.66) 0.143 0.098 0.143 (0.36) (0.26) (0.36) 0.023 -0.073 0.025 (0.15) (-0.79) (0.15) 0.019 0.015 0.019 (0.25) (0.20) (0.25) 0.283 0.283 (1.47) (1.47) 0.022 (0.37) 0.49 0.51 0.49 120 166 119 Cumulative return for intervals 0 to 3 1 3 4 -0.083 -0.041 -0.052 (-1.06) (-0.95) (-0.55) 0.045 0.003 0.014 (0.51) (0.06) (0.14) -0.021 -0.036 -0.028 (-0.52) (-1.04) (-0.67) -0.049 -0.056 -0.049 (-2.15) (-2.28) (-2.15) -0.235 (-0.18) -1.621 (-2.25) -0.536 -0.269 -0.330 (-0.74) (-0.67) (-0.42) -0.192 -0.215 -0.192 (-0.47) (-0.53) (-0.47) 0.259 -0.187 0.301 (0.78) (-1.52) (0.90) 0.058 0.050 0.058 (0.94) (0.82) (0.94) 0.164 0.164 (1.34) (1.34) -0.063 (-0.53) 0.43 0.46 0.42 105 131 105 Source: http://www.doksinet TABLE 11 (contd) Panel B: Overnight announcements Dependent variable: INTERCEPT NYAM OFFSIZE*NASD OFFSIZE*NYAM Interval 0 return 1 -0.030 (-0.69) 0.026 (0.54) -0.031 (-0.93) -0.021 (-1.09) ABNSPRD*NASD+ ABNSPRD*NYAM+ TRNOVER*NASD+ TRNOVER*NYAM+ MSHRTRD*NASD+ MSHRTRD*NYAM+

BIDDEPTH*NYAM+ -0.214 (-0.52) 0.055 (0.28) -0.126 (-0.66) 0.003 (0.07) 0.611 (2.85) 2 -0.038 (-1.47) 0.025 (0.72) -0.013 (-0.69) -0.016 (-0.77) -2.161 (-2.16) -1.218 (-2.30) -0.217 (-0.85) -0.027 (-0.13) -0.113 (-0.73) -0.020 (-0.42) NMKR+ Adj-R-square No. of Observations + (coefficient*100) 0.38 99 0.46 130 3 -0.122 (-2.77) 0.118 (2.41) 0.003 (0.10) -0.021 (-1.09) -0.863 (-2.18) 0.055 (0.28) 0.006 (0.04) 0.003 (0.07) 0.611 (2.85) 0.111 (3.37) 0.42 96 Cumulative return for intervals 0 and 1 1 2 3 -0.014 -0.013 -0.106 (-0.43) (-0.55) (-2.42) 0.005 -0.005 0.097 (0.14) (-0.16) (2.02) -0.010 -0.012 0.007 (-0.34) (-0.62) (0.25) -0.011 -0.006 -0.011 (-1.24) (-0.63) (-1.24) -0.725 (-0.84) -1.930 (-2.55) 0.005 0.019 -0.754 (0.02) (0.09) (-1.88) 0.043 -0.044 0.043 (0.22) (-0.21) (0.22) -0.227 -0.177 -0.107 (-3.10) (-1.96) (-1.50) -0.024 -0.033 -0.024 (-0.32) (-0.40) (-0.32) 0.456 0.456 (3.48) (3.48) 0.075 (1.90) 0.41 0.42 0.41 103 137 99 Cumulative return for intervals 0 to 3 1 3 4

-0.014 -0.017 -0.114 (-0.39) (-0.64) (-2.08) -0.035 -0.052 0.065 (-0.74) (-1.14) (1.02) -0.009 -0.002 0.011 (-0.34) (-0.12) (0.41) 0.014 0.020 0.014 (0.75) (0.96) (0.75) -1.210 (-0.95) -1.582 (-1.53) 0.027 0.029 -0.811 (0.08) (0.11) (-1.49) -0.268 -0.429 -0.268 (-0.90) (-1.25) (-0.90) 0.007 -0.059 0.067 (0.06) (-0.46) (0.56) -0.034 -0.032 -0.034 (-0.50) (-0.42) (-0.50) 0.378 0.378 (3.07) (3.07) 0.074 (1.52) 0.36 0.34 0.36 107 144 103 Source: http://www.doksinet TABLE 11 (contd) The t-statistics are based on White’s heteroscedasticity robust standard errors and are presented within parenthesis. Variable Definitions: NYAM: 1 for NYSE/AMEX stocks, 0 otherwise. NASD: 1 for NASDAQ stocks, 0 otherwise. TRNOVER: logarithm of the daily average number of shares traded in the benchmark period divided by the pre-offering shares outstanding, measured at the close of the quarter immediately before the offering. OFFSIZE: ratio of the number of shares offered to the pre-offering shares

outstanding, measured at the close of the quarter immediately before the offering. MSHRTRD: Abnormal frequency of 1000 share trades in event interval i, measured by the frequency of 1000 share trades in event interval i (number of 1000 share trades relative to all trades in event interval i) minus the frequency of 1000 share trades in interval i of the benchmark period, relative to the daily average frequency of 1000 share trades in the benchmark period. ABNSPRD: Abnormal percentage bid-ask spread is measured by the time weighted average percentage bid-ask spread in interval i minus the average time weighted percentage bid-ask spread in interval i for the benchmark period, all divided by the daily average time weighted percentage spread in the benchmark period. BIDDEPTH: Abnormal depth at the bid is measured by the time weighted average depth at the bid in the ith interval minus the time weighted average depth at the bid in the ith interval of the benchmark period, all divided by the

daily average time weighted depth at the bid in the benchmark period. NMKR: Average number of NASDAQ dealers in the stock over the benchmark period. The benchmark period is defined as day -30 to day -5, where day 0 is defined as the SEO announcement day, if the announcement occurs before 4pm. If the announcement occurs after 4pm, then day 0 is defined as the following trading day. Time weights are calculated as the proportion of the 15 minute trading interval i for which the jth quote is outstanding. Source: http://www.doksinet FIGURE 1 Intraday distribution of announcements of equity offerings on NASDAQ number of observations 35 30 25 20 15 10 5 17.301800 16.001630 14.301500 13.001330 11.301200 10.001030 8.30-900 7.00-730 0 Announcem ent tim e Intraday distribution of announcements of equity offerings on NYSE/AMEX 30 25 20 15 10 5 Announcem ent tim e 17.301800 16.001630 14.301500 13.001330 11.301200 10.001030 8.30-900 0 7.00-730 number of observations 35

Source: http://www.doksinet Figure 2 Abnormal number of trades around daytime equity offering announcements Abnormal number of trades around overnight equity offering announcements 0.25 0.18 0.16 abnormal no. of trades 0.06 0.04 0.02 12 9 6 3 -0.02 -0.05 0 0 -3 12 9 6 3 0 -3 -6 -9 0 0.1 0.08 -6 0.05 0.12 -9 0.1 0.14 -12 0.15 -12 abnormal no. of trades 0.2 event interval ev ent interval NYSE/AMEX FIRMS NASDAQ FIRMS This figure presents the mean number of trades in the fifteen-minute event intervals around equity offering announcements, where event interval 0 is defined as the interval containing the announcement. The abnormal number of trades is defined as: Abnormal number of trades = (Ki - ATRDi)/DTRD where Ki is the number of trades in event interval i, ATRDi is the benchmark period average for number of trades in the fifteen-minute interval corresponding to interval i and DTRD is the benchmark period daily average number of trades. Source:

http://www.doksinet Figure 3 Abnormal spreads around overnight equity offering announcements 0.25 0.2 0.2 0.15 abnormal spreads abnormal spreads Abnormal spreads around daytime equity offering announcements 0.15 0.1 0.05 0 -0.05 -12 -10 -8 -6 -4 -2 0 2 4 6 8 10 12 0.1 0.05 0 -12 -10 -8 -6 -4 -2 0 2 4 6 8 10 -0.05 -0.1 -0.1 event interval event interval NYSE/AMEX NASDAQ This figure presents the mean abnormal spreads in the fifteen-minute event intervals around equity offering announcements. The interval containing the equity offering announcement is defined as event interval 0. Abnormal spread is defined as: Ji Abnormal spread i = [ ∑ (SPRD ij − ASPRD i )Tij ] / DSPRD j=1 where SPRDij is the percentage bid-ask spread for the jth quote in interval i and Tij is the proportion of interval i for which the jth quote is outstanding, ASPRDi is the benchmark period average for percentage spreads in the fifteen-minute interval corresponding to interval

i and DSPRD is the benchmark period daily average percentage spread. 12