Trading Halts and Price Discovery

J Finan Serv Res (2006) 30: 311–328 DOI 10.1007/s10693-006-0421-x Trading Halts and Price Discovery Jeff Madura & Nivine Richie & Alan L. Tucker Received: 10 September 2004 / Revised: 21 November 2005 / Accepted: 26 July 2006 # Springer Science + Business Media, LLC 2006 Abstract Trading halts have their proponents and opponents. Recent literature has examined the benefits of halts, if any, by studying the consequences of halts on order flow and price volatility. This study complements existing literature by examining the consequences of trading halts on price discovery. Our results indicate that stock price adjustments surrounding trading halts are conditioned on the underlying event that caused the halt. The weighted price contributions of all subsamples are concentrated in the halt period and some subsamples show significant price contribution in the prehalt period as well. We find minimal evidence of price discovery continuing after the halt is removed. In addition, cross-sectional analysis shows that price discovery in the pre-halt and halt periods are more pronounced for larger firms and for firms with specific news events. Keywords Trading halts . price discovery . market efficiency . event study JEL Classifications G-14 1 Introduction Proponents of halts maintain that halts ensure a level playing field among investors when news that could materially affect the value of the stock is released. Halts give investors time to react to the news and to adjust their orders. For example, a halt may occur due to a temporary order imbalance resulting from the incomplete interpretation of new information disclosed about a stock. Proponents of trading halts argue that halts allow time for price and order imbalances to interact to occasion a new equilibrium stock price. Either new orders restore order balance at the prevailing J. Madura Florida Atlantic University, Boca Raton, FL, USA N. Richie (*) Sigmund Weis School of Business, Susquehanna University, 514 University Drive, Selinsgrove, PA 17870, USA e-mail: [email protected] A. L. Tucker Pace University, New York, NY, USA 312 J Finan Serv Res (2006) 30: 311–328 price, or the price adjusts and eliminates the imbalance. In some cases, new orders and a price adjustment interact to allow for a new equilibrium. Trading halts serve to stabilize markets by allowing market participants more time before responding to new information. Halts presumably can reduce potential overreaction that may be fueled by emotional and uninformed trading. They are not intended to protect market makers specifically, but to reduce the potential price instability that could be caused by any traders who might act too quickly on information. On the other hand, opponents of halts suggest that halts represent a temporary fix and may even be harmful because they delay the price from moving quickly and efficiently to its new equilibrium. Halts may be disruptive and may prevent investors from unwinding undesirable trading positions. Simply put, opponents suggest that halts merely Binterfere with the self-correcting forces that bring the market back to equilibrium^ (Wilmouth, 2001).1, 2 Recent research into the non-trading period associated with trading halts assesses the value of such halts by empirically examining the effects of halts on stock prices, order flow, and volatility. Specifically, Corwin and Lipson (2000) find that market and limit orders increase during trading halts implemented by the NYSE and that the opening price following the halt is a good predictor of future prices. Lee et al. (1994) find excess volatility following trading halts imposed by the NYSE. Christie et al. (2002) assess a sample of NASDAQ stocks and find that inside quoted spreads as well as stock volatility increase after trading halts when trading resumes on the same day. On the other hand, spread and volatility effects are attenuated when trading resumes at the open on the following day. Their results for halts when trading is resumed in the same day raise the question of whether halts stabilize market prices; if market prices were more stable, opening spreads should be narrower.3 Market microstructure literature has greatly improved our understanding of price discovery, the process by which information is reflected in security prices. Research by Cao et al. (2000) examines price discovery surrounding periods of non-trading and finds that significant price discovery occurs in the pre-opening session where overnight information is impounded into stock prices. Cao et al. (2000) extend the investigation of Greene and Watts (1996) into overnight earnings announcements.4 1 See, for example, the views of the late Merton Miller in BSmall Stock Focus: Proposed Halt in Trading Volatile Nasdaq Stocks is Criticized by Some Traders and Academics^, Wall Street Journal, 16 February 1999, p c7. 2 A similar debate exists regarding circuit breakers, that is, whether these breakers serve a useful economic function or merely serve to reduce market efficiency and delay preordained outcomes, c.f. Greenwald and Stein (1991) and. 3 Regarding foreign equity markets, Brailsford (1995) finds excess volatility following trading halts imposed in the Australian market. And Kryzanowski and Nemiroff (1998) find that volatility following trading halts imposed by the Montreal Stock Exchange is excessive, but subsides over the next two days. 4 Earlier studies of price formation surrounding trading halts find mixed results. Kryzanowski (1979) provides early documentation of abnormal returns surrounding Canadian trading halts and finds abnormal returns in the post-halt period for unfavorable news but not for favorable news. Hopewell and Schwartz (1978) examine abnormal returns surrounding NYSE trading halts but find no significant daily abnormal returns either during the halt period or during the post-halt period. Howe and Schlarbaum (1986) find significant negative abnormal returns following SECinitiated trading suspensions. Ferris et al. (1992) examine excess returns surrounding SEC mandated suspensions for a sample of AMEX and NYSE listed securities and find some reversal of residuals in the period shortly following the suspension. J Finan Serv Res (2006) 30: 311–328 313 Overall, research on halts indicates that (1) volume and volatility are high following halts, (2) bid–offer spreads typically remain high following halts; (3) traders reposition their trades during halts; (4) indicator quotes converge toward the reopening price (NYSE); (5) longer NASDAQ quotation periods dampen volatility effects; and (6) reopening prices are good indicators of future prices. Of course, a caveat here is that conclusions drawn from previous studies regarding the value of halts are necessarily limited because one does not know what would have happened in the absence of a halt. This caveat applies to the current study as well. Our study extends the literature on price discovery by using daily and intraday data to examine the price contribution of the trading halt period relative to total price discovery. Earlier studies are forced to use weekly return data (Kryzanowski, 1979; Howe and Schlarbaum, 1986) or have no access to the exact time and nature of each trading halt (Hopewell and Schwartz, 1978). Furthermore, exchangemandated trading halts differ substantially from SEC-mandated trading suspensions. Finally, we are able to employ more current methods of investigation, including a more refined measure of price discovery. Using data from NASDAQ, we assess 656 trading halts during 1998. We assess a pre-halt period to determine whether pricing behavior may have triggered the halt, and assess a halt period to determine how dealers in the stock respond to the trading halt. We also assess a post-halt period to determine how market participants respond to the trading halt. Our analysis is conducted on nine different subsamples that are separated by type of news, since the impact of a trading halt may vary with the type of news event. Finally, we assess whether the halt period or post-halt period pricing is associated with halt-specific characteristics or with firm-specific characteristics. Our results show significant abnormal returns in the halt period for the full sample and for anticipated events. Pre-halt periods find some abnormal returns while post-halt periods show no significant abnormal returns. An examination of the weighted price contribution in the different periods finds that the price discovery is concentrated in the halt period for all news events. There is some evidence that pre-halt periods experience significant, but lower, price contribution while post-halt periods show minimal or no price contribution. Our halt period contains the 5-min quotation period that exists when the stock is allowed to resume trading. Consequently, our results are consistent with the Barclay and Hendershott (2003) findings that significant price discovery is experienced in the pre-open trading session following the limited trading of the night session. Our results are not consistent with the Howe and Schlarbaum (1986) finding that abnormal returns remain significant in the weeks following an SEC suspension of trading. On balance, our results indicate that trading halts serve a useful purpose as the post-halt period contributes little, if at all, to the price discovery process. This study proceeds as follows: Section 2 describes the trading halt process. Section 3 presents the testable hypotheses and Section 4 describes the data. Section 5 presents the research design and results of price discovery analysis and Section 6 presents the cross sectional analysis. Section 7 provides a summary and conclusion. 2 Trading halt process The NASDAQ Stock Watch Department seeks to ensure that no investor is at a relative disadvantage in receiving information that could impact the valuation of a firm_s stock. The Stock Watch Department periodically implements trading halts 314 J Finan Serv Res (2006) 30: 311–328 that allow for news to be fully disclosed to all interested parties. While investors may continue to submit orders, trading is suspended until the halt is lifted. NASDAQ normally notifies a firm whose stock is subjected to a trading halt, although it has the right since April 1999 to impose the halt regardless of its ability to contact the firm. Many trading halts have duration of less than 30 min, but others can last several trading days. If trading is resumed during the trading day, there is a 5-min quotation period (pre-resumption) in which dealers establish their price quotations. Conversely, if the halt is lifted after the close then trading is resumed following a 90-min pre-opening quotation period the next trading day. 3 Testable hypotheses Market microstructure models suggest that traders are motivated by asymmetric information or by liquidity needs (see Kyle 1985). Since trading halts are designed to allow time for information to be disseminated to all participants, a test of the price discovery associated with the trading halt offers insight into the success or lack thereof of these market mechanisms. Barclay and Hendershott (2003) examine the role of private information in periods of non-trading, namely the pre-open and the post-close sessions on the NASDAQ market, and find more asymmetric information in the preopen session than in other periods. They find most total price discovery during the regular trading session and little price discovery in the post-close session. Their results motivate our investigation into the price discovery surrounding trading halts. 3.1 Pre-halt period Since a trading halt may be prompted by high asymmetric information among market participants, we expect the pre-halt trading period to contribute significantly to total price discovery. In cases where the information arrives as a surprise to market participants, we can expect limited asymmetric information whereas information that is seemingly anticipated by market participants faces greater asymmetric information. Consequently, pre-halt pricing is expected to vary with the type of news event associated with each firm subjected to a trading halt. Surprise announcements should be met with limited price discovery in the pre-halt period while anticipated announcements should be met with significant price discovery in the pre-halt period. 3.2 Halt period Barclay and Hendershott (2003) find that though the pre-open session experiences high price discovery per trade, the regular trading session experiences the greatest total price discovery. In fact, they find that asymmetric information declines over the course of the trading day. Since the trading halt period allows time for information to be more completely disseminated before trading resumes, it should contribute significantly to price discovery. Since the 5-min quotation period (or 90 min quotation period if overnight) is included in this halt period, we should see price discovery along the lines of that experienced in the pre-open session of the NASDAQ market. Furthermore, halts that come as a surprise to market participants should experience more pronounced price discovery during the halt period than would stocks where the information that prompted the halt was anticipated by the market. J Finan Serv Res (2006) 30: 311–328 315 3.3 Post-halt period We expect that most of the price discovery will be completed during the halt period due to the quotation period preceding the first transaction. To the extent that markets are not efficient in the semi-strong form, we should identify further price contribution during the post-halt period. The ability of dealers to properly price the halted stocks during the halt period may be conditioned on the type of news associated with each firm. Thus the contribution of the post-halt period to price discovery may vary by the type of news. 4 Data Our sample includes NASDAQ listed firms whose stocks experienced a trading halt during 1998. These data, as supplied by NASDAQ, include the date and time of each halt, the reason for each halt, the last trade price before the halt, and the first trade price after the halt.5 For stocks where trading is resumed the same day, these prices are the first transaction prices following the 5-min quotation period once trading is resumed. For stocks where trading is resumed the following day, these prices are the first transaction prices following the 90 min quotation period at the start of the next trading day. More recent data do not provide us with the detailed reasons for each halt. Rather, halts today are classified according to general codes such as Fnews pending_ or Fnews released_ but the specifics of the news that trigger the halts are no longer available. These intraday values are combined with the surrounding closing values from the Center for Research in Securities Prices (CRSP) daily files to arrive at pre-halt and post-halt returns. The sample returns are matched to intraday values on the NASDAQ composite index gathered from NASDAQ and are matched to the nearest 1-min increment. For the pre-halt and post-halt periods, the closing price of the NASDAQ composite index are extracted and used to calculate daily returns. The sample period of 1998 precedes extended session and night trading. After hours markets are likely to distort the estimation of abnormal returns and so we standardize the pre and post halt periods to daily frequency.6 Table 1 describes the sample by the type of news prompting the trading halt. Of the initial sample of 949 trading halts, we eliminate any stocks with transaction prices below $5. We further screen the sample to eliminate trading halts that last more than 24 h or less than 10 min. Our final sample includes 553 firms with a total of 656 trading halts. Of the screened sample, 631 halts are announced and lifted within the same calendar date while 25 are lifted the following calendar day. Many of the halts are either announced or lifted in the pre-open or post-close sessions. Of the 656 halts in the sample, 412 of them are announced after 4:00 P.M. but before 9:30 A.M. while 314 halts are lifted during these post close or pre-open hours. Of the total sample, 253 are both halted and resumed in the after hours sessions. The 5 We thank Timothy McCormick of NASDAQ for his assistance in providing the data and imparting the rules of the exchange as they pertain to trading halts. 6 Limiting the pre and post halt periods to a finite number of hours created estimation difficulties in cases where trading was halted or resumed at the beginning or end of the trading day. 316 J Finan Serv Res (2006) 30: 311–328 Table 1 Descriptive statistics. Average and median duration of trading halt in hours and the minimum and maximum values. Panel B describes duration of halt when trading resumes the same day vs when trading resumes the next day. Panel C describes duration of halt when the event is anticipated vs when the event is a surprise. An anticipated event has an average abnormal return prior to the event that is statistically significant while a surprise event is not statistically significant. Halt duration Median Panel A: full sample and by reason Fullsample (N=656) 0.92 Acquire (N=49) 1.10 Target (N=106) 1.06 Highearn (N=51) 0.78 Lowearn (N=257) 0.85 Opport (N=38) 1.08 Badnews (N=42) 0.96 Reorg (N=18) 0.86 Merger (N=42) 1.16 Buyback (N=11) 0.75 Other (N=42) 1.00 Panel B: subsamples Same day (N=631) 0.88 Next day (N=25) 15.93 Anticipated (N=447) 0.92 Surprise (N=209) 0.90 Favorable (N=224) 0.87 Unfavorable (N=348) 0.93 Mean Min Max 1.80 2.25 1.85 0.89 1.57 3.44 1.70 0.88 3.06 1.05 1.69 0.18 0.62 0.43 0.53 0.23 0.18 0.52 0.53 0.43 0.48 0.48 24.00 17.40 21.62 2.05 17.00 24.00 18.27 1.45 17.18 2.92 16.10 1.19 17.20 1.79 1.84 1.68 1.79 0.18 14.90 0.23 0.18 0.23 0.18 8.37 24.00 21.62 24.00 18.27 24.00 number of halts per firm range from a minimum of one to a maximum of four. For firms that experience more than one halt in 1998, the median number of halts is two. Table 1 shows that the median halt duration is 0.92 h while the mean is just 1.8 h. The length of the halt ranges from 0.18 h to a maximum of 24 h. Based on NASDAQ_s stated reason, we classify each event according to the following codes: & & & & & & & & & & Acquire: Firm intends to engage in an acquisition, Target: Firm is a target of an acquisition, Highearn: Mention of an unexpectedly high earnings report, Lowearn: Mention of an unexpectedly low earnings report, Opport: Firm is pursuing an opportunity (such as a new drug), Badnews: Firm is announcing operational difficulties, Reorg: Firm is announcing reorganization or workforce restructuring, Merger: Firm will merge without the distinction of being the target or acquirer, Buyback: Firm is announcing the repurchase of its equity, Other: Missing and/or miscellaneous announcements such as delays in earnings reports, unknown abrupt stock price movement, or issuance of new equity. The single largest subsample is low earnings (257 trading halts), while the smallest subsample is for buybacks (11 halts). We partition our sample by whether the reason is anticipated by market participants or whether the halt comes as a surprise. We define a surprise event as one where the cumulative abnormal returns J Finan Serv Res (2006) 30: 311–328 317 described in the next section are significant in the days immediately preceding the trading halt. Using this categorization method, we identify the anticipated trading halts as those associated with mergers, targets, bad news, and low earnings. All other categories come as a surprise to market participants. We further partition our sample by whether the news event is favorable or unfavorable to prevent the results from being clouded when positive and negative events are considered jointly. Accordingly, we include Target, Highearn, Opport, Reorg, and Buyback events as favorable. The Acquire events are classified as unfavorable since research has found a negative market response to acquisitions, on average. Lowearn and Badnews events are also classified as unfavorable events. 5 Price discovery surrounding trading halts To investigate price discovery surrounding trading halts, we begin with the abnormal returns in the three intervals described above. We then examine the contribution of each period to the total price discovery. 5.1 Measurement of abnormal returns We examine the pricing behavior of stocks subjected to trading halts over three separate intervals surrounding the halts. First, a pre-halt period extends from the market close the previous day until the time at which the halt is imposed. This period should be subject to a high level of information asymmetry, which may trigger the decision to impose a trading halt. Second, the trading halt period begins when the trading halt is imposed and ends when trading is resumed. The price can change over the halt period because dealers may adjust opening prices at the time trading resumes. If price discovery exists during the trading halt period, the abnormal return of the stock should be different from zero. Third, the post-halt period extends from the resumption of trading until the close on that trading day. The post-halt period is assessed separately from the halt period because it accounts for investor reaction to the trading halt. If a signal is emitted by the trading halt that is not fully accounted for in the opening price established by dealers at the resumption of trading, said signal should be evident within the post-halt period. Since the signal from a halt may be conditioned on the underlying news that prompts the trading halt, we partition the sample into nine categories by news type. For each category, we follow the price discovery process over these three intervals. Trading is halted and resumed at the discretion of the exchange and, consequently, each halt window is unique. We employ the following methodology to standardize all periods to daily frequencies. To isolate the effect of the halt itself, we capture the percent change in price from the time when the halt is imposed to the time when the halt is removed, and use this interval as our event window. The percent change in price from the time when the halt is removed until the following closing value then becomes our post-event window of t+1. All subsequent post-event windows (i.e., t+2 through t+3) follow the traditional definition of return as percent change in price from one day_s closing value to the next. We apply the same definitions to our pre-event windows where we define the tj1 window as the percent change in price from the previous closing price to the time when the halt is imposed. All other pre-event windows follow the traditional calculations of percent change in price from one day_s closing price to the next. AR (j3) Panel A: full sample Full sample (N=656) Favorable events (N=224) Unfavorable events(N=348) Same Day (N=631) Next day (N=25) Merger (N=42) Target (N=106) Badnews (N=42) AR (j1) AR (0) AR (+1) AR (+2) AR (+3) 0.0025 (0.731) 44:56% 0.0152 (1.65)* 49:51% j0.0063 (j2.72)*** 38:62% 0.0002 (0.106) 44:56% 0.0610 (0.798) 44:56% j0.0038 (j1.902)* 47:53% 0.0043 (1.34) 52:48% j0.0103 (j3.73)*** 43:57% j0.0034 (j1.668)* 48:52% j0.0141 (j1.319) 40:60% j0.0081 (j2.331)** 41:59% 0.0158 (2.71)*** 50:50% j0.0258 (j6.26)*** 33:67% j0.0072 (j2.114)** 41:59% j0.0301 (j1.014) 48:52% j0.0398 (j5.719)*** 42:58% 0.0684 (6.51)*** 79:21% j0.1224 (j16.50)*** 14:86% j0.0357 (j5.082)*** 42:58% j0.1446 (j3.617)*** 24:76% 0.0028 (0.575) 49:51% j0.0027 (j0.45) 47:53% 0.0074 (1.03) 51:49% 0.0004 (0.078) 48:52% 0.0631 (1.341) 64:36% j0.0037 (j1.626) 46:54% j0.0096 (j2.70)*** 38:62% j0.000026 (j0.01) 51:49% j0.0043 (j1.814)* 45:55% 0.0096 (0.912) 56:44% j0.0042 (j2.11)** 43:57% j0.0042 (j1.55) 42:58% j0.0033 (j1.07) 46:54% j0.0043 (j2.138)** 43:57% j0.0012 (j0.104) 52:48% j0.0031 (j1.357) 40:60% 0.0030 (0.396) 55:45% 0.0119 (2.148)** 50:50% j0.0098 (j1.894)* 38:62% j0.0040 (j1.631) 47:53% 0.0123 (1.523) 60:40% 0.0124 (2.641)*** 58:42% j0.0113 (j1.358) 36:64% j0.0102 (j2.595)*** 39:61% 0.0174 (1.781)* 57:43% 0.0336 (3.697)*** 52:48% j0.0434 (j2.997)*** 19:81% j0.0676 (j7.86)*** 30:70% 0.0411 (1.821)* 64:36% 0.1092 (6.538)*** 84:16% j0.1198 (j6.345)*** 10:90% 0.0060 (0.986) 51:49% 0.0064 (0.219) 45:55% j0.0007 (j0.073) 48:52% j0.0124 (j0.71) 36:64% j0.0019 (j0.678) 50:50% j0.0033 (j0.446) 45:55% j0.0018 (j0.417) 42:58% j0.0112 (j1.116) 45:55% j0.0030 (j1.176) 43:57% j0.0027 (j0.463) 33:67% j0.0055 (j1.637) 38:62% j0.0078 (j1.234) 45:55% J Finan Serv Res (2006) 30: 311–328 Panel B: anticipated events Total anticipated events (N=447) AR (j2) 318 Table 2 Average abnormal returns. Average abnormal returns estimated using market adjusted model. Abnormal returns (AR) presented with cross-sectional tstatistics in parentheses. Proportion positive: proportion negative shown in italics. Panel A presents ARs for the full sample as well as ARs for halts that resume trading on the same calendar date (same day) vs halts that resume trading the following trading day (next day). Panel B shows the ARs for anticipated events and the respective subcategories while Panel C shows the ARs for surprise events and the respective subcategories. Panel C: surprise events Total surprise events(N=209) Reorg (N=18) Opport(N=38) Other(N=42) Acquire(N=49) buyback (N=11) Highearn(N=51) j0.0092 (j3.318)*** 33:67% j0.0122 (j3.857)*** 42:58% j0.0273 (j6.221)*** 34:66% 0.0145 (1.505) 53:47% 0.0078 (0.661) 44:56% 0.0460 (0.917) 39:61% 0.0076 (1.058) 50:50% 0.0118 (2.001)* 65:35% j0.0081 (j0.353) 55:45% 0.0066 (0.773) 55:45% j0.0034 (j0.979) 47:53% j0.0160 (j1.116) 44:56% j0.0050 (j0.783) 39:61% j0.0096 (j1.056) 45:55% 0.0010 (0.136) 51:49% 0.0006 (0.085) 45:55% 0.0023 (0.344) 53:47% j0.0036 (j0.519) 46:54% j0.0067 (j0.541) 39:61% 0.0180 (1.093) 58:42% j0.0141 (j0.586) 45:55% j0.0029 (j0.227) 39:61% 0.0018 (0.118) 45:55% j0.0118 (j1.223) 47:53% j0.1498 (j18.1)*** 6:94% 0.0197 (1.837)* 66:34% j0.0267 (j1.298) 39:61% 0.0455 (2.099)** 84:16% j0.0138 (j0.364) 50:50% 0.0191 (1.574) 57:43% 0.0621 (2.205)* 91:9% 0.0358 (1.707)* 76:24% 0.0117 (1.436) 55:45% j0.0042 (j0.548) 46:54% 0.0141 (0.839) 56:44% j0.0151 (j0.969) 37:63% j0.0102 (j0.708) 52:48% 0.0019 (0.081) 41:59% j0.0164 (j0.933) 45:55% j0.0007 (j0.058) 49:51% j0.0001 (j0.028) 54:46% j0.0012 (j0.306) 47:53% j0.0077 (j1.846)* 38:62% j0.0178 (j1.233) 56:44% j0.0185 (j1.959)* 34:66% j0.0038 (j0.553) 50:50% 0.0100 (0.954) 37:63% j0.0280 (j2.174)* 36:64% j0.0121 (j1.394) 25:75% j0.0067 (j2.244)** 43:57% 0.0123 (0.797) 39:61% j0.0089 (j1.134) 39:61% j0.0124 (j1.939)* 38:62% j0.0107 (j1.94)* 41:59% 0.0024 (0.286) 45:55% j0.0052 (j1.011) 51:49% J Finan Serv Res (2006) 30: 311–328 Lowearn(N=257) ***Statistical significance at 10% level **Statistical significance at 5% level *Statistical significance at 1% level 319 320 J Finan Serv Res (2006) 30: 311–328 We use a market-adjusted model of Brown and Warner (1985) 7 and following the methodology employed by Lee et al.(1994) in their analysis of NYSE trading halts. We estimate our abnormal return in each interval with the abnormal return (ARj,t) given as: ARj;t ¼ Rj;t  Rm;t ð1Þ where Rj,t=the interval return for firm j over interval t and Rm,t=the corresponding return for the NASDAQ composite index over interval t. We calculate the market interval returns by matching intraday values of each sample firm_s pre-halt and post-halt value to the intraday values of the NASDAQ composite index to the nearest minute. After hours values for the NASDAQ composite index are not available as early as 1998, and yet many of our sample halts are either announced or lifted outside of normal trading hours (i.e., after 4:00 P.M. or before 9:30 A.M.). To reconcile pre-halt and post-halt index values for after hours announcements, we use the following criteria. If the halt is announced after 4:00 P.M. but before midnight, then the index pre-halt value is the closing value that same day. If the halt is announced after midnight but before 9:30 A.M., then the index pre-halt value is the closing value the previous day. If trading is resumed between midnight and 9:30 A.M., then the post-halt index value is the opening value that same day. If trading is resumed between 4:00 P.M. and midnight, then the post-halt index value is the opening value the next morning. In this manner we are able to match the event interval for each sample firm with an event interval for the NASDAQ composite index. For the pre- and post-event index intervals, we follow the traditional computation of return as percent change in the index from one day_s close to the next.8 For a number of N firms, the interval average abnormal return, (ARt), is: ARt ¼ N 1X ARj;t : N j¼1 ð2Þ Following Brown and Warner (1985), the cross-sectional t-statistic is computed as: N P  CARi N i¼1 : t¼ N  P CARi pffiffiffiffi N i¼1 ð3Þ 5.2 Results from computing abnormal returns Table 2 presents the event study results for the full sample and for the various subgroups. Panel A shows that the full sample experiences significant negative 7 Following Ferris et al. (1992), a market model estimation of abnormal returns is estimated using the CRSP equally weighted index as well as the NASDAQ composite index as proxies for the market. The results are qualitatively similar but due to the intraday nature of the trading halts, the results are not reported here. 8 Due to the irregular event window lengths in our intraday data, the assumption of identically distributed residuals is violated. To address this issue, we follow the Boehmer et al. (1991) procedure to standardize the abnormal return and the cumulative abnormal return by the estimation period variance and arrive at a standardized cross-sectional test statistic. The results are qualitatively similar and are therefore not reported. J Finan Serv Res (2006) 30: 311–328 321 abnormal return during the halt period and that reaction is pronounced for the small group of stocks that resume trading the following trading day. There is modest evidence of significant negative effects in the pre- and post-halt periods, but the magnitude of the market response is clearly most pronounced in the halt period. The subsamples of favorable events and unfavorable events as well as the subsamples of same day and next day halts show similar results. These results provide preliminary support for our hypotheses that pre-halt and halt periods should contribute to price discovery while post halt prices provide limited price discovery. Panel B of table 2 shows that anticipated events experience significant abnormal returns in the pre-halt period and these abnormal returns are consistent with they type of news being disseminated. That is, firms whose stocks were halted for reasons of Merger or Target related announcements show a positive abnormal return in both the pre-halt and halt periods. On the other hand, firms whose stocks were halted for reasons of Badnews or Lowearn related announcements show negative abnormal returns in both the pre-halt and halt periods. In all cases, the post-halt period does not exhibit any significant abnormal returns; again, this result is consistent with our hypotheses. Panel C of table 2 shows the event study results for surprise events. These abnormal returns are not significant in the pre-halt period and show only minor evidence of significance during the halt period. In the post-halt period, there is limited price contribution. The event study results provide preliminary support for our hypotheses. We more closely investigate the price contribution associated with each halt interval. 5.3 Weighted price contribution Following the methodology used by Barclay and Warner (1993); Cao et al. (2000), and Barclay and Hendershott (2003), we identify the cumulative price change in each time interval using weighted price contribution (WPC). The WPC is the contribution to price discovery associated with each of the three intervals surrounding the trading halt and is estimated as: 1   n B  C  DP  X i;j B DPj  C WPC ¼ C BN  A @ P DP j i¼1 DPj  0 ð4Þ j¼1 where iZ{prehalt, halt, posthalt} and DPij = the change in price over interval i for stock j. The first term in parentheses is the weight while the second term is the proportion of the total price change associated with the specific interval. The cross-sectional mean is tested for significance and results are reported in table 3. Table 3 shows the weighted price contribution for the full sample (panel A) and for the sample of stocks that halt and resume trading within the same day (panel B). In all cases, the majority of the price discovery occurs in the halt period. For the full sample, approximately 80% of price discovery occurs in the halt period with 15% occurring in the pre-halt period and very little price discovery occurring in the post 322 J Finan Serv Res (2006) 30: 311–328 Table 3 Weighted price contribution. The weighted price contributions (WPC) for the pre-halt, halt, and post-halt periods are given for both the full sample and for the sample of firms where trading was halted and resumed on the same calendar date. The WPC is cross-sectionally averaged across the sample. Sample Panel A: full sample WPC Full sample Favorable Unfavorable Anticipated Merger Target Badnews Lowearn Surprise Reorg Opport Other Acquire Buyback Highearn Panel B: same day halts Full sample Favorable Unfavorable Anticipated Merger Target Badnews Lowearn Surprise Reorg Opport Other Acquire Buyback Highearn Pre-halt Halt Post-halt 0.1503* 0.1262* 0.1525* 0.1295* 0.0411 0.1366* 0.2898* 0.1003* 0.2111 0.0357 0.1946* 0.4663 0.2977 0.1122 0.0895 0.7924* 0.7466* 0.7981* 0.8063* 0.5524* 0.8161* 0.5809* 0.8893* 0.7516* 0.6032* 0.6134* 1.4167* 0.5058* 0.8761* 0.6837* 0.0573 0.1273* 0.0494 0.0642 0.4064* 0.0472 0.1293 0.0104 0.0372 0.3611 0.192 j0.8829 0.1965 0.0117 0.2268 0.1372* 0.1331* 0.1249* 0.1266 0.0268 0.1445* 0.2611* 0.1014* 0.1677 0.0357 0.2336* 0.4821 0.0725 0.1122 0.0895 0.7877* 0.7383* 0.7914* 0.8016 0.5657* 0.8075* 0.6141* 0.8722* 0.7473 0.6032* 0.5565* 1.4364* 0.4948* 0.8761* 0.6837* 0.0752 0.1286 0.0837 0.0718 0.4075* 0.048 0.1248 0.0265 0.085 0.3611 0.2099 j0.9185 0.4327* 0.0117 0.2268 *Indicates statistical significance at 5% level halt period. Anticipated events experience 13% of price discovery in the pre-halt period, which is statistically different than zero. Surprise events experience 21% of price discovery during the pre-halt period, although that mean is not statistically significant at conventional levels. These results are consistent with those of Barclay and Hendershott (2003), who find that significant price discovery exists in the pre-open period following a limited night trading session. The accumulation of information in the non-trading session is then reflected in the price as soon as trading begins and is followed by reduced price contribution as the day progresses. Similarly, our results show that the non-trading session (i.e. the halt period) which includes a 5-min quotation period experiences a large part of the price discovery, despite the limited ability to trade. As trading J Finan Serv Res (2006) 30: 311–328 323 enters the post-halt period, most of the price contribution is eliminated and in only a few cases (merger related announcements) is there any lingering price discovery in the post-halt period. Our results are not consistent with Howe and Schlarbaum_s (1986) finding that price discovery occurs well beyond the SEC suspension period. Though they find significant abnormal returns in periods prior to the suspensions, they do not find significant abnormal returns in the week immediately preceding the suspension. They do, however, find significant price discovery both during and following the SEC suspension, on average, which contradicts market efficiency. The difference in results between the two studies is at least partially attributed to the sample selection. The SEC mandated suspensions are regulatory in nature and address such issues as fraud or other illicit behavior, while exchange-mandated trading halts are imposed to correct temporary trading imbalances or allow for further news dissemination. 6 Cross-sectional analysis To investigate further how price discovery may vary according to the type of news that is pending at the time of the trading halt, we perform a cross-sectional analysis of halt period abnormal returns and repeat the analysis using the weighted price contributions surrounding the trading halts. 6.1 Cross-sectional model This cross-sectional analysis is conducted using dummy variables for each of nine subsamples, defined by the news that existed at the time of the halt. The analysis allows us to determine whether the cross-sectional differences in price discovery, if any, during the halt period are conditioned on firm-specific or halt-specific characteristics. Howe and Schlarbaum (1986) find that longer SEC suspensions are correlated with greater stock price devaluation so we take the duration of the trading halt as endogenously determined. Consequently we conduct a weighted least squares regression where the weights are the length of time that the halt remains outstanding. The analysis is repeated using only those firms where trading was halted and resumed on the same day. The dependent variable (abnormal return during the halt or the weighted price contribution during the halt period) is hypothesized to be affected by the following characteristics: & & Trading volume. The trading volume of a stock serves as a measure of the stock_s liquidity. Since more liquid stocks may be more closely monitored by the market, a stock that has more liquidity is expected to be shocked less by a halt signal. This implies that more market participants in liquid stocks would rely less on a trading halt as a signal. Thus, price discovery is expected to be less pronounced for stocks that have relatively large trading volumes. Market capitalization. Larger stocks are more closely monitored than smaller stocks and, thus, are expected to be shocked less by a halt signal. Market capitalization of the firm serves as a measure of the degree of attention paid to it by investors. While this proxy is similar to the trading volume in that it attempts to measure the degree of market monitoring, it may serve as a more or less 324 & & J Finan Serv Res (2006) 30: 311–328 reliable proxy. Trading volume and market capitalization serve as imperfect substitutes. Surprise dummy. Based on an examination of the abnormal returns in the periods preceding the trading halt, we identify the following categories as surprise events: Merger, Target, Badnews, and Lowearn. We include a dummy variable to indicate surprise events in order to test our hypothesis that anticipated events experience less information asymmetry and consequently should experience less price discovery during the halt period. Also, surprise events experience relatively more information asymmetry and thus should witness greater price discovery during the halt period. Category dummies. A dummy variable is designated for each of the categories partitioned by the event that triggered the trading halt, with the category of Other serving as the default. We test the null hypothesis that the coefficients on all the dummy variables are equal. We further test the null hypothesis that the Acquiredum coefficient is equal to the Targetdum coefficient as well as the null that the Highearndum coefficient is equal to the Lowearndum coefficient. 6.2 Results of cross-sectional analysis Table 4 shows the weighted least squares regression results of the halt period abnormal return on the firm specific variables, the surprise dummy variable, and the category dummies. The results show that halt period abnormal returns are higher for surprise events and for target events as well as for merger-related events. For all other sub-categories, the halt period abnormal returns show no relation or mixed results with category dummies. These results provide preliminary support for greater price discovery for surprise events during the halt period. The tests on the dummy variables show that the Acquiredum and Targetdum coefficients are significantly different from one another while the Highearndum and Lowearndum are not. Table 5 shows the weighted least squares regression results of the pre-halt period weighted price contribution (panel A), the halt period weighted price contribution (panel B), and the post-halt period price contribution (panel C). In all cases, the regressions using the full sample have better explanatory power as seen in the higher adjusted R2 and F-statistic than the regressions using same day halts only. Consequently, we will limit the remaining discussion to the full sample results. The results offer three key findings. First, for the full sample, pre-halt price discovery and halt-period price discovery are directly related to size, indicating that larger firms experience greater price discovery in these periods than do smaller firms. Following the resumption of trading, price discovery is inversely related to size, indicating that the relationship reverses after a trading halt relative to the period prior to a trading halt. Second, the Surprise dummy coefficient is negative and significant in the pre-halt period and the post-halt period, but is not statistically significant in the halt period itself. Thus, surprise events experience less price discovery in the pre-halt and posthalt periods. This is consistent with our hypothesis that surprise events will have their price discovery concentrated in the halt period. Third, the category dummy variables are typically not significant for the most part during the halt period. Target, Lowearn, and Merger dummies are negative and significant in the pre-halt period, indicating lower pre-halt price discovery for these J Finan Serv Res (2006) 30: 311–328 325 Table 4 Weighted-least-squares analysis of abnormal returns during halt period. WLS regression of market adjusted abnormal return on Logvol (ln of trading volume), Logsize (ln of market cap), and dummy variables representing different reasons given by Nasdaq for the trading halt and a dummy variable indicating whether the halt was a surprise or anticipated. The weights are the length of time that trading was halted. Same day halts includes firms where trading is halted and resumed on the same calendar date. N is the number of non-missing observations used in the analysis and the sample period covers halts announced by the Nasdaq during 1998. Full sample (N=644) Estimate Intercept Logvol Logsize Acquiredum Targetdum Highearndum Lowearndum Opportdum Reorgdum Mergerdum Buybackdum Surprise Adj-R2 F value acquiredum=targetdum highearndum=lowearndum t-statistics Same day halts (N=619) Estimate t-statistics j0.0570 (j0.84) j0.0244 (j0.39) 0.0057 (1.60) j0.0008 (j0.25) j0.0130 (j2.24)** j0.0078 (j1.39) j0.0143 (j0.41) 0.0248 (0.84) 0.2228 (7.01)*** 0.2132 (8.32)*** 0.0232 (0.55) 0.0448 (1.41) j0.0454 (j1.57) j0.0184 (j0.76) j0.0860 (j2.63)*** 0.0764 (2.58)** j0.0489 (j0.74) j0.0313 (j0.65) 0.1331 (3.93)*** 0.2085 (6.47)*** 0.0194 (0.27) 0.0362 (0.69) 0.1736 (4.64)*** 0.1332 (4.34)*** 0.2903 0.3302 24.92 28.69 Tests on dummy variables (p value) <0.0001 <0.0001 0.1787 0.1130 *Statistical significance at 10% level **Statistical significance at 5% level ***Statistical significance at 1% level three categories. Target and Lowearn are positive and significant in the post-halt period, indicating greater price discovery after the resumption of trading. This finding suggests that the relative timing of price discovery for stocks subjected to trading halts varies by type of information. 7 Summary and conclusions Our study examines the value of trading halts by focusing on the price discovery process of stocks subjected to halts. We measure abnormal returns and weighted price contribution just before the halt, during the halt, and after the halt to determine the relative contribution to price discovery among these three periods. Our principal result is that most of the price discovery occurs while trading is halted. There is modest evidence of price discovery in the pre-halt period for selected subsamples, but even for these subsamples, the price discovery is dominant in the halt period. For those subsamples in which there is price discovery in the prehalt period, there is price discovery in the same direction during the halt period. This suggests that the pre-halt price adjustments move toward but do not reach 326 J Finan Serv Res (2006) 30: 311–328 Table 5 Weighted-least-squares analysis of weighted price contribution. WLS regression of weighted price contribution (WPC) on Logvol (ln of trading volume), Logsize (ln of market cap), and dummy variables representing different reasons given by Nasdaq for the trading halt and a dummy variable indicating whether the halt was a surprise or anticipated. The weights are the length of time that trading was halted. Same day halts includes firms where trading is halted and resumed on the same calendar date. N is the number of non-missing observations used in the analysis and the sample period covers halts announced by the Nasdaq during 1998. Panel A: pre-halt weighted price contribution Full sample Intercept Logvol Logsize Surprise Acquiredum Targetdum HighearnDum Lowearndum Opportdum ReorgDum MergerDum BuybackDum Adj-R2 F value acquiredum=targetdum highearndum=lowearndum Estimate t-statistics j0.0052 0.0000 0.0006 j0.0016 0.0014 j0.0016 j0.0005 j0.0018 j0.0002 j0.0006 j0.0017 0.0002 0.3148 27.52 (j8.31)*** (j1.07) (11.66)*** (j4.63)*** (4.34)*** (j5.30)*** (j1.19) (j6.69)*** (j0.76) (j1.05) (j5.33)*** (0.32) Same day halts Estimate t-statistics 0.00034 (0.97) 0.00013 (6.96)*** j0.00008 (j2.50)** j0.00056 (j3.17)*** j0.00021 (j1.21) j0.00057 (j3.83)*** j0.00011 (j0.58) j0.00062 (j4.43)*** j0.00006 (j0.33) j0.00024 (j0.86) j0.00079 (j4.29)*** j0.00001 (j0.04) 0.1098 7.84 Tests on dummy variables (p value) <0.0001 0.1094 0.0053 0.0272 Panel B: halt period weighted price contribution Full sample (N=636) Same day halts (N=611) Estimate t-statistics Estimate t-statistics Intercept j0.0026 (j2.71)*** j0.0022 (j2.13)** Logvol 0.0000 (j0.91) j0.0001 (j2.17)** Logsize 0.0004 (j5.00)*** 0.0004 (4.60)*** Surprise j0.0004 (j0.84) j0.0002 (j0.33) Acquiredum j0.0011 (j2.25)** j0.0010 (j1.98)** Targetdum j0.0004 (j0.98) j0.0003 (j0.77) HighearnDum j0.0005 (j0.82) j0.0005 (j0.93) Lowearndum 0.0001 (0.24) j0.0002 (j0.47) Opportdum 0.0001 (0.17) j0.0008 (j1.69) ReorgDum j0.0010 (j1.12) j0.0011 (j1.38) MergerDum j0.0013 (j2.67)*** j0.0009 (j1.62) BuybackDum j0.0005 (j0.48) j0.0006 (j0.69) 0.0819 0.0377 Adj-R2 F value 6.15 3.17 Tests on dummy variables (p value) acquiredum=targetdum 0.3100 0.3189 highearndum=lowearndum 0.4100 0.6421 J Finan Serv Res (2006) 30: 311–328 327 Table 5 (continued). Panel C: post-halt period weighted price contribution Full sample (N=636) Same day halts (N=611) Estimate t-statistics Estimate t-statistics Intercept 0.0064 (7.97)*** j0.0004 (j0.54) Logvol 0.0003 (6.01)*** 0.0001 (2.06)** Logsize j0.0007 (j10.68)*** 0.0000 (j0.57) Surprise j0.0014 (j3.01)*** j0.0010 (j2.34)** Acquiredum j0.0003 (j0.71) 0.0013 (3.31)*** Targetdum j0.0005 (j1.24) 0.0001 (0.28) HighearnDum 0.0016 (3.20)*** 0.0013 (3.09)*** Lowearndum j0.0005 (j1.52) 0.0000 (j0.02) Opportdum 0.0008 (2.00)** 0.0008 (2.14)** ReorgDum 0.0015 (1.91)* 0.0011 (1.79)* MergerDum 0.0000 (j0.02) 0.0004 (0.99) BuybackDum 0.0007 (0.84) 0.0010 (1.41) 0.2005 0.0133 Adj-R2 F value 15.47 1.75 Tests on dummy variables (p value) acquiredum=targetdum 0.7553 0.0204 highearndum=lowearndum 0.0005 0.0132 *Statistical significance at 10% level **Statistical significance at 5% level ***Statistical significance at 1% level equilibrium. The mean abnormal returns for most subsamples are not significant in the pre-halt period. The price discovery process is completed by the end of the halt period. There is no evidence of a reversal in the halt or post-halt period, which implies that any price adjustments do not reflect corrections to market overreaction. This evidence is corroborated by the measurement of the relative contributions of price discovery among the three periods surrounding the halt. The contribution to price discovery during the halt period is significant among the subsamples of halts partitioned by the events that triggered the halts. Most of the weighted price contribution occurs in the halt period, which includes the 5-min quotation period prior to the resumption of trading. For anticipated news events, pre-halt price discovery is significant while for surprise news events, pre-halt price discovery is not significant. The results from cross-sectional analyses of abnormal returns and weighted price contributions vary among subsamples. Abnormal returns during trading halts are more pronounced for anticipated events and for their respective subsamples. Weighted price contributions before and during the halts are directly related to firm size. Pre-halt weighted price contributions are lower for surprise events and for some specific news categories. Overall, the price discovery is concentrated in the halt period, but the timing of the price discovery varies by the type of information. Acknowledgments The authors thank Haluk Unal (JFSR editor), an anonymous referee of JFSR, and Shane Corwin, University of Notre Dame for helpful comments. 328 J Finan Serv Res (2006) 30: 311–328 References Barclay, M., and J. Warner. BStealth Trading and Volatility: Which Trades Move Prices?^ Journal of Financial Economics, 34 (1993), 281–305. Barclay, M., and T. Hendershott. BPrice Discovery and Trading After Hours.^ Review of Financial Studies, 16 (2003), 1041–1073. Boehmer, E., J. Musumeci, and A. Poulsen. BEvent-study Methodology Under Conditions of Eventinduced Variance.^ Journal of Financial Economics, 30 (1991), 253–272. Brailsford, T. BMarket Closures and Time-Varying Volatility in the Australian Equity Market.^ Journal of Empirical Finance, 2 (1995), 165–172. Brown, S., and J. Warner. BUsing Daily Stock Returns: The Case of Event Studies.^ Journal of Financial Economics, 14 (1985), 3–31. Cao, C., E. Ghysels, and F. Hatheway. BPrice Discovery Without Trading: Evidence From the Nasdaq Preopening.^ Journal of Finance 3 (2000), 1339–1365. Christie, W., S. Corwin, and J. Harris. BThe Impact of Market Mechanisms on Prices, Trading Activity, and Execution.^Journal of Finance 57 (2002), 1443–1478. Corwin, S., and M. Lipson. BOrder Flow and Liquidity Around NYSE Trading Halts.^ Journal of Finance 55 (2000), 1771–1805. Ferris, S., R. Kumar, and G. Wolfe. BThe Effect of SEC-ordered Suspensions on Returns, Volatility, and Trading Volume. ^Financial Review 27 (1992), 1–34. Greenwald, B., and J. Stein. BTransactional Risk, Market Crashes, and the Role of Circuit Breakers.^ Journal of Business 64 (1991), 443–462. Greene, J., and S. Watts. BPrice Discovery on the NYSE and the NASDAQ: The Case of Overnight and Daytime News Releases.^ Financial Management 25 (1996), 19–42. Hopewell, M., and A. Schwartz, Jr. BTemporary Trading Suspensions in Individual NYSE Securities.^ Journal of Finance 33 (1978), 1355–1373. Howe, J., and G. Schlarbaum. BSEC Trading Suspensions: Empirical Evidence.^ Journal of Financial and Quantitative Analysis 21 (1986), 323–333. Kryzanowski, L. BThe Efficacy of Trading Suspensions: A Regulatory Action Designed to Prevent the Exploitation of Monopoly Information.^ Journal of Finance 34 (1979), 1187–1200. Kryzanowski, L., and H. Nemiroff. BPrice Discovery Around Trading Halts on the Montreal Stock Exchange using Trade-by-trade Data.^ Financial Review 34 (1998), 195–212. Kyle, A. BContinuous Auctions and Insider Trading.^ Econometrica 53 (1985), 1315–1336. Lee, C., M. Ready, and P. Seguin. BVolume, Volatility, and New York Stock Exchange Trading Halts.^ Journal of Finance 49 (1994), 183–214. Wilmouth, R. BRemarks Given to Standing Committee 2, IOSCO Consultative Committee, Regarding Trading Halts, Rome, Italy on 26 October, 2001, National Futures Association.^ (2001), available online at http://www.nfa.futures.org/news/newsTestimony.asp?ArticleID=289.