Large price movements and short-lived changes in spreads, volume, and selling pressure

Transcription

1 The Quarterly Review of Economics and Finance 39 (1999) Large price movements and short-lived changes in spreads, volume, and selling pressure Raymond M. Brooks a, JinWoo Park b, Tie Su c, * a Oregon State University, 200 Bexell Hall, Corvallis, OR 97330, USA b Hankuk University of Foreign Studies, College of Business and Economics, Imun-Dong, Seoul , Republic of Korea c Department of Finance, University of Miami, PO Box , Coral Gables, FL , USA Abstract In this paper we examine changes in dollar and relative bid-ask spreads of stocks following large price movements. We investigate large increases and decreases separately and link our results to current market microstructure theories on trading activities and spreads. We also look at changes in volume and selling pressure to interpret the changes in trading activity. Our results show that the market reacts differently to price increases and decreases. For large price decreases, trading increases on the sell side even when spreads have increased. For large price increases, trading increases on the buy side during a period of higher spreads. However, the increases in dollar spreads and price pressure are most pronounced at the end of trading day. Our results are consistent with microstructure models that link trading activities and costs to the level of asymmetric information Board of Trustees of the University of Illinois. All rights reserved. Keywords: Market microstructure; Bid-ask spread; Trading volume. 1. Introduction In this paper we examine changes in dollar and relative bid-ask spreads of stocks following large price movements. We investigate large increases and decreases separately * Corresponding author. Tel.: address: tie@miami.edu (T. Su) /99/$ see front matter 1999 Board of Trustees of the University of Illinois. All rights reserved. PII: S (99)

2 304 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) and link our results to current market microstructure theories on trading activities and spreads. We also look at changes in volume and selling pressure to interpret the changes in trading activity following large price changes. Our results show that the market reacts differently to large price increases and decreases. For large price decreases, trading increases on the sell side, even though spreads have increased. For large price increases, trading increases on the buy side during a period of higher spreads. Microstructure models on trading activity rely on an association with the level of trading costs, and more specifically on the size of the bid-ask spread. One set of models (Admati and Pfleiderer, 1989; Glosten, 1989; Leach and Madhavan, 1993) predict an increase in spreads following large price changes to speed up price discovery. Market makers increase the spread so that a higher percentage of trades will be initiated by informed traders. They incur temporary losses, but these losses are offset by the benefit of finding the new equilibrium price quickly. Another set of models (Kyle, 1985; Glosten and Milgrom, 1985; Admati and Pfleiderer, 1988) predict that following large price changes, asymmetric information is reduced and bid-ask spreads should narrow. In these models, spreads decrease as the adverse selection component of spreads is reduced during periods of lower asymmetric information. The two different sets of microstructure models essentially define different sets of trading agents in the market following large price changes. The price discovery models imply a high bid-ask spread and a trading profile with a greater percentage of trading initiated by informed traders to take advantage of the higher spreads. The asymmetric information models imply a lower bid-ask spread and an increase in volume as more liquidity traders choose to re-balance their portfolios when trading costs and information asymmetry is lowest. We examine changes in spreads, volume, and selling pressure following both large price increases and decreases to test whether markets have similar trading patterns following these events and which set of microstructure models better explains the observed trading patterns. We find that both relative and dollar spreads have a short-lived increase following large price decreases, consistent with strategic market makers microstructure models. Furthermore, we find that dollar spreads temporarily increase following large price increases, and that trading volume increases significantly after a large price movement. Finally, we find that selling pressure increases following a price reduction, but selling pressure decreases following a price increase. A curious question is why price increases lead to an increase in dollar spreads and a decrease in selling pressure, while price decreases result in an increase in both dollar spreads and selling pressure. Our findings seem to reach back into the literature on good news versus bad news releases (Patell and Wolfson, 1982; Penman, 1987; Fishe et al., 1993.) On the one hand, firms releasing good news want to credibly disclose their information. Following a good news release, information asymmetry is reduced among all traders, and uninformed buying increases while information asymmetry is low. Increased volume with a balance of trading from the seller-initiated trades and buyer-initiated trades is consistent with a good news release and microstructure models predicting a reduction in asymmetric information. On the other hand, firms releasing bad news have an incentive to delay or soften the impact of the bad news. Firms often announce some good news with their bad news. Market makers, aware of a firms incentive to minimize the impact of bad news, may require additional time to analyze and separate any confounding elements of the release. Market

3 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) makers may also require additional information from more informed traders to interpret the bad news. One way to acquire information quickly is to temporarily increase spreads and thereby discourage uninformed or noise trading. Increased volume during periods of wider bid-ask spreads and an increase in seller-initiated trades are consistent with a bad news release and a price discovery explanation. We also partition the sample into two groups. The first group, an overnight sample, consists of firms with 75% or more of the price changes taking place from the previous days closing price to the opening price on the event day. The second group, the continuous sample, includes all other firm observations. Our results for the overnight sample show that sellerinitiated and buyer-initiated trades in the day depend on the direction of price movement. In general, after large price decreases, selling pressure increases with volume significantly higher early in the day. Then, late in the day, even though trading volume is similar to those in the prior days, selling pressure is significantly higher. However, after large price increases at the beginning of the day, increases in buying pressure are insignificant even though volume is significantly higher. At the end of the trading day, spreads, volume, and buying pressure are significantly higher than those in prior trading days. The rest of the paper is organized as follows. The next section of this paper describes the data and sample. Section 2 presents research procedures and test statistics. Section 3 discusses the results. The summary and conclusion are included in the final section. 2. Data and sample A sample of large price movements is constructed by screening for absolute daily returns in excess of 10% for all firms on the Center for Research of Security Prices (CRSP) tapes during years Next, the firms with large price movements are removed from the sample if they do not appear in the database with intraday trading data on the NYSE and AMEX Trades and Quotes Transaction File prepared by the Institute for the Study of Security Prices (ISSM). The ISSM tapes contain intraday time stamped trades and quotes. The trade transactions include the price and size of the transaction. Quote revisions contain bid and ask quote prices and support for each quote revision. We reorder the timing of quote revisions that precede trades by five seconds or less to restore the order of quotes and trades as they were realized. 1 Additionally, firm observations are screened for low value stocks. A ten-dollar share minimum price is used so that low-priced stocks are excluded from the sample. We further eliminate all firms with stock splits sixty days around the large price movement. We also limit the frequency of each firm in the sample to one so that we can assume independence in statistical analysis. The screening process leaves 1,351 firm observations, which include 461 large price decreases and 890 large price increases. The sample is partitioned by the timing of the price movement. An overnight period covers from the close on day 1 to the open on day zero. A continuous market period expands from the close on day 1 to the close on day zero. If 75% or more of the large price movement takes place in the overnight period, then the observation is also assigned to the overnight sub-sample. Otherwise the observation is assigned to the continuous sub-sample.

4 306 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) Table 1 Characteristics of large stock price changes and pre-event period trading measures N Mean S.D. Minimum Q1 Median Q3 Maximum Large price decreases Return DSP RSP VOL SP Large price increases Return DSP RSP VOL SP Notes: This table presents statistics of large positive and negative return distributions, and measures of trading activity during the pre-event ( 10 to 4 day) period. Large returns are measured from the closing price of Day 1 to that of Day 0. DSP: pre-event mean dollar spread in dollars. RSP: pre-event mean relative (percent) spread in percent. VOL: mean trading volume in round lot of shares in a 30-minute trading interval. SP: mean selling pressure in round lot of shares in a 30-minute trading interval. SP is the difference between trading volume at the bid price and volume at the ask price. For each measure, we report the mean, standard deviation (S.D.), minimum, first quartile (Q1), median, third quartile (Q3), and maximum. All tests for changes in the spreads, volume, and selling pressure are conducted for the full sample and the overnight sub-sample. In the full sample, the 461 price decrease observations have a mean event day return of 13.34% and a median return of 11.94%. The largest decrease is 37.88% and, by sample selection, the smallest decrease is 10%. The 890 large price increases have a mean of 13.58% and a median of 11.76%. The largest increase is 74.29% and the smallest increase is 10.0%. A summary of return distributions is presented in Table 1. We use two different measures of the bid-ask spread: dollar spread (DSP) and relative spread (RSP). Dollar spread is the difference between the lowest ask quote and the highest bid quote. Relative spread is dollar spread divided by the bid-ask midpoint. While dollar spreads are widely used in market microstructure studies, relative spreads are more appropriate for cross-sectional samples. We use both dollar and relative spreads to measure changes in cost of trading. We use two measures of trading activities in this study, trading volume and selling pressure. Trading volume (VOL) is measured by the number of round lot shares traded. Selling pressure (SP) is the difference in volume between seller-initiated trades and buyerinitiated trades. A trade below the bid-ask spread midpoint is assigned a positive selling pressure (for example a trade of 300 shares at the bid price is 3 round lots), a trade at the bid-ask midpoint is assigned a zero selling pressure, and a trade above the bid-ask spread midpoint is assigned a negative selling pressure (for example a trade of 500 shares at the ask price is 5 round lots). If there is a higher seller-initiated volume during a period of time, then

5 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) the selling pressure is positive. If there is more buying volume, then the selling pressure is negative. The selling pressure takes a zero value when there are balanced buying and selling. The rest of Table 1 presents the pre-event sample average, standard deviation, minimum, maximum, median, and first and third quartiles of the above four variables: dollar spreads, relative spreads, volume, and selling pressure. Statistics for firms with large price increases and decreases are reported separately. The average dollar spreads are around 27 cents per share in both samples. There is little difference in distributions of dollar spreads across the two samples. The average relative spreads are 1.29% and 1.65% for the large price decrease and large price increase samples, respectively. Average volume is lower for the large price increase sample. On average, for each thirty-minute trading period the volume is 665 round lots for firms in the large price decrease sample and 505 round lots for firms in the large price increase sample. In both samples, there is a slight indication of buying pressure (negative selling pressure) before large price changes. The average selling pressure in a thirty-minute trading interval is 17 and 20 round lots of shares for price decrease and increase sample, respectively. However, the medians are essentially zero for both samples, indicating that in the pre-event period there is little price pressure. 3. Procedures We use intraday data to examine both the impact of large price changes and duration of changes in trading activities as a result of a large price movement. We partition each trading day into thirty-minute trading periods and report results for each thirty-minute interval. The advantage of this partition is that we can analyze the changes in trading activity by comparing the pre-event and post-event measures in similar clock time thirty-minute windows. Thus, we effectively avoid the potential noise created by intraday trading patterns such as the reverse J of relative spreads reported by McInish and Wood (1992). We define the event day as day zero and the pre-event period as days 10 to 4. The event day is the day of the large price change measured from the closing price on day 1 to the closing price on day zero. Changes in each variable for each firm are the differences between the pre-event average and the event day measure. The duration of a change is determined by the number of consecutive thirty-minute trading periods that have sustained a significant change in cross-sectional means. We use both standard parametric t-test and non-parametric sign test to compute the significance of changes in variables. First, we compute cross-sectional means of changes in each of the four variables. Next, we use parametric t-test to calculate t-statistics corresponding to the changes in means. Finally, we use non-parametric sign test to report the significance of number of firms with changes in the right direction under our hypotheses. Following a large price increase (decrease), we expect dollar spreads to increase, relative spreads to decrease (increase), volume to increase, and selling pressure to decrease (increase). The non-parametric sign test utilizes a binomial distribution with the number of trials as the number of firms in the sample, and the probability of.50 as the probability of getting a change in the direction of our expectation for each individual firm. On dollar spreads, we predict that dollar spreads widen after large price changes regardless

6 308 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) of the direction. Increase in dollar spread is caused by the added asymmetric information. As the asymmetric information increases for an individual firm market makers increase dollar spreads (market makers increase the adverse selection component of the spread to compensate for trading losses to informed traders). We also expect relative spreads to increase after large price decreases as dollar spreads increase and prices fall. However, change in relative spread is less clear for the sample of large price increases. Dollar spreads increase and stock prices also increase. The relative spread, the ratio of the two, changes according to the percentage increases in the dollar spread and stock price. We expect volume to increase after large price movements as a result of additional information arriving at the markets. Following large stock price decreases (increases), we expect selling pressure to increase (decrease) as investors act upon the bad (good) news. Another measure of price pressure, quote depth, is not used in this study. The ISSM tapes contain the depth of the highest bid and lowest ask quotes. However, by inspection, it is not clear if the current depth of the lowest ask and highest bid is always updated to the tape. When additional limit orders are submitted for the same price as the current best quote (highest bid or lowest ask), the additional depth does not always appear to be posted to the tape. Thus the current depth of quotes are stale depth. As a result, we choose to use the number of buyer-initiated and seller-initiated trades to proxy price pressure. However, the depth of the quote around large price changes is addressed by Handa (1995.) Handa finds the depth of quotes declines sharply in response to large price changes. The decline in the depth of a quote is an indication that the level of information asymmetry is higher because traders commit to trading fewer number of shares at a given (quote) price. 4. Results In this section, we first discuss results related to the full sample. We examine dollar spreads and relative spreads following large price movements. We also examine the volume and selling pressure following these events. Secondly, we specifically study the reaction of firms in the overnight sample because firms in this sample experience large price changes when the market is closed. After the market opens on event day zero, the impact of large price changes should be strongest. Thus, we expect to detect clearer results Changes in bid-ask spreads Our empirical results suggest that dollar spreads temporarily increase following large price changes and that relative spreads increase (decrease) following large price decreases (increases). Table 2 summarizes changes in dollar and relative spreads by thirty-minute periods during the event day. For dollar spreads, we detect an average one-cent increase in the mean dollar spreads, regardless of the direction of the price change. For large price decreases, the change is more pronounced at the beginning and the end of the trading day. For large price increases, the changes are more significant during the afternoon hours of trading than in the morning hours. For relative spreads, we detect highly significant increases in the mean relative spreads after large price decreases. The results are consistent with our

7 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) Table 2 Average dollar spreads and relative spreads around large price changes: Full sample results TIME DSP K N RSP K N Large price decreases 9:30 a.m. to 10:00 a.m *** *** 202*** :00 a.m. to 10:30 a.m *** *** 239*** :30 a.m. to 11:00 a.m * *** 241*** :00 a.m. to 11:30 a.m * *** 249*** :30 a.m. to 12:00 noon ** *** 268*** :00 noon to 12:30 p.m ** *** 267*** :30 p.m. to 1:00 p.m * *** 247*** 392 1:00 p.m. to 1:30 p.m *** 238*** 384 1:30 p.m. to 2:00 p.m ** *** 253*** 392 2:00 p.m. to 2:30 p.m ** *** 273*** 404 2:30 p.m. to 3:00 p.m *** *** 277*** 414 3:00 p.m. to 3:30 p.m *** 247*** *** 306*** 420 3:30 p.m. to 4:00 p.m *** 242** *** 321*** 439 Average *** *** Large price increases 9:30 a.m. to 10:00 a.m * :00 a.m. to 10:30 a.m ** 332*** :30 a.m. to 11:00 a.m *** 310*** :00 a.m. to 11:30 a.m *** *** :30 a.m. to 12:00 noon *** 396** * :00 noon to 12:30 p.m *** *** 302*** :30 p.m. to 1:00 p.m *** *** 721 1:00 p.m. to 1:30 p.m *** 377* ** 706 1:30 p.m. to 2:00 p.m *** * 721 2:00 p.m. to 2:30 p.m *** 391** ** 731 2:30 p.m. to 3:00 p.m *** 404** * 336*** 745 3:00 p.m. to 3:30 p.m *** 415** * 350*** 772 3:30 p.m. to 4:00 p.m *** 472*** *** 360*** 810 Average *** *** Notes: This table reports average changes in bid-ask spreads within each 30-minute interval during the event day (Day 0) in the full sample. DSP: pre-event mean dollar spread in dollars. RSP: pre-event mean relative (percent) spread in percent. : changes in a variable during the event day (Day 0) from its pre-event period. K: number of successful events out of total number of trials (N) that are in favor of our hypothesis. *, **, ***: significant at the 0.05, 0.01, and levels. predictions because of the increased dollar spread (numerator) and the decreased stock price (denominator). In the case of large price increases, the mean relative spreads decreases with an average of.05% of stock price. Therefore, the dollar spread increase percentage (numerator) is smaller than the price increase percentage (denominator). However, a parametric t-test does not suggest much statistical significance. Table 2 contains pre-event averages of dollar spreads and relative spreads, changes in variables during the event day, total number of firm observations, and frequency of firms with changes in favor with our predictions. All statistics are recorded by each thirty-minute window and significance are indicated by the asterisk notation.

8 310 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) In the case of large price decrease, average dollar spread across thirteen thirty-minute trading windows is 27 cents per share. After the large price decrease, dollar spreads increase one cent per share on average. Parametric t-tests suggest that twelve out of thirteen changes are statistically significant at least at the.05 significance level. Non-parametric sign test utilizes a binomial distribution with the probability of increasing dollar spread being.50. The variable K indicates the frequency of firms experience increases in dollar spread. The variable N represents the total number of firm observations in a particular thirty-minute trading window. The significance for the non-parametric test is weaker than the parametric results. Only two out of thirteen trading periods are significant at the.05 level. Following large price decreases, average relative spreads increase dramatically from the pre-event mean of 1.26% to 1.49%, a change of.23%. The results across all trading periods are significant at the.001 level for both parametric and non-parametric tests. Following large price increases, average dollar spreads increase by one cent per share. Ten out of thirteen trading periods produce significant changes by parametric t-tests. Six of these are judged significant by non-parametric sign tests. The average relative spreads decrease by.05% after price increases. Six out of thirteen trading periods are significant using t-tests. All thirteen periods show significance by non-parametric sign tests Changes in volume and selling pressure Changes in volume and selling pressure following large price movements are also examined using the same set of statistical procedures. Both volume and selling pressure are measured in round lots of shares. Results are reported by each thirty-minute period. Volume increases significantly following large price changes. Selling pressure increases significantly after large price reductions and decreases significantly after large price increases. Table 3 summarizes results of volume and selling pressure changes. Table 3 reports pre-event averages of volume and selling pressure, average changes during the event day, and frequency of firms experiencing changes in favor of our hypothesis. The changes in average volume for the two different types of price changes are similar. On average, trading volume is significantly higher throughout the event day, suggesting that the market is more liquid after a price change. If increased volume suggests increased liquidity, then the asymmetric information market microstructure models predict a reduction in relative spreads as liquidity increases. While price increases weakly support this notion, price decreases cause higher volume but also higher relative spreads. After a large price reduction, trading volume increases by an average of 123 round lot of shares in each thirty-minute trading interval. Similarly, trading volume increase by an average of 173 round lot of shares following a price increase. A significant number of firms experience increased volume. Both parametric t-tests and non-parametric sign tests show high significance. However, the pattern is not as consistent for large price decreases as large price increases. Every thirty-minute trading period has significant average volume increases in the price increase sample but only six of the thirteen periods are significant in the price decrease sample. Although trading volume is a measure of liquidity, another measure is price pressure. If volume increases on one side (more buying with constant selling, or more selling with

9 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) Table 3 Average volume and selling pressure around large price changes: Full sample results TIME VOL K N SP K N Large price decreases 9:30 a.m. to 10:00 a.m *** 280*** ** 205*** :00 a.m. to 10:30 a.m * 246*** :30 a.m. to 11:00 a.m *** *** 228*** :00 a.m. to 11:30 a.m *** *** 239*** :30 a.m. to 12:00 noon *** *** 234*** :00 noon to 12:30 p.m *** ** :30 p.m. to 1:00 p.m ** 244*** * 221*** 392 1:00 p.m. to 1:30 p.m *** 244*** *** :30 p.m. to 2:00 p.m ** 244*** *** 224*** 392 2:00 p.m. to 2:30 p.m * 264*** *** 228*** 404 2:30 p.m. to 3:00 p.m *** *** 252*** 414 3:00 p.m. to 3:30 p.m *** *** 261*** 420 3:30 p.m. to 4:00 p.m *** *** 271*** 439 Average *** 19 91*** Large price increases 9:30 a.m. to 10:00 a.m *** 540*** * 293*** :00 a.m. to 10:30 a.m * 460*** * :30 a.m. to 11:00 a.m *** 472*** * 340*** :00 a.m. to 11:30 a.m *** 469*** *** :30 a.m. to 12:00 noon *** 483*** *** :00 noon to 12:30 p.m *** 451*** *** 296*** :30 p.m. to 1:00 p.m *** 459*** *** 721 1:00 p.m. to 1:30 p.m *** 459*** * 706 1:30 p.m. to 2:00 p.m *** 461*** :00 p.m. to 2:30 p.m *** 466*** ** 317*** 731 2:30 p.m. to 3:00 p.m *** 480*** *** :00 pm. to 3:30 p.m *** 497*** *** 318*** 772 3:30 p.m. to 4:00 p.m *** 517*** *** 333*** 810 Average *** 24 28*** Notes: This table reports average changes in trading volume and selling pressure within each 30-minute interval during the event day (Day 0) in the full sample. VOL: pre-event mean trading volume in round lot of shares. SP: difference between trading volume at bid price and volume at ask price. : changes in a variable during the event day (Day 0) from its pre-event period. K: number of successful events out of total number of trials (N) that are in favor of our hypothesis. *, **, ***: significant at the 0.05, 0.01, and levels. constant buying), then spreads should increase despite larger volume. Our proxy for price pressure is selling pressure (SP), which is volume of seller-initiated trades minus volume of buyer-initiated trades. In the pre-event period, days 10 to 4, the large price decrease sample has slight buying pressure (median SP is 4). The large price increase sample essentially has no price pressure (median SP is zero). Selling pressure increases significantly following a price reduction, with an average of 91 more round lots seller-initiated trades during day zero than previous trading days. The statistic suggests that after a large price reduction, net seller-initiated trades increase from the pre-event average of 19 to event day 72 round lots. Eleven out of thirteen trading periods produce significant changes. Following a large price increase, selling pressure decreases significantly, with an average of 28 more

10 312 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) round lots of buyer-initiated trades. Thus, following a price increase, selling pressure changes from 24 to 52, or an average of 52 more round lots are buyer-initiated than seller-initiated on day zero. A majority of changes in the thirteen trading periods are statistically significant at the.005 significance level. A curious part of the result is that price pressure is greatest at the end of the day in both price decrease and price increase samples. Part of this late day price pressure can be explained by the design of the study. Since firms realize large price movements throughout the trading hours of the event day, by the end of the day large price changes have been realized. Therefore the late day price pressure could be a build up of intraday price pressure. However, part of this late price pressure can also be derived from buy or sell decisions of portfolio managers. They increase trading late in the day to accommodate order flows to their portfolios. For example, a mutual fund manager receives buy and sell orders during the day and trades at the end of the day to re-balance the portfolio based on the net order flow. This late day trading is consistent with the increased selling pressure for price decreases and increased buying pressure for price increases. We also examine all transaction prices with respect to the contemporaneous quotes. Trades are classified as above the standing ask, at the ask, between the ask and the spread midpoint, at the midpoint, between the midpoint and the bid, at the bid, and below the bid. All transactions are also classified by volume. Trades are partitioned into size groups of 1 to 5 round lots, 6 to 10 round lots, 11 to 25 round lots, 26 to 50 round lots, 51 to 99 round lots, and 100 or more round lots. It is interesting to compare the volume at the bid before and after large price decreases. The volume at the bid averages 23% of daily transactions prior to the price decrease but increases to over 31% after a price decrease. Trading inside the spread falls following a large price decrease. Coupled with a larger spread following a price decrease, increased trading at the quotes may indicate less consensus in price and higher levels of asymmetric information. The results are consistent with the strategic agent microstructure models and the bad news story. Although volume significantly increases following large price increases, we do not detect any significant changes in the proportion of trading at or near the quote prices. The increased volume is consistent across all trading sizes and locations (with respect to the contemporaneous bid and ask quotes). This result is consistent with the good news story. Again, the difference in trading at the quotes and inside the spread following price increases and decreases indicates that the agents trading following price changes may behave differently for price increases and price decreases Results of overnight sample Perhaps the cleanest investigation of the data is afforded right after market opens following a large price movement in the overnight period. In the overnight sample, at least 75% of the price movement takes place when the market is closed. After the market opens all firm observations in this sample have just realized a large price movement. We expect the greatest changes in the first few thirty-minute trading periods following the opening. Tables 4 and 5 document results for the overnight sample and parallel the results presented in Tables 2 and 3. The results for the changes in dollar and relative spreads are more

11 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) Table 4 Average dollar spreads and relative spreads around large price changes: Overnight sample results TIME DSP K N RSP K N Large price decreases 9:30 a.m. to 10:00 a.m * ** 126* :00 a.m. to 10:30 a.m ** *** 117** :30 a.m. to 11:00 a.m *** 118* :00 a.m. to 11:30 a.m ** *** 128*** :30 a.m. to 12:00 noon *** 121*** *** 136*** :00 noon to 12:30 p.m *** *** 137*** :30 p.m. to 1:00 p.m *** *** 116*** 197 1:00 p.m. to 1:30 p.m *** 116*** 190 1:30 p.m. to 2:00 p.m *** *** 124*** 196 2:00 p.m. to 2:30 p.m * *** 134*** 209 2:30 p.m. to 3:00 p.m *** 124*** *** 145*** 211 3:00 p.m. to 3:30 p.m *** 141*** *** 169*** 219 3:30 p.m. to 4:00 p.m *** 146*** *** 181*** 229 Average *** *** Large price increases 9:30 a.m. to 10:00 a.m ** :00 a.m. to 10:30 a.m :30 a.m. to 11:00 a.m * :00 a.m. to 11:30 a.m *** * :30 a.m. to 12:00 noon *** 230*** :00 noon to 12:30 p.m *** 215*** :30 p.m. to 1:00 p.m *** 212*** :00 p.m. to 1:30 p.m *** 217*** :30 p.m. to 2:00 p.m *** 224*** :00 p.m. to 2:30 p.m *** 232*** ** :30 p.m. to 3:00 p.m *** 247*** :00 p.m. to 3:30 p.m *** 251*** :30 p.m. to 4:00 p.m *** 274*** Average *** * Notes: This table reports average changes in bid-ask spreads within each 30-minute interval during the event day (Day 0) in the overnight sample. DSP: pre-event mean dollar spread in dollars. RSP: pre-event mean relative (percent) spread in percent. : changes in a variable during the event day (Day 0) from its pre-event period. K: number of successful events out of total number of trials (N) that are in favor of our hypothesis. *, **, ***: significant at the 0.05, 0.01, and levels. pronounced in the overnight sample than the full sample. For the large price decrease sample, changes in dollar spreads are higher with an average of three cents per share. The changes are much higher at the end of the trading day than the beginning of the day. The same results hold for the large price increase sample. As expected, relative spreads increase in the price decrease sample but remain about unchanged in price increase sample. In summary, the overnight sample produces similar dollar spread results for both price increase and decrease samples. The results are consistent with the microstructure models that predict higher spreads to entice more informed trading. However, the dollar spreads are more pronounced later in the day suggesting higher selling or buying pressure late in the day.

12 314 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) Table 5 Average volume and selling pressure around large price changes: Overnight sample results TIME VOL K N SP K N Large price decreases 9:30 a.m. to 10:00 a.m *** 190*** *** 135*** :00 a.m. to 10:30 a.m *** 162*** * :30 a.m. to 11:00 a.m *** 154*** * 122** :00 a.m. to 11:30 a.m *** 163*** * 130*** :30 a.m. to 12:00 noon *** 161*** *** 127*** :00 noon to 12:30 p.m *** 152*** * :30 p.m. to 1:00 p.m *** 137*** ** 197 1:00 p.m. to 1:30 p.m *** 136*** :30 p.m. to 2:00 p.m *** 132*** * 123*** 196 2:00 p.m. to 2:30 p.m ** 147*** *** 127*** 209 2:30 p.m. to 3:00 p.m *** * 211 3:00 p.m. to 3:30 p.m *** *** 142*** 219 3:20 p.m. to 4:00 p.m *** 158*** 229 Average *** 22 81*** Large price increases 9:30 a.m. to 10:00 a.m *** 355*** *** :00 a.m. to 10:30 a.m *** 304*** ** :30 a.m. to 11:00 a.m *** 293*** * :00 a.m. to 11:30 a.m *** 271*** *** 168*** :30 a.m. to 12:00 noon *** 274*** *** :00 noon to 12:30 p.m *** 251*** *** 154*** :30 p.m. to 1:00 p.m *** 256*** *** 384 1:00 p.m. to 1:30 p.m *** 254*** *** 379 1:30 p.m. to 2:00 p.m * 265*** :00 p.m. to 2:30 p.m ** 256*** * 170*** 393 2:30 p.m. to 3:00 p.m * 279*** :00 p.m. to 3:30 p.m * 275*** *** 172*** 428 3:30 p.m. to 4:00 p.m ** 286*** *** 182*** 442 Average *** 21 34*** Notes: This table reports average changes in trading volume and selling pressure within each 30-minute interval during the event day (Day 0) in the overnight sample. VOL: pre-event mean trading volume in round lot of shares. SP: difference between trading volume at bid price and volume at ask price. : changes in a variable during the event day (Day 0) from its pre-event period. K: number of successful events out of total number of trials (N) that are in favor of our hypothesis. *, **, ***: significant at the 0.05, 0.01, and levels. The results on volume and selling pressure are interesting. Following large price decreases, volume is significantly higher early in the day but returns to normal by the end of the day. However, selling pressure is the greatest at the end of the day. This is consistent with the portfolio manager trading story as the daily order flows would have more sell orders and be executed late in the day. Following large price increase, volume increases significantly throughout the trading hours. The largest increases are in the early trading periods of the day. However, buying pressure is greatest late in the day. As the day progresses, more buy orders are processed in comparison to sell orders even though increase in volume is not as large. The results of the overnight sample point to a correlation between price pressure and

13 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) dollar spreads. For both large price increases and decreases, as the price pressure increases late in the day, dollar spreads increase in reaction to one side of trading. For example, we find that in the morning following a large price increase, dollar spreads do not significantly change despite higher volume. However this is also a period of trading when selling pressure has not significantly changed. In the morning following a large price decrease, dollar spreads have significantly increased but so have both volume and selling pressure. Yet, as the day progresses following a large price decrease, volume returns to its pre-event level but selling pressure continues to increase. The increase in selling pressure is manifested in higher dollar spreads late in the trading day. Similarly, dollar spreads are significantly higher later in the trading day following large price increases when buying pressure is highest and volume is lowest. These results are consistent with a good news, bad news reaction by traders. The results are also an institutional aspect of the market. When prices fall, a bad news event, mutual fund owners submit sell orders. When prices rise, a good news event, the mutual fund owners submit buy orders. The execution of the orders are not completed early in the day when the market is most liquid, but rather orders are executed late in the day when the accumulated order flow is known to the mutual fund manager. The dollar spreads change in reaction to one sided trading late in the day and not to the price change early in the day. 5. Summary and conclusions We examine changes in spreads, volume, and price pressure of firms experiencing large price movements. When prices increase, perhaps in response to good news, dollar spreads increase but not at the same rate as the price increase resulting in a fall in relative spreads. Volume is higher and buying pressure increases. When a large price decrease takes place, perhaps in response to bad news, both relative and dollar spreads increase and volume is higher. Selling pressure increases and is more pronounced as the day progresses. The widening of the dollar spread coupled with increased volume following large price changes is consistent with strategic agent models which predict strategic quote setting. Market makers temporarily increase the size of the spread to speed up price discovery during periods of higher information asymmetry. However, examining price changes that occur while the market is closed (the overnight sample) shows a delayed response in buying and selling pressure. Increased trading activity at the end of day is consistent with the institutionalization of the market. Mutual fund managers balance their portfolios based on the daily order flows and have discretion when to submit orders. Placing buy or sell orders late in the day when the net order flow is known is probably more efficient than trading throughout the day. Thus, the late day trading activity of mutual fund managers may influence both the dollar spreads and selling pressure late in the day. Notes 1. Based on Lee and Ready (1991), we move quote revisions that precede a trade by five seconds or less to follow the trade. Reordering quote data is important when classifying

14 316 R.M. Brooks et al. / The Quarterly Review of Economics and Finance 39 (1999) trades as buys or sells with contemporaneous bid and ask prices. Failing to reorder the quotes that are incorrectly recorded (time stamped ahead of a trade) can result in classification of buys as sells or vice versa. Recall, a trade above the bid-ask spread midpoint is classified as a buyer-initiated trade, and a trade below the bid-ask spread midpoint is classified as a seller-initiated trade. A second classification method using up-ticks and down-ticks (Lee and Ready, 1991) partitions trades into the same buy and sell categories as the method using bid-ask spread midpoints. References Admati, Anat and Paul Pfleiderer A Theory of Intraday Patterns: Volume and Price Variability, Review of Financial Studies 1:3 40. Admati, Anat and Paul Pfleiderer Divide and Conquer: A Theory of Intraday and Day of the Week Mean Effects, Review of Financial Studies 2: Fishe, Raymond, Thomas Gosnell, and Dennis Lasser Good News, Bad News, Volume and the Monday Effect, Journal of Business Finance and Accounting 20: Glosten, Lawrence Insider Trading, Liquidity, and the Role of the Monopolist Specialist, Journal Business 62: Glosten, Lawrence and Paul Milgrom Bid, Ask and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders, Journal of Financial Economics 14: Handa, Puneet Bid-Ask Dynamics During Large Price Adjustments: An Empirical Investigation, Working Paper, University of Iowa. Leach, Chris and Ananth Madhavan Price Experimentation and Market Structure, Review of Financial Studies 6: Kyle, Albert Continuous Auctions and Insider Trading, Econometrica 53: Lee, Charles and Mark Ready Inferring Trade Direction from Intraday Data, The Journal of Finance 46: McInish, Thomas and Robert Wood An Analysis of Intraday Patterns of Bid-Ask Spreads for NYSE Stocks, Journal of Finance 47: Patell, James and Mark Wolfson Good News, Bad News, and the Intraday Timing of Corporate Disclosures, The Accounting Review 57: Penman, Stephen The Distribution of Earnings News Over Time and Seasonalities in Aggregate Stock Returns, Journal of Financial Economics 18: