Raymond M. Brooks Oregon State University

Ajay Patel Wake Forest University

Tie Su University of Miami

How the Equity Market Responds to Unanticipated Events*

I. Introduction

One of the major premises of efficient market theory is that the market quickly impounds any publicly available information, including macroeconomic in- formation, that might be used to predict stock prices. It is only new-and especially new and unpredicta- ble-information that moves prices, and yet many studies examine only announcements that have a pre- dictable component. Researchers typically select a proxy for the anticipated portion of the news an- nouncement and then test the market's reaction to the unanticipated portion of the announcement.

However, the process of separating the anticipated and unanticipated portions of news announcements is critical to conclusions that can be drawn about price changes, the speed of adjustment, and trading activ- ities. We avoid this separation problem by looking at fully unanticipated events.

* We would like to thank Laura Starks and an anonymous referee for helpful suggestions, Ron Howren for computer support, Eric Schuster and Bob Hebert for gathering the sample, and Sandra Sizer Moore for editorial assistance. Ajay Patel thanks the Research Fel- lowship Program at Wake Forest University's Babcock Graduate School of Management for partial support of this project. Tie Su acknowledges financial support from the Research Council, Uni- versity of Miami. The usual disclaimer applies.

(Journal of Business, 2003, vol. 76, no. 1) ? 2003 by The University of Chicago. All rights reserved. 0021-9398/2003/7601-0005$ 10.00

We examine the market reaction of prices, vol- ume, spreads, and trading location when firms ex- perience events that are totally unanticipated by the equity market in terms of both timing and content. We find that the response time is longer than previous studies have reported. Selling pressure, wider spreads, and higher volume re- main significant for over an hour. We also find an immediate price reaction for overnight events; however, the market takes longer to react to events that occur when it is open. These findings may shed light on the efficacy of trading halts.

109

Journal of Business

We collect a set of unanticipated events and then examine the equity market's reaction to these events. The events we examine range from the Exxon Valdez running aground on Prince William Sound to plane crashes (United Airlines [UAL] and USAir) and plant explosions (Quantum Chemical, Phillips Petroleum, and ARCO) and even include the death of a CEO and a chairman (of McClatchy Newspapers and Gilette, respectively). Some of these events take place when the equity market is open, but others take place when the equity market is closed. For events that occur when the market is open, investors have an immediate opportunity to trade on the information. For other events, such as those that happen overnight, there is a period of no trading before the information can be impounded in prices.

One of the unique features about the U.S. equity market is the structure of the opening of the New York and American stock exchanges. Both exchanges begin trading with a call auction and then switch to a continuous trading process. This structure gives us an interesting question: Does the opening call market handle unanticipated information differently from the continuous trad- ing market and, if so, how? We partition our sample into events that take place when the market is closed and when it is open. We find different speeds of adjustment and trading processes for these partitioned events. Partitioning also lets us examine whether trading is necessary for the resolution of un- certainty with new public information.

Our general findings are that unanticipated events have an impact on prices. What is different from previous studies (Dann, Mayers, and Raab 1977; Patell and Wolfson 1984; Jennings and Starks 1985; and Ederington and Lee 1993, 1995) is the speed of the adjustment. Prior studies show that the price reaction to announcements of scheduled events takes place within 1-15 minutes. In our study, the initial price reaction to announcements of unanticipated events takes over 20 minutes. However, we find that prices tend to reverse over the following 2 hours.

Since the events in our sample are negative news events to a firm, the reversal pattern is consistent with findings that show the market overreacts to bad news (DeBondt and Thaler 1985, 1987; and Brown, Harlow, and Tinic 1988). We also find evidence of an increase in selling pressure, trading volume, and quoted dollar spreads following these announcements. In fact, selling pressure and volume remain significantly higher for more than 90 minutes. Dollar spreads are significantly wider for over an hour.

We also find that for those events that occur when the market is open, the increase in volume, selling pressure, and dollar spreads takes about 8-10 minutes to catch up to those events that occur when the market is closed. However, prices for events that occur during the day take at least 14 minutes to reach the levels for events that occur when the market is closed.

Once changes in volume, selling pressure, and prices for daytime events catch up to those of overnight events, subsequent changes in price, volume, and trading pressure are similar to those of overnight events for at least the next 3 hours. Only spreads behave differently: the size of the spread remains

110

Equity Market

much wider for those events that take place when the market is open. This

pattern continues for at least the next 3 hours. The extended pattern of wider

spreads for daytime events suggests an inventory control problem for the specialist.

Finally, we find a 3-minute delay in the market's reaction to daytime events. This short delay might be necessary to analyze the impact of new information or, perhaps, to clean out stale quotes from the limit order book. Overnight events do not have this delay, suggesting that trading might not be necessary to resolve uncertainty about new public information. What is needed is for traders to reposition their trading interests by submitting new orders and canceling old orders and for market makers to set market-clearing prices by using the information in order submissions and cancellations.

This article is organized as follows: Section II presents a literature review. Section III describes our data, sample, and variables. Section IV outlines the research procedures. Section V discusses the results. The last section sum- marizes and concludes.

II. Literature Review

Fama's (1965) introduction of an efficient market and the event study meth- odology by Fama et al. (1969) were the first articles to examine financial market efficiency and the speed with which markets adjust to new information. Empirical research since Fama et al. (1969) shows how the equity market reacts to unanticipated information. However, as the surveys of Fama (1970, 1991) and LeRoy (1989) indicate, the equity market overreacts to new in- formation (DeBondt and Thaler 1985, 1987; and Brown et al. 1988), under- reacts to earnings announcements (Bernard and Thomas 1990), and, given economic fundamentals, is too volatile (Shiller 1981). Therefore, although academics generally agree that equity markets are reasonably efficient, the debate on market efficiency is kept alive by the discovery of market anomalies.

Dann et al. (1977) were among the first researchers to examine how quickly the equity market adjusts to new information using intraday data. Dann et al. study the equity market's reaction to announcements of block trades and find that a trader would have to react within 5 minutes of an announcement to earn a positive return and that transaction prices adjust completely 15 minutes after block trades.

Several studies have examined the equity market's response to dividend and earnings announcements. Patell and Wolfson (1984) and Jennings and Starks (1985) find that the ability to earn excess returns lasts no longer than 10-15 minutes. However, volatility remains high for several hours following the announcement.

Greene and Watts (1996) examine differences in the equity market's re- sponse to earnings announcements made during trading and nontrading hours on the New York Stock Exchange (NYSE) and Nasdaq. Their results show that the two markets impound information in prices differently. For earnings

111

Journal of Business

announcements during nontrading hours, the first postannouncement trade (i.e., the opening trade) on the NYSE impounds most of the price response, but, during trading hours, the price response is spread evenly over several of the initial postannouncement trades. Greene and Watts attribute the speed of ad- justment at the opening trade on the NYSE to the procedure used by specialists to determine the opening price.' However, on Nasdaq, the first postannoun- cement trade impounds most of the price response regardless of whether the announcement occurs during trading or nontrading hours. Moreover, the price adjustment at the first postannouncement trade is greater on Nasdaq than on the NYSE for both sets of announcements.

Cao, Ghysels, and Hatheway (2000) examine Nasdaq market makers' ac- tivities during the preopening period. They find that price discovery on the Nasdaq during the preopening period is conducted via price signaling rather than the auction process used at the opening on the NYSE or the continuous market used during the trading day. Cao et al. also shed light on Nasdaq's greater speed of adjustment to overnight news announcements. They find that the preopening period facilitates greater price discovery than does the call auction process on the NYSE; hence the faster price adjustment for Nasdaq stocks at the open.

Ederington and Lee (1993, 1995) examine the impact of scheduled mac- roeconomic news releases on the interest-rate and foreign-exchange markets. They find that, in these markets, prices react within 10 seconds and that the major price adjustment occurs within 1 minute of the scheduled news releases. Their findings suggest that the interest-rate and foreign-exchange markets react much faster to public announcements than the equity markets.

Fleming and Remolona (1999) study the impact of scheduled macroeco- nomic news releases on the U.S. Treasury market. They find that the arrival of public information results in a two-stage adjustment process. In the first stage, prices react immediately, trading volume drops, and bid-ask spreads widen. In the longer second stage, volume surges, volatility persists, and spreads remain wide. Their first-stage results are the same as the NYSE's and Nasdaq's responses to the arrival of public information. The second-stage findings indicate disagreement among investors on the information content of the public announcement. Liquidity and volatility return to their normal levels once the Treasury market reaches a consensus.

Some event studies focus on the overreaction hypothesis using abnormal returns to signify the arrival of new information. These studies begin with a

change in abnormal returns and then investigate the market's efficiency by examining reversing trends in the market. For example, Bremer and Sweeny (1991) find that following large price decreases, prices rebound in subsequent periods, but following large price increases, prices remain at the new level.

1. In n. 7 of Greene and Watts (1996), they explain how specialists announce trial clearing prices before settling on an opening price. Based on the findings of Greene and Watts, the use of these "indications" allows for price discovery in a manner similar to the preopening process on Nasdaq.

112

Equity Market

The implied conclusion in their study is that the market does not handle bad news efficiently.

More recently, Hong and Stein (1999) propose a model, namely, of het- erogeneity across investors, whereby investors observe different pieces of private information at different times. One key assumption they make is that firm-specific information diffuses gradually across the investing public, thereby resulting in underreaction and positive return correlations. Hong, Lim, and Stein (2000) test the Hong and Stein model and find that negative in- formation diffuses gradually through the investing public-that is, bad news travels slowly.

Although previous studies of equity, interest-rate, and foreign-exchange markets show that information is quickly incorporated in prices, each of these studies examines an event that is anticipated by the market. For example, the equity market often anticipates the timing and amount of earnings and dividend announcements. Ederington and Lee (1995) examine the interest rate and foreign currency markets, while Fleming and Remolona (1999) examine the U.S. Treasury market's reaction to scheduled announcements of macroeco- nomic news. Since the events are anticipated, spreads, volume, and volatility can, and probably do, change during the preannouncement period. In fact, Fleming and Remolona document a widening of the spread in the U.S. Trea- sury market before the release of macroeconomic news.2 Each market should quickly adjust to news releases for anticipated events because traders can plan strategies for different potential scenarios. Then, as one of the scenarios is revealed by the announcement, traders can quickly implement the appropriate trading strategy.

In contrast to previous studies, we focus on the equity market's adjustment to fully unanticipated firm-specific events. Thus, this article provides new, unique evidence on how the equity market processes information and on the trading activity that follows the arrival of unanticipated public information. By comparing the adjustment process for unanticipated announcements that occur when the market is open to those announcements that occur when the market is closed, we provide information about the speed of adjustment and necessity of trading for the resolution of uncertainty. It may also be possible to use these results to draw inferences into the efficacy of trading halts.

III. Data, Sample, and Variables

Our study uses the intraday transaction data for 1989 through 1992, which we obtain from the NYSE and Amex Trades and Quotes Transaction File compiled by the Institute for the Study of Security Markets (ISSM). Intraday data from the ISSM tape include time-stamped transactions and bid and ask quotes. The database also contains the price and size of each transaction, the support for a quote, and opening and closing prices. We use intraday data to

2. Krinsky and Lee (1996) document a widening of the spread before earnings announcements.

113

Journal of Business

TABLE 1 Event Firms, Time and Date of Event, and Competing Firms

Event Firm*

Exxon Chevron McClatchy Newspapers UAL Quantum Chemical USAir Phillips Petroleum ARCO Gillette UAL Dominion Resources Conoco People's Natural Gas Amocot Enront Union Carbidet Burlington Northern Texaco Panhandle Eastern Pipeline TimeWarner McDonnell Douglas

Date

3/24/89 4/10/89 4/16/89 7/19/89 9/12/89 9/21/89

10/23/89 7/5/90

1/25/91 3/4/91

8/27/91 9/4/91

1/17/92 4/5/92 4/5/92 4/5/92

6/30/92 10/8/92 10/9/92

12/20/92 12/21/92

Time

5:20 A.M. 5:57 P.M.

7:33 A.M. 5:26 P.M.

10:06 A.M.

7:32 A.M. 2:38 P.M. 7:31 A.M. 3:17 P.M. 9:00 A.M.

3:50 P.M. 9:03 A.M. 6:11 P.M. 8:56 A.M. 8:56 A.M. 8:56 A.M.

11:34 A.M. 8:44 A.M.

12:22 P.M. 7:05 A.M. 7:30 A.M.

Competitor

Chevron Amoco Lee Enterprises Delta Rohm Haas Delta Unocal Hanson PLC Warner Lambert Delta Virginia Electric Mobil Questar Chevron Vastar Dow Chemical Consolidated Rail Amoco El Paso Natural Gas Disney Boeing

NOTE.-We construct our sample of event firms by using various keywords to search the Lexis/Nexis database for unanticipated events. We obtain the exact time of the event from the Dow Jones Newswire. The time of the event is based on Eastern Standard Time. A competing firm is from the same industry (based on two-digit SIC code) and is similar in size to its corresponding event firm.

* UAL has two events. t One event affected three firms. A fire and explosion on April 5, 1992, affected refineries for Amoco,

Enron, and Union Carbide, with common boundaries.

calculate returns, classify trades (buy or sell), determine trade size, and mea- sure bid-ask spreads.

We construct a sample of event firms by searching Lexis/Nexis, using various keywords, such as "unexpected," "unanticipated," "surprised," and "shocked," that highlight unanticipated events. We omit announcements re- lated to key corporate events that are partially anticipated by the equity market, such as earnings or dividend announcements, mergers and acquisitions, and so on, from the sample. Thus, our sample contains only unexpected negative news events about firms.

Once we identify a news announcement as a potentially "unanticipated" event, we determine the exact time of the event from the Dow Jones Newswire. All firms that have unanticipated events and intraday trading data on the ISSM tapes qualify for the event sample. We examine 21 fully unanticipated events, ranging from the Exxon Valdez running aground in Prince William Sound to an explosion at a Texaco refinery in Los Angeles that injured more than 16 employees. The events are negative news in that they are unpredictable ac- cidents that occur in a specific firm. Table 1 contains the list of events examined in this study.

We also look at the equity market's reaction to see if the unanticipated event has a contagion effect for the event firm's competitors. We define a

114

Equity Market

competing firm as one from the same industry and similar in size to its corresponding event firm. To control for the impact that macroeconomic news announcements might have on the trading process for each event firm and its competitor, we construct a portfolio of control firms. Each firm in this portfolio must have a two-digit SIC code that is different from that of the event firm, a market value that is within 10% of the event firm, and intraday data on the ISSM tapes. We construct a separate control-firm portfolio for each event firm.

The sample sizes in the control-firm portfolios range from 11 to 32 firms. We use competing firms to examine any contagion effects and use the control- firm portfolio to benchmark trading variables not subject to the firm-specific event. Each firm in the portfolio must also have intraday trading data on the ISSM tapes. Therefore, we examine 21 event firms, 21 competing firms, and 21 control-firm portfolios.

We use five variables to examine the trading process after announcements of unanticipated events. The first variable is the stock price (Price), which we define as the transaction price when we analyze long-term average prices and as the bid-ask spread midpoint when we examine prices on a minute-by- minute basis. The second variable is the spread in quoted prices (Spread), which we measure in both relative and nominal terms. The third variable is trading volume (Volume), which we measure by both total trading dollars and number of shares traded. The fourth variable is stock return volatility (Vol- atility), which we compute as the standard deviation of percentage changes in bid-ask spread midpoints. We measure volatility over each 10-minute win- dow around event announcements.

We use Keim's (1989) trading location measure (Location), which he defines as the transaction price minus the bid price divided by the current bid-ask spread. Trades at the bid receive a value of zero, and trades at the ask receive a value of one. The average expected value of the trading location variable is 0.5, indicating that half the trades are above and half are below the bid- ask spread midpoint. Selling pressure is indicated when the average trading location is below 0.5, and buying pressure is indicated when the average trading location is above 0.5.

IV. Procedures

We choose the event day and time for an unanticipated event based on the time stamp given by the Dow Jones Newswire. We then calculate pre-event averages for trading price, volume, nominal spread, relative spread, and trading location. We use a 10-day pre-event period beginning 11 days before the event and ending the day before the event. We compute the pre-event averages on an hourly basis over the 10-day pre-event period.

Table 2 presents the average values for these measurement variables during the pre-event period for the event firms, competing firms, and control-firm portfolios.

For the pre-event period, the average value of each of our variables is not

115

TABLE 2 Pre-event Averages for Event and Competitor Firms, and Control-Firm Portfolios

Pre-event Average (Standard Deviation)

Variable Event Firms Competitor Firms Control-Firm Portfolios

Price $45.54 $47.24 $46.07 ($33.07) ($20.21) ($26.06)

Volume 30,178 46,220 49,723 (25,566) (55,614) (54,666)

Location .48 .51 .50 (.19) (.18) (.17)

Dollar spread $.29 $.30 $.30 ($.13) ($.14) ($.13)

Relative spread .84 .81 .85 (.66) (.59) (.47)

NOTE. -We construct our sample of event firms by using various keywords to search the Lexis/Nexis database for unanticipated events. A competing firm is from the same industry (based on the two-digit SIC code) and is similar in size to its corresponding event firm. A control-firm portfolio is constructed for each event firm. Each control-firm portfolio comprises firms that are of a similar size but outside the industry of the event firm. We define Price as the average transactions price on an hourly basis. We measure Volume as the average number of shares traded on an hourly basis. We define Location, based on Keim (1989), as the transaction price minus the bid price standardized by the current bid-ask spread. We calculate the Dollar Spread as the average nominal spread, using all quotes available on an hourly basis. Relative Spread is the average of the nominal spread standardized by the most recent transaction price. We define the pre-event period as the 10 days beginning 11 days before and ending 1 day before the unanticipated event. The pre-event averages are the average hourly estimates for our variables over the pre-event period.

statistically different across the three sets of firms. Also, the trading location is close to 0.5 for all three samples, indicating that there is no buying or selling pressure during the pre-event period. Since prices, volume, and spreads are similar for the three sets of firms, the data suggest that the market does not anticipate the event. If the market had partially anticipated the event, spreads for the event firms would have widened during the pre-event period relative to those of the competing firms and those in the control-firm portfolios.

We examine abnormal changes in price, spreads, volume, and location to determine the magnitude and speed of adjustment to announcements of un- anticipated events. We compute abnormal changes as the difference between actual and expected levels of the variables following the announcement. To ensure robustness of our results, we examine three different sets of expec- tations. First, if the event occurred at 10 A.M., we compute the abnormal change 15 minutes after the event as the difference between the variable's value at 10:15 A.M. on the event day and the average value for that variable at 10:15 A.M. over the 10-day pre-event period.3 Mindful of previous research that suggests that spreads and volume vary through the trading day, we control for the time of day when we determine the abnormal change. We examine the abnormal change 30 minutes after the event by using data at 10:30 A.M. each day. We process for all time periods by using the same method.

As our second measure of expectations, we use the level of each variable 30 minutes before each event. This approach virtually eliminates any mac-

3. Lee, Ready, and Seguin (1994) also control for time of day when they examine abnormal changes in volume and volatility around NYSE trading halts.

116 Journal of Business

Equity Market

roeconomic effects that could affect the event day or the pre-event window of 10 days.

Finally, we use regressions to construct our third set of abnormal changes. We run a set of regressions to determine the relation between price, volume, spreads, and trading location that can be explained by the control-firm port- folio. We construct each control-firm portfolio on an equal-weighted basis. Our regression model is:

Pricei = a + 1Pricej + 12Volumej + ,33Spreadj

+ /34Locationj + 5 Pricei30 + (36Volumei30 (1)

+ -37Spreadi30 + L8Location,30 + ci,

where Pricei is the price of the event or competing firm i before the event, Pricej is the price of the associated control-firm portfolio, Volumej is the volume of the control-firm portfolio, Spreadj is spread of the control-firm portfolio, Locationj is the trading location parameter of the control-firm port- folio, and Ei is the relevant error term. All lagged values of Price, Volume, Spread, and Location are observations 30 minutes prior to the event.

We compile data for the dependent and independent variables on a minute-

by-minute basis. Using data from the pre-event period, we obtain regression coefficients of the independent variables. We compute postevent predicted values of event firms and their corresponding competing firms by using post- event price, volume, spread, and location values of the control-firm portfolios. We then use the parameter estimates from the regression to calculate the abnormal postevent measures of price, volume, spread, and trading location as the difference between the actual realized postevent levels and the predicted values.

Finally we separate the events into those that occur when the equity market is open (daytime events) and those that occur when the equity market is closed (overnight events).

V. Results

We find that the equity market does not react quickly to announcements of

unanticipated events. Table 3 documents changes in variables at 15, 30, 60, 90, and 120 minutes after the announcements and gives the statistical signif- icance of the differences between post- and pre-event levels.

The control-firm portfolios do not show a significant change in any of the variables. However, for the event firms, trading prices fall by an average of about $0.80 per share (or 1.60%) during the first 15 minutes following the announcement. Prices continue to move downward for another 7 minutes and

117

Journal of Business

TABLE 3 Average Changes in Variables from Pre-event Means

15 Minutes 30 Minutes 60 Minutes 90 Minutes 120 Minutes

-.80** -.14

.05B

-1.60* -.23

.OOA

-.68* .31 .04A

-1.31* .31

-.03A

1,687*** 1,673*** 1,873*** 1,822**

508B 345B

5.20** 4.91** 4.11* 4.74*

.21B .15B

.13** .07**

.06 .03

.03B .04A

.06 .10 -.02 -.02

.05 .04

-.13** -.16** .02 .02 .02A .02A

NOTE. -We construct our sample of event firms by using various keywords to search the Lexis/Nexis database for unanticipated events. A competing firm is from the same industry (based on the two-digit SIC code) and is similar in size to its corresponding event firm. A control-firm portfolio is constructed for each event firm. Each control-firm portfolio comprises similar-size firms, but outside the industry of the event firm. For each firm, we obtain the value of each variable at 15-, 30-, 60-, 90-, and 120-minute marks following the event. In addition, we compute the average value for each variable at the same time each day over the pre-event period, defined as the 10 days beginning 11 days before and ending 1 day before the unanticipated event. We define the difference between the postevent value and the pre-event average as the abnormal change in each variable. We then average the abnormal change in each variable across all firms in the subset. The figures reported below are the average abnormal changes in each variable.

* Significantly different from zero at the .10 level. ** Significantly different from zero at the .05 level. *** Significantly different from zero at the .01 level. A Significantly different from the event-sample variables at the .10 level. B

Significantly different from the event-sample variables at the .05 level. c

Significantly different from the event-sample variables at the .01 level.

then start to rebound.4 These results are based on the minute-by-minute ex- amination of the data over the entire trading day.

After 30 minutes of trading, prices are lower than the pre-event price by nearly $0.68 (or 1.31%). By the end of the first hour, prices are still down by about $0.28 (or 0.55%) per share for event firms. After another hour of trading, prices return to $0.07 (or 0.14%) below their pre-event average. Prices

4. These results have not been presented in this article, in order to conserve space, but are available from the authors on request.

Variable

Price: Event Competitor Control

% Price: Event Competitor Control

Volume: Event Competitor Control

% Volume: Event Competitor Control

$ Spread: Event Competitor Control

% Spread: Event Competitor Control

Location: Event Competitor Control

-.28 .27

-.11

-.55 .25

-.14

1,205** 1,659**

232A

4.24* 4.51*

.23A

.08**

.03 -.02

.06 -.04 -.09

-.07* .03 .01

-.18 .46

-.20

-.32 .52

-.16

611* 1,027* -335

2.98 2.33

.08

.02

.06 -.04

.05

.00 -.03

-.09* .04

-.02

-.07 .04 .14

-.14 .02 .09

345 421

9

1.78 .98

-.03

.05 -.01

.01

.02 -.08

.00

-.12* .04 .01

118

Equity Market

continue a slight upward trend for the day, finishing about $0.12 per share higher for the entire trading day.

For competing firms, prices fall by about $0.14 (or 0.23%) a share during the first 15 minutes. Then prices reverse; there is a positive spillover effect, and prices drift upward for the next 105 minutes, reaching as high as $0.60 above the pre-event average. The price increase does partially reverse, before reaching about $0.46 (or 0.52%) above the pre-event level at the 90-minute mark. At the end of 2 hours, prices end up at about $0.04 above the pre- event level. Prices remain higher for the entire day and continue to drift upward, averaging $0.60 per share higher at the end of the first trading day. The postevent prices are not statistically different from the pre-event level after 1 hour of trading but appear to be economically significant over the first day of trading.

The first 30 minutes of trading show both a negative reaction for event firms and a positive spillover effect for competing firms. The positive spillover benefits to a competing firm could be due to its expected gain in market share at the expense of the event firm's loss. Our findings also suggest that the equity market's reaction to unexpected announcements is much slower than it is in the equity, interest rate, foreign currency, and U.S. Treasury markets for scheduled announcements (see Jennings and Starks 1985; Ederington and Lee 1993, 1995; and Fleming and Remolona 1999). While the interest rate, foreign currency, and Treasury markets are much more liquid than the equity market, the findings in Jennings and Starks should provide a meaningful benchmark for our findings on the unanticipated events. However, in the event the stocks examined by Jennings and Starks are more liquid than those ex- amined in this study, we also control for liquidity effects in our subsequent analyses.

We also show that during the 90 minutes following announcements, event firms experience a price reversal, but prices for competing firms continue to show an upward drift. These findings support the overreaction results noted by Brown et al. (1988) and Bremer and Sweeny (1991). However, in our study, price recovery takes much longer than it does in earlier studies. Volume, although significantly higher during the first 90 minutes of trading for both event firms and competing firms, begins to return to pre-event levels by the end of the second hour of trading. The control-firm portfolios do not show a significant change in volume over the 2-hour window or for the entire trading day.

For event firms, the dollar bid-ask spread increases significantly over the first hour of trading and then starts to reverse. However, the spread remains wider throughout the first 2 hours of trading. The competing firms see a small increase in dollar spreads, but the increase is not statistically significant. Again, the control-firm portfolios show no changes in dollar spreads.

Using data from tables 2 and 3, we see that event firms exhibit a significant increase in selling pressure as the trading location measure decreases from

119

Journal of Business

0.48 to 0.32 during the first 30 minutes of trading. Competing firms experience an insignificant increase in buying pressure as the trading location measure increases from 0.51 to 0.53. Similarly, the control-firm portfolios show no significant change in trading location either during the first hour or throughout the first trading day.

During the second hour of trading, selling pressure continues for event firms. It returns to normal by the end of the third hour (0.46).5 For competing firms, buying pressure stays marginally higher for the first 3 hours but returns to normal by the fourth trading hour. The control-firm portfolios show no distinctive pattern of buying or selling pressure throughout the trading day.

We use price, spreads, volume, and location levels 30 minutes before the events as our second proxy for expected levels following the events and to determine the robustness of our findings. The results in table 4 strongly suggest that our conclusions are robust: for the event firms, prices are significantly lower over the first 30 minutes following the events, volume is significantly higher for the following 90 minutes, dollar spreads significantly increase over the first hour, and the trading location parameter is significantly lower over the first 2 hours. The competing firms show a significant increase in volume. The control-firm portfolios have no significant changes in any of the four variables. The levels of the abnormal changes are similar in magnitude to those reported in table 3.

Next, we examine whether the positive drift to competing firms, in tables 3 and 4, is due to the expected gains in market share at the expense of the event firm. We construct two sets of competing firms from among all firms with the same two-digit SIC code as the event firm. First, we select the firm whose sales level is closest to that for the event firm. We call this set of firms "Competitor 1." Next, we select the firm with the smallest level of sales and call this set "Competitor 2." We use sales in this part of the analysis as a

proxy for market share. Using prices, spreads, volume, and location levels 30 minutes before the event as our proxy for the expected level following the event, we recompute abnormal changes in each of the four variables for the event firms and the two sets of competitor firms. Table 5 indicates that our

previous results continue to hold for "Competitor 1" firms. However, "Com-

petitor 2" firms appear to be unaffected by the events. This result suggests that any positive spillover effect appears to be more likely for similar size firms and for firms with similar market share. Smaller firms, or firms with significantly smaller market shares, do not appear to benefit from the unan- ticipated bad-news event.

As a final check for robustness, we reconfirm our results by using the predicted values from the control-firm portfolios to construct changes in the postannouncement levels of the variables. The control firms in table 6 are based on size and are defined as in tables 3 and 4. We use the parameter estimates from equation (1) to compute postevent predicted values for the

5. These results are available from the authors on request.

120

TABLE 4 Average Changes in Variables from 30 Minutes before Event Announcement

Variable

Price: Event Competitor Control

% Price: Event Competitor Control

Volume: Event Competitor Control

% Volume: Event Competitor Control

$ Spread: Event Competitor Control

% Spread: Event Competitor Control

Location: Event Competitor Control

15 Minutes 30 Minutes 60 Minutes 90 Minutes 120 Minutes

-.81** -.15

.06B

-1.61* -.28

.08A

1,750** 1,804**

334B

5.45** 4.06*

.35B

.14**

.05

.03B

.08 -.03

.07

-.15** .03 .04A

-.69* .27 .05A

-1.30* .33 .07A

1,722** 1,853**

459A

5.21* 4.34*

.27B

.08**

.02

.04

.10 -.04

.03

-.17** .03 .04A

-.30 .24

-.11

-.51 .29

-.08

1,339** 1,689**

375A

4.20* 3.95*

.37A

.09**

.02 -.02A

.09 -.06 -.08

-.08* .03 .02

-.22 .07

-.18

-.28 .63

-.29

749** 958**

-401A

3.24 2.47

.20

.02

.06* -.04

.06 -.02 -.03

-.10* .05

-.01

-.11 .05 .16

-.13 -.09

.33

421 334 27

2.22 1.09 .08

.06 -.02

.01

.18 -.10

.02

-.13* .05 .02

NOTE. -We construct our sample of event firms by using various keywords to search the Lexis/Nexis database for unanticipated events. A competing firm is from the same industry (based on the two-digit SIC code) and is similar in size to its corresponding event firm. A control-firm portfolio is constructed for each event firm. Each control-firm portfolio comprises similar-size firms, but outside the industry of the event firm. For each firm, we obtain the value of each variable at 15-, 30-, 60-, 90-, and 120-minute marks following the event. We define the difference between the postevent value and the value of the variable 30 minutes before the event as the abnormal change in each variable. We then average the abnormal change in each variable across all firms in the subset. The figures reported below are the average abnormal changes in each variable.

* Significantly different from zero at the .10 level. ** Significantly different from zero at the .05 level. *** Significantly different from zero at the .01 level. A Significantly different from the event-sample variables at the .10 level. B Significantly different from the event-sample variables at the .05 level. c

Significantly different from the event-sample variables at the .01 level.

levels of each variable. As the third measure of abnormal changes following the event, we use the differences between the actual postevent variables and the predicted values based on the control-firm portfolios.

The results in table 6 indicate that our findings are robust.6 The magnitude of the abnormal changes is similar across all three sets of results in tables 3, 4, and 6.

6. As a final check for robustness, we included hour-of-the-day and day-of-the-week dummy variables in the regression. Almost all the dummy variable coefficients were insignificant. Most importantly, the abnormal changes based on this regression are qualitatively similar to those reported in table 6. As such, these findings have not been reported but are available upon request.

Equity Market 121

TABLE 5 Average Changes in Variables from 30 Minutes before Event Announcement for Two Sets of Competitor Firms

Variable 15 Minutes 30 Minutes 60 Minutes 90 Minutes 120 Minutes

Price: Event -.81** -.69* -.30 -.22 -.11 Competitor I -.10 .22 .26 .04 .04 Competitor 2 .02 .01A .03 -.02 -.02

% Price: Event -1.61* -1.30* -.51 -.28 -.13 Competitor 1 -.34 .27 .19 .40 .06 Competitor 2 .04A .17A -.01 -.11 .24

Volume: Event 1,750** 1,722** 1,339** 749** 421 Competitor 1 1,432** 1,567** 1,333** 720* 559 Competitor 2 -120B 301A 229A 109 115

% Volume: Event 5.45** 5.21* 4.20* 3.24 2.22 Competitor 1 3.98* 4.12* 3.71* 2.59 .88 Competitor 2 .19B . .7 24 -.05 .13

$ Spread: Event .14** .08** .09** .02 .06 Competitor 1 .05 .01 .02 .04* .01 Competitor 2 .01B .02 -.01A -.02 .01

% Spread: Event .08 .10 .09 .06 .18 Competitor 1 .00 -.02 -.03 -.02 -.06 Competitor 2 .04 -.01 .03 .01 .02

Location: Event -.15* .17** -.08* -.10* -.13 Competitor 1 .01 .02 .01 .04 .06 Competitor 2 .02 .03 .02 .00 -.01

NOTE.- We construct our sample of event firms by using various keywords to search the Lexis/Nexis database for unanticipated events. A "Competitor 1" firm is from the same industry (based on the two-digit SIC code) and is similar in sales to its corresponding event firm. A "Competitor 2" firm is from the same industry (based on the two-digit SIC code) and has a much smaller sales level than its corresponding event firm. For each firm, we obtain the value of each variable at 15-, 30-, 60-, 90-, and 120-minute marks following the event. We define the difference between the postevent value and the value of the variable 30 minutes before the event as the abnormal change in each variable. We then average the abnormal change in each variable across all firms in the subset. The figures reported below are the average abnormal changes in each variable.

* Significantly different from zero at the .10 level. ** Significantly different from zero at the .05 level. *** Significantly different from zero at the .01 level. A Significantly different from the event-sample variables at the .10 level. B Significantly different from the event-sample variables at the .05 level. c

Significantly different from the event-sample variables at the .01 level.

The small size of our sample lets us look at each event firm's price reaction over the first 2 hours. Table 7 presents the percentage price changes from the

pre-event level for each of the 21 event firms. The firm with the largest initial price reaction is ARCO, which has a negative

price reaction of 2.50% in the first 15 minutes. The firm with the smallest negative price reaction is McClatchy Newspapers, which has a 0.79% decrease in the first 15 minutes. All 21 firms have a negative price reaction over the entire first 30 minutes of trading. This finding indicates that the equity market's response to unexpected announcements takes much longer than the 10-15

122 Journal of Business

TABLE 6 Average Changes in Variables from Control-Firm Portfolio Regression Predicted Values

Variable 15 Minutes 30 Minutes 60 Minutes 90 Minutes 120 Minutes

Price: Event -.87* -.79* -.41 -.26 -.20 Competitor -.20 .17 -.05 .22 .14

% Price: Event -1.55 -1.32 -.50 -.30 -.17 Competitor -.20 .25 .36 .57 -.20

Volume: Event 1,777** 1,742* 1,008* 609 524 Competitor 1,625* 1,544* 986 205 192

% Volume: Event 4.40* 4.32* 3.77 2.95 1.54 Competitor 3.72* 4.50* 3.33 2.50 1.44

$ Spread: Event .16* .10* .09* .03 .04 Competitor .06 .03 .03 .04 .00

% Spread: Event .11 .09 .07 .05 .06 Competitor .01 -.01 -.03 -.03 -.01

Location: Event -.14* -.14* -.10 -.11 -.11 Competitor .01 .02 .02 .04 .03

NOTE. -We construct our sample of event firms by using various keywords to search the Lexis/Nexis database for unanticipated events. A competing firm is from the same industry (based on the two-digit SIC code) and is similar in size to its corresponding event firm. A control-firm portfolio is constructed for each event firm. Each control-firm portfolio comprises similar-size firms, but outside the industry of the event firm. For each firm, we obtain the value of each variable at 15-, 30-, 60-, 90-, and 120-minute marks following the event. We define the difference between the postevent value and a regression-predicted value from the control-firm portfolio as the abnormal change in each variable. We then average the abnormal change in each variable across all firms in the subset. The figures reported below are the average abnormal changes in each variable.

* Significantly different from zero at the .10 level. ** Significantly different from zero at the .05 level. *** Significantly different from zero at the .01 level.

minutes suggested in Jennings and Starks (1985), or within 1 minute in the interest-rate and foreign-exchange markets, as in Ederington and Lee (1993, 1995) and the U.S. Treasury market studied in Fleming and Remolona (1999). It is not until after an hour of trading that some firms begin to reverse the early trading trends. At the 60-minute mark, the percentage price changes of 15 of the 21 firms are still negative. We note that the firms with smallest negative price reactions are not the first firms to reverse.

After 2 hours of trading, 14 firms still have a negative price reaction. Six of these had a complete reversal before the 2-hour price change measurement. Seven firms show a complete reversal and are trading with a positive price change. Therefore, only eight firms had negative prices across the entire 2- hour window.

The size of the initial negative price reaction does not predict which firms will have negative prices across the 2-hour window. The immediate price reaction is not an indicator of how long it will take for the price to reverse, or even if the price will reverse. In fact, ARCO, which has the largest initial

Equity Market 123

Journal of Business

TABLE 7 Percent Price Change by Firm at 15, 30, 60, 90, and 120 Minutes

Firm Open/Closed 15 Minutes 30 Minutes 60 Minutes 90 Minutes 120 Minutes

ARCO Closed -2.50 -.70 -.71 .06 -.63 Burlington Open -2.43 -.62 .40 .36 -1.04 McDonnell Closed -2.40 -2.37 -1.45 -.17 -1.04 Amoco Closed -2.27 -2.01 -.10 -1.19 .01 USAir Closed -2.11 -1.05 -.98 .06 .78 People's Closed -2.04 -2.22 -.11 -1.18 .40 Panhandle Closed - 1.99 -.76 -1.59 .31 .13 Quantum Open -1.86 -1.40 .06 -.28 .54 Exxon Closed - 1.86 -2.10 -1.29 -1.06 -.95 Enron Closed - 1.78 -1.67 -1.44 .52 -.16 Chevron Closed - 1.65 -1.06 -.40 -.93 -.68 UAL Closed -1.60 -.88 -1.07 -1.13 -.04 Gillette Open - 1.36 -1.11 -.05 -.55 -.12 Dominion Open - 1.35 -1.35 .37 -1.09 -.10 UAL Closed -1.31 -.84 .30 .61 -.92 Carbide Closed -1.26 -1.98 -.83 -.90 -.78 Texaco Closed -1.18 -1.44 -.96 -.02 .62 Phillips Open -1.14 -.68 .28 .67 .21 TimeWarner Closed -1.14 -2.12 -1.38 -.91 -.24 Conoco Open -.82 -2.08 .26 -.31 -.19 McClatchy Closed -.79 -.50 -.97 -.78 -.52 Average -1.66 -1.38 -.56 -.38 -.22

NOTE. -We construct our sample of event firms by using various keywords to search the Lexis/Nexis database for unanticipated events. We obtain the exact time of the event from the Dow Jones Newswire. Open indicates that the unanticipated announcement occurred when the equity market was open, and Closed indicates that the announcement occurred when the equity market was closed for trading. We compute the percent price changes relative to the last trade before the news announcement. We list the firms in descending order based on the magnitude of the percent price change over the first 15 minutes of trading following the unanticipated announcement.

negative price reaction, reverses after 90 minutes of trading. McClatchy News- papers, which has the smallest initial reaction, does not reverse during the 2- hour window.

Next, we partition the events into two subsamples: initial announcements of an event that cross the newswire when the exchanges (NYSE and Amex) are open and those that happen when the exchanges are closed. Figures 1 through 4 show minute-by-minute changes in price, volume, trading location, and spreads for the event sample, beginning 5 minutes before and ending 20 minutes after the unanticipated announcements. Although the small size of our sample severely restricts many statistical tests, graphs of the results are instructive.

Price changes are immediate for the overnight events, which suggests that it does not take trading to have an impact on prices. This finding supports Greene and Watts (1996), who examine price adjustments following earnings announcements that come during nontrading hours.

The daytime events show a delay of 3 minutes before prices move. Fol- lowing daytime events, prices take nearly 15 minutes to show the same price impact as the overnight events. We note that minute -1 is different for the two partitioned subsamples. The overnight event minute -1 is the last minute

124

of trading for the prior day. The daytime event minute -1 is 1 minute before the announcement during the trading day.

One explanation for the 3-minute price reaction delay is that investors are trading against stale quotes. A majority of the trades in the first 3 minutes following the event are small volume trades transacted at the bid quote. Ap- plying Occam's Razor, the simplest explanation would be that immediately following the news announcements, specialists match market sell orders to limit orders on the limit-order book. The 3-minute delay causes minimal price movement because initial trades are simply removing stale bid quotes. Once the stale quotes are removed, transactions start to move prices quickly.

Volume immediately jumps up for the overnight events. However, it takes time to build up for the daytime events. After about 8-10 minutes of trading, the daytime events show trading volume similar to that of the overnight events.

Selling pressure, as measured by the trading location variable, behaves almost the same as the volume variable. After about 8-10 minutes, the location variable is essentially the same for both daytime and overnight samples.

The minute-by-minute results for prices, volume, and selling pressure in- dicate that the equity market adjusts slowly to unexpected announcements that may have a material impact on the firm. These findings contrast with the

speed of adjustment following announcements of scheduled events (e.g., earn- ings or dividend announcements, or announcements of macroeconomic news) documented in previous studies.

Our fourth measure, the bid-ask spread, responds to overnight events by jumping in the first minute. It remains at this level over the next 20 minutes of trading. Again, the daytime events take about 8-10 minutes to reach the same level but continue to widen over the next 5-6 minutes. The spreads remain wider for the next 2 hours.

Why do spreads remain wider for daytime events? One of the features of the equity markets is the difference between the opening trade and continuous trading throughout the day. At the opening trade, the specialist examines the overnight order flow and the current limit-order book. He opens the stock at a price that incorporates the order information. Based on the order flow, the specialist may or may not take an inventory position to increase the number of orders executed at the opening. Thus, when an event takes place when the equity market is closed, the specialist can use the order flow to pick the opening price and, simultaneously, his inventory position.

When a daytime event takes place in the continuous market, the specialist is acting in a different trading environment. In both cases, the specialist is responsible for maintaining an orderly market. However, during continuous trading, the specialist might need to take the opposite side of a market order to complete this charge. Thus, as selling pressure and volume increases, the specialist is probably participating in more and more trades to maintain an orderly market. Madhavan and Sofianos (1998) find that the specialist par- ticipates more actively when bid-ask spreads are wide and previous price

Equity Market 125

Journal of Business

movements are volatile. Knowing that his posted bid price will probably be the highest bid, the specialist increases the spread (lowers his bid) to minimize his long and growing inventory position. It will also take the specialist longer to unwind this long inventory position throughout the trading day, so he needs to maintain a wider bid-ask spread. Hence, the spread is wider and stays wider longer for daytime events. A wider spread reflects the inventory cost com- ponent of the bid-ask spread.

Although we note that the NYSE does allow the specialist to request a trading halt for a firm with a large order imbalance or for a firm with an important announcement, we did not find any trading halts for our sample of events.

The extended reaction to unanticipated events when the equity market is open suggests that circuit breakers or trading halts might be appropriate. One argument for trading halts is that a nontrading period allows information to be transmitted to all market participants before trading. This information might otherwise give one set of traders an advantage over another.

Although our results indicate that overnight events have consensus at the opening, empirical investigations of trading halts do not support such action. Lee et al. (1994) find that trading halts increase, rather than reduce, volume and volatility. The increased volume and volatility after the trading halt might not result from the dissemination of information but reflect the process and timing used to reopen the market. Therefore, using trading halts would be a function of both the information and time sequence necessary to inform all potential traders, so that the call market to reopen trading would fully reflect market demand.

Corwin and Lipson (2000) study order flow and liquidity around NYSE trading halts. They find that traders take advantage of a halt to submit new orders and cancel old orders. Halts thus allow traders to reposition their interests around the new price following the news release. This result may be consistent with NYSE' s objectives of using trading halts. However, Corwin and Lipson also find that market makers are reluctant to provide liquidity during these unusual times.

As do Lee et al. (1994), Corwin and Lipson (2000) find that volume and volatility increase significantly following the trading halt, but their results

suggest that decreased liquidity explains a small portion, at best, of the in- creased volatility. Even though they find that the reopening price is a good indicator of the future price, they state, "It should be emphasized that our results cannot resolve the question of whether trading halts and related actions like 'spreading the quote' are desirable. The central problem is that we cannot know what would have occurred in the absence of a halt or what equilibrium trading patterns would be in a market where halts are not permitted" (Corwin and Lipson 2000, p. 1773).

In figures 1-4, our results provide some interesting insights in comparison with the results in Corwin and Lipson (2000). Our overnight events occur when there is no trading activity and allow market participants to reposition

126

Equity Market

-5 0 5 10 15 20

Event Time in Minutes

I-- Overight Events -- Daytime Events

FIG. 1.-Price change by minute for event firms. We define "price" as the bid-ask spread midpoint. We compute an average price for each minute, beginning 5 minutes before the event and ending 20 minutes after the event. The change in price is the average price during each minute surrounding the event less the average price during the pre-event period. We also compute the average price during the pre-event period on a minute-by-minute basis. Overnight events occur when the equity market is closed, and daytime events occur when the equity market is open.

2500 2000 1500 1000 500

0 -500

-1000 I I I I I I I I I I I I I

/'? _i ^ ?*??*

-5 0 5 10 15 20

Event Time in Minutes

- -Overnight Events - Daytime Events

FIG. 2.-Volume changes by minute for event firms. We compute average volume for each minute, beginning 5 minutes before the event and ending 20 minutes after the event. The change in volume is average volume during each minute surrounding the event less average volume during the pre-event period. We compute the average volume during the pre-event period on a minute-by-minute basis. Overnight events occur when the equity market is closed, and daytime events occur when the equity market is open.

* 0.5 U)

c 0

0 -0.5 0 C -1 CL

O

0) 0 E 5 O

127

Journal of Business

* 0.1 0) 0.05

-0.05 ? -0.1 'U -0.15 ? -0.2 1 , , , , , , , , , ,

-5 0 5 10 15 20

Event Time in Minutes

-- Overnight Events - Daytime Events |

FIG. 3.-Trading location changes by minute for event firms. Keim (1989) defines his trading location measure as the transaction price minus the bid price divided by the current spread. We use the bid-ask spread midpoint instead of the transaction price. We compute the average trading location measure for each minute, beginning 5 minutes before the event and ending 20 minutes after the event. Change in location is the average location during each minute surrounding the event less the average location during the pre-event period. We compute the average location during the pre-event period on a minute-by-minute basis. Overnight events occur when the equity market is closed, and daytime events occur when the equity market is open.

their interests before the reopen. Thus, they are similar to what occurs during a trading halt. Moreover, market makers use similar processes to set the open- ing price at the beginning of the trading day, as they do following a trading halt. Figure 1 suggests that the average opening price following an overnight event is a good indicator of prices for at least the next 20 minutes. In contrast, daytime events take 15 minutes to catch up to those following overnight events. Figures 2, 3, and 4 suggest that, following an overnight event, volume, trading location, and dollar spreads at the open are good indicators of the level of the variable for at least the first 20 minutes. This result suggests that it might be beneficial to have a period of no trading, during which investors could reposition their trading interests, and a reopening process that would allow market makers to apply the information in order submissions and cancellations to gauge the pulse of the market and set prices.

However, we wish to add two additional thoughts: first, even though the exchange could have called for a trading halt due to the news event, our daytime events apparently were not substantial enough to warrant a trading halt. In spite of this, dollar spreads remain wider, and volume and volatility remain higher for a longer period following daytime events. These results contrast with the findings in Corwin and Lipson (2000). Second, although it takes time for the market to react to daytime events, the time lag is, at most, 15 minutes. In contrast, the average news halt in Corwin and Lipson's sample

128

Equity Market

0.3

0.2

0.1 0

-0.1I . I I i I I

-5 0 5 10 15 20

Event Time in Minutes

I- Ovemight Events -- Daytime Events

FIG. 4.-Spread changes by minute for event firms. We compute the average bid- ask spread for each minute, beginning 5 minutes before the event and ending 20 minutes after the event. The change in spread is the average spread during each minute surrounding the event less the average spread during the pre-event period. We also compute the average spread during the pre-event period on a minute-by-minute basis. Overnight events occur when the equity market is closed, and daytime events occur when the equity market is open.

lasts over 85 minutes. Therefore, although our results are informative, we do not believe they allow us to make definitive statements about the efficacy of trading halts.

Figure 5 presents 10-minute volatility changes for the partitioned event firms. We define volatility as the standard deviation of returns during each 10-minute interval. We use bid-ask spread midpoints to compute returns during each interval. Volatility increases by 50% in the first 10 minutes. The daytime event volatility increase has a slight lag in adjustment during the first 20 minutes but is nearly the same for the next 3 hours.

This increased volatility pattern is another indication of the price uncertainty faced by the specialist. High volatility following the events indicates either that there is uncertainty about the impact of the events on the firms or that the specialist has inventory control problems. The specialist could react by widening the bid-ask spread during this period of greater uncertainty.

The slight lag for daytime events (over the first 20 minutes) is consistent with the wider spread for daytime events. A specialist has time to evaluate an event that took place overnight before the market opens. With the overnight order flow, there is less uncertainty about the impact of the event on prices and spreads when trading begins.

Overnight events do not show a delay in price, volume, selling pressure, or spreads. Our results suggest that trading might not be necessary to resolve uncertainty about new public information. Similar to Corwin and Lipson's (2000) recent study, Cao et al. (2000) also document price discovery without trading. They show how quote behavior in the preopening of the top 50 Nasdaq

129

Journal of Business

0.6

0.4

E 0.3

0.2 -6 0 6 12 18

10-Minute Time Intervals

'-4Night " Day

FIG. 5.-Trading volatility by 10-minute intervals for event firms. We construct our sample of event firms by using various keywords to search the Lexis/Nexis database for unanticipated events. We obtain the exact time of the event from the Dow Jones Newswire. We define volatility as the standard deviation of returns during each 10- minute window. We compute returns during each 10-minute window by using the bid- ask spread midpoint.

stocks influences price changes between the close and the opening prices. For

unanticipated events that are announced when the market is closed, price discovery on NYSE takes place via overnight order flow before the market

opens. The specialist announces trial clearing prices ("indications") before

settling on an opening price instead of making any formal preopening price quotations.

Our results show that there is a significant market reaction to announcements of unanticipated events. Are these events economically significant for firms and traders? To address this question, we first assess the economic impact of the announcement by calculating the change in at-the-money option prices for these firms. We look at a 1-month at-the-money call (put) option on a

typical event firm in our sample. We assume a risk-free interest rate of 6%. From our figure 5, we find that pre-event volatility is around 35% and that

postevent volatility is roughly 50%. Using only the change in volatility over the first 10 minutes, the Black-Scholes at-the-money call and put prices in- crease by 40% and 46%, respectively. If we incorporate the stock price drop with the volatility increase, the increase in put premium is further exacerbated, and the increase in call premium is somewhat mitigated, by the offsetting price factor. Therefore, unanticipated events have a strong economic impact on both the equity and options markets.

We can estimate the economic impact on the specialist by the product of

130

Equity Market

the trading volume and the price change minus the spread over the first 30 minutes. If the specialist is making the market and thus is required to take a

position that is opposite to the sell orders, then he would be buying at the bid price. Therefore, the estimated average initial loss is around $5,400 over the first 30 minutes. Since the estimated first half-hour of trading volume is 16,762 shares and the average price change ($0.68) minus the spread ($0.36) is $0.32, the product is just under $5,400 for the first half-hour. We also calculate profit to the specialist over a normal hour of trading. We use the NYSE Fact Book for the Year 1999 (2000) to determine typical dealer activity of the specialist and assume the specialist makes the spread on a round-trip transaction. We find that a typical hour's profit is around $225. In other words, the specialist could take more than 3 trading days to recover this one half- hour loss.7 This finding, too, suggests that the economic impact of the un- anticipated event is significant.

VI. Summary and Conclusion

We examine 21 fully unanticipated news events. We measure the equity mar- ket's reaction to these events by using prices, volume, trading location, spreads, and volatility.

We find that the equity market's reaction to unanticipated information is not impounded in prices as quickly as suggested by previous research on scheduled events. However, the market's response to unanticipated events that take place when the market is closed suggests that trading is not necessary for prices to react. When the market has time to digest the information before the trading day opens, the reaction is immediate in price, volume, selling pressure, and bid-ask spreads. When the news comes to the market during the trading day, the market reacts more slowly on price, volume, selling pressure, and bid-ask spreads than previous research would indicate.

Second, our study shows that, following announcements of bad news events, prices reverse following the initial negative price reaction, a pattern consistent with the finding of overreaction in previous studies. The initial price reaction drifts downward for at least 15-20 minutes, but often completely reverses over the next hour-and-a-half for event firms. The competing firms apparently receive a positive spillover from the bad news, but this price increase is not reversed in subsequent trading.

Daytime events induce wider bid-ask spreads and longer duration of wider spreads compared with overnight events. We suspect that this result is due to the inventory cost that the specialist faces for managing an orderly market across daytime events.

Overall, our results are not only statistically significant and robust but also

7. If this hourly profit appears small, it is important to keep in mind that the specialist has more than one stock (typically six to 10), and that the specialist is paid for other functions such as brokering an order left at the post.

131

Journal of Business

economically significant. Our results suggest that if traders can reposition their trading interests (by submitting new orders and canceling old orders) and market makers can set prices using information in order submissions and cancellations, trading may not be needed to reach market-clearing prices.

Market-clearing prices appear to be better indicators of future prices than if trading were allowed to continue following the news event. However, we cannot conclusively state that trading halts are superior, since trading halts tend to last much longer than the 15 minutes for prices following daytime events to catch up to those following overnight events.

References

Bernard, V. J., and Thomas, J. 1990. Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting and Economics 13:305-0.

Bremer, M., and Sweeny, R. 1991. The reversal of large stock-price decreases. Journal of Finance 46:747-54.

Brown, K.; Harlow, W.; and Tinic, S. 1988. Risk aversion, uncertain information and market efficiency. Journal of Financial Economics 14:6-13.

Cao, C.; Ghysels, E.; and Hatheway, F. 2000. Price discovery without trading: Evidence from the Nasdaq pre-opening. Journal of Finance 55:1339-66.

Corwin, S. A., and Lipson, M. L. 2000. Order flow and liquidity around NYSE trading halts. Journal of Finance 55:1771-1801.

Dann, L.; Mayers, D.; and Raab, R. 1977. Trading rules, large blocks and the speed of adjustment. Journal of Financial Economics 4:3-22.

DeBondt, W F. M., and Thaler, R. H. 1985. Does the stock market overreact? Journal of Finance 40:793-805.

DeBondt, W. F. M., and Thaler, R. H. 1987. Further evidence on investor overreaction and stock market seasonality. Journal of Finance 42:557-81.

Ederington, L. H., and Lee, J. H. 1993. How markets process information: News releases and volatility. Journal of Finance 48:1161-91.

Ederington, L. H. and Lee, J. H. 1995. The short-run dynamics of the price adjustment to new information. Journal of Financial and Quantitative Analysis 30:117-34.

Fama, E. F. 1965. The behavior of stock market prices. Journal of Business 38:34-105. Fama, E. F. 1970. Efficient capital markets: A review of theory and empirical work. Journal of

Finance 25:383-417. Fama, E. F. 1991. Efficient capital markets: II. Journal of Finance 46:1575-1617. Fama, E. F.; Fisher, L.; Jensen, M. C.; and Roll, R. 1969. The adjustment of stock prices to new

information. International Economic Review 10:1-21. Fleming, M., and Remolona, E. 1999. Price formation and liquidity in the U.S. Treasury market:

The response to public information. Journal of Finance 54:1901-27. Greene, J. T, and Watts, S. G. 1996. Price discovery on the NYSE and the Nasdaq: The case

of overnight and daytime news releases. Financial Management 25:19-42. Hong, H.; Lim, T.; and Stein, J. 2000. Bad news travels slowly: Size, analyst coverage, and the

profitability of momentum strategies. Journal of Finance 55:265-95. Hong, H., and Stein, J. 1999. A unified theory of underreaction, momentum trading, and over-

reaction in asset markets. Journal of Finance 54:2143-84. Jennings, R., and Starks, L. 1985. Information content and the speed of stock price adjustment.

Journal of Accounting Research 23:336-50. Keim D. B. 1989. Trading patterns, bid-ask spreads, and estimated security returns: The case of

common stock at calendar turning points. Journal of Financial Economics 25:75-98. Krinsky, I., and Lee, J. 1996. Earnings announcements and the components of the bid-ask spread.

Journal of Finance 51:1523-35. Lee, C. M.; Ready, M.; and Seguin, P. 1994. Volume, volatility, and New York Stock Exchange

trading halts. Journal of Finance 49:183-214. LeRoy, S. F. 1989. Efficient capital markets and martingales. Journal of Economic Literature

27:1583-1621.

132

Equity Market 133

Madhavan, A., and Sofianos, G. 1998. An empirical analysis of NYSE specialist trading. Journal of Financial Economics 48:189-210.

New York Stock Exchange (NYSE). 2000. Fact Book for the Year 1999. New York: NYSE. Patell, J., and Wolfson, M. 1984. The intraday speed of adjustment to stock price earnings and

dividend announcements. Journal of Financial Economics 13:223-52. Shiller, R. J. 1981. Do stock prices move too much to be justified by subsequent changes in

dividends? American Economic Review 71:121-36.