Trading Volume, Information Asymmetry, and Timing Information Joon Chae∗ Massachusetts Institute of Technology First Draft: July, 2001 Current version: November 27, 2002
ABSTRACT This paper investigates trading volume before scheduled and unscheduled corporate announcements to explore how traders respond to private information. I show that cumulative trading volume decreases by more than 15% prior to scheduled announcements. The decline in trading volume is largest when information asymmetry is high, while the opposite relation holds for volume after the announcement. In contrast, trading volume before unscheduled announcements increases dramatically and shows little relation to proxies for information asymmetry. All the results for scheduled announcements are consistent with asymmetric information theories, where discretionary liquidity traders (DLTs) decrease volume when they know there is much adverse selection. However, DLTs do not seem to read information embedded in prices before unscheduled announcements. I further investigate the behavior of market makers and find that they act appropriately by increasing price sensitivity before all announcements. This implies that market makers extract timing information from their order books.
JEL classification: G14 Keywords: Trading volume; Timing information; Information asymmetry ∗
Joon Chae (
[email protected]) is from the Sloan School of Management at MIT, 50 Memorial Drive, Cambridge, MA 02139, USA. Telephone number: 617-233-6660; Fax: 617-452-2199. I thank Andrew Lo, Ken French, S.P. Kothari, Jonathan Lewellen, Dimitri Vayanos, Jiang Wang, and workshop participants at MIT for helpful comments and suggestions. I also thank Albert Wang for detailed suggestions that greatly improved the exposition.
1
Trading Volume, Information Asymmetry, and Timing Information
ABSTRACT This paper investigates trading volume before scheduled and unscheduled corporate announcements to explore how traders respond to private information. I show that cumulative trading volume decreases by more than 15% prior to scheduled announcements. The decline in trading volume is largest when information asymmetry is high, while the opposite relation holds for volume after the announcement. In contrast, trading volume before unscheduled announcements increases dramatically and shows little relation to proxies for information asymmetry. All the results for scheduled announcements are consistent with asymmetric information theories, where discretionary liquidity traders (DLTs) decrease volume when they know there is much adverse selection. However, DLTs do not seem to read information embedded in prices before unscheduled announcements. I further investigate the behavior of market makers and find that they act appropriately by increasing price sensitivity before all announcements. This implies that market makers extract timing information from their order books.
2
Trading volume plays a critical role in financial markets. It facilitates the price discovery process, enables investors to share financial risks, and ensures that corporations can raise funds needed for investment. Trading volume is generally characterized as either informed or uninformed (liquidity trading). Although both types of trades are studied in the finance literature,1 their relative importance remains poorly understood. In this paper, I study trading volume prior to important corporate announcements, when information asymmetry is likely to be the greatest. The analysis provides insights into how informed and liquidity traders, along with market makers, respond to asymmetric information in the market. Finance theory provides ambiguous predictions about trading volume prior to corporate announcements. When liquidity trading is exogenous and inelastic to price, as in Kyle (1985), trading volume increases in information asymmetry. This is because informed traders attempt to exploit their private information. However, if liquidity traders have timing discretion, as in Admati and Pfleiderer (1988) or Foster and Viswanathan (1990), trading volume can decrease in information asymmetry. In these models, when discretionary liquidity traders (DLTs) receive exogenous trade demands prior to announcements, they will postpone trading until the announcement is made and the information asymmetry is resolved. Therefore, total volume can decrease before announcements and correspondingly increase after. While theoretical models offer ambiguous predictions, little empirical evidence exists to distinguish between the predictions. In this article, I offer such evidence. I consider trading volume before two types of announcements: with and without timing information. Timing information refers to the public availability of when an announcement will be issued, and I designate these as scheduled announcements. When timing information is available, DLTs know that a large flow of information will be released on a specific date. Although they do not know what the information is, they can still optimize the timing of their trades to minimize adverse selection costs related to the information release. When 1
See Karpoff (1987) or Lo and Wang (2000) for general references.
3
timing information is not available, DLTs might not change the timing of their trades. I designate these as unscheduled announcements. In this article, I use earnings announcements as scheduled announcements. Earnings announcements involve a release of information in which the timing is publicly known. Even though “pre” earnings announcements have become increasingly popular, the actual earnings announcement still resolves much uncertainty related to stock pricing. For unscheduled announcements, I use acquisition, target, and Moody’s bond rating announcements. Neither the timing nor the magnitude and direction of these announcements are public information. I choose these four types of announcements because they represent major corporate events that have substantial impacts on prices.2 The empirical analysis explores four related hypotheses to study trading dynamics around these announcements. Hypothesis 1 states that trading volume should decrease before scheduled announcements.
The test compares abnormal trading volume before
scheduled and unscheduled announcements, and the main results confirm Hypothesis 1. Cumulative trading volume decreases more than 15% before scheduled announcements but steadily increases before unscheduled announcements. This implies that DLTs do change their behavior depending on the existence of timing information. Hypothesis 2 states that in a cross-section of stocks, the decrease in trading volume before scheduled announcements should be correlated with the extent of information asymmetry. Moreover, no such relation needs to hold before unscheduled announcements. I test and confirm Hypothesis 2 by regressing abnormal trading volume before each type of announcement on commonly used proxies for information asymmetry (size, number of analysts, bid-ask spread, and industry dummies). In Hypothesis 3, I further investigate the relation between information asymmetry and volume, but after scheduled announcements instead of before. George et al. (1994) propose a 2
See Bamber(1987), Foster, Olsen and Shevlin (1984), Jensen and Ruback(1983), Jarrell and Poulsen (1989a), and Hand, Holthausen, and Leftwich(1992), etc.
4
theoretical argument that high trading volume immediately after corporate announcements is a result of increased liquidity trading. Once the announcement has been made, DLTs perceive lower adverse selection costs and are finally willing to submit their trade demands. Thus, Hypothesis 3 states that information asymmetry before scheduled announcements should be positively correlated with abnormal trading volume after the announcements. I find that the hypothesis holds for most specifications. Hypothesis 4 states that market makers should increase price sensitivity only before scheduled announcements. Market makers are uninformed with respect to corporate information, so they should increase their price sensitivity when they perceive adverse selection costs. To test this, I estimate the Kyle (1985) lambda for each stock during different periods around the announcement. These tests confirm that price sensitivity increases before scheduled announcements. However, I find that price sensitivity increases even before unscheduled announcements. This suggests that market makers are able to extract and react to information related to timing. The rest of the paper is organized as follows. In Section I, I formally state the four Hypotheses. In Section II, I describe the datasets and test the main (first) hypothesis. In Section III, I test the remaining 3 follow-on hypotheses. In Section IV, I suggest possible implications of the results, and finally in Section V, I offer concluding remarks.
I
Hypotheses
Based on various event studies about corporate announcements, (e.g., earnings or takeover announcements), I assert that these announcements release a considerable amount of information and therefore spur large price changes. For example, the magnitude of the daily price changes on an earnings announcement is about 56% higher than the average daily price change during the announcement month.3 Therefore, immediately before these an3 45% more price change exists on acquisition announcement days, 287% on target announcements, and 18% on Moody’s bond rating change days.
5
nouncements there is likely severe information asymmetry between informed and uninformed investors. As shown in Milgrom and Stokey (1982), Black (1986), and Wang (1994), if there is a high possibility of trading with an informed counterparty, then uninformed traders will not participate in the market. In an extreme case with only one informed investor and one uninformed investor, the informed investor wants to trade the stock before the information is revealed. However, this is almost impossible since the uninformed investor will not trade without urgent liquidity needs. Therefore, it is a reasonable hypothesis that trading volume decreases before announcements. However, a necessary condition for this hypothesis to work is that the uninformed investors should know or infer the timing of the impending information issuance. Because uninformed investors’ participation in the market depends on the expected adverse selection cost, there should be a reasonably high level of informed trading to justify decreasing trading volume before an announcement.4 If informed trading and uninformed liquidity trading are always concurrent in the market, uninformed traders have difficulty in determining information trades from liquidity trades. If there is a scheduled announcement (such as an earnings announcement), uninformed investors can easily predict a high probability of informed trading before the announcement. Based on these arguments, trading volume before scheduled announcements should decrease. On the other hand, trading volume before unscheduled announcements depends on uninformed investors’ imperfect inference about information asymmetry. Therefore, trading volume before unscheduled announcements should be higher than trading volume before scheduled announcements. Hypothesis 1 Before an information-revealing announcement, trading volume should decrease only if uninformed investors expect the announcement. According to Hypothesis 1, uninformed investors should avoid trading when they perceive adverse selection. The amount of adverse selection costs depends on the extent of 4
As an effort to estimate the probability of informed trading, see Easley, Hvidkjaer and O’Hara (2002).
6
ex ante information asymmetry. Therefore, in the cross-section, trading volume before scheduled announcements should be negatively correlated with levels of ex ante information asymmetry. Such relation should not hold before unscheduled announcements unless they can infer the upcoming announcements. Hypothesis 2 Trading volume before scheduled announcements is negatively correlated with levels of ex ante information asymmetry. If the above hypothesis is true and liquidity traders must eventually fulfill their trading needs, then the decrease in trading volume represents the amount of delayed trading. Thus, after information asymmetry is resolved, DLTs should not only resume “normal” trading but also make up for delayed trading. This phenomenon will cause an increase in expost trading volume, which should be cross-sectionally correlated with ex-ante information asymmetry. In other words, high ex-ante information asymmetry causes a large amount of delayed trading, and thus ex-post trading volume should be positively related to ex-ante information asymmetry. This prediction is based on Goerge et al. (1994). They develop a theoretical model in which high trading volume on announcement days can be explained by the relatively lower adverse selection costs immediately after information announcements. Hypothesis 3 If adverse selection costs drive trading volume down before announcements, there should be a corresponding increase in trading volume after those announcements. Admati and Pfleiderer (1988) and Foster and Viswanathan (1990) argue that DLTs try to select trading periods when price sensitivity to order flow is the lowest. This implies that when uninformed investors avoid trading (before scheduled announcements), price sensitivity to order flow should be higher. Hypothesis 4 Price sensitivity to order flow before scheduled announcements will be higher than during “average” periods. 7
II
Tests
A. Data and description of variables This research uses four types of corporate announcements: earnings, acquisition, target, and Moody’s bond rating change announcements. Since I could not have analyzed all corporate announcements, I selected from those with well documented effects on return and trading volume, as in Bamber (1987), Bamber and Cheon (1995), Foster et al. (1984), and Hand et al. (1992). As Chordia et al. (2001) mention, an earnings announcement is one of the best candidates for scheduled announcements involving a release of pricing information. Therefore, I use earnings announcements to proxy for scheduled announcements. Among other corporate announcements, acquisition and target announcements have as much effect on the return and trading volume as earnings announcements. For the most part, I use acquisition and target announcements to proxy for unscheduled announcements. I also use a set of Moody’s bond rating change announcements for unscheduled announcements. Unlike the other announcements investigated in this article, Moody’s ratings are issued outside of the firm. This may imply that insider trading over this information is less than in the other cases. As mentioned before, these few types of announcements cannot represent every corporate announcement. The particular characteristics of these four announcements might affect the results. Therefore, I also randomly select “events” from a sample of days with large absolute returns.5 These random observations are meant to proxy for all other announcements, and should mitigate the problems inherent in using only four different kinds of announcements in the analysis. I/B/E/S data from 1986 to 2000 are used for the earnings announcement sample. From the I/B/E/S summary files, the number of analysts for each stock and reporting dates are 5
The exact procedure for this data will be stated in the following section.
8
extracted. The total number of earnings announcements during this period is 65,912, all from NYSE and AMEX companies. Acquisition and target announcements for NYSE and AMEX stocks are collected from the SDC database compiled by Thomson Financial Securities Data. To match the I/B/E/S period, only the data between 1986 and 2000 are included in the study. The total numbers of acquisition and target announcements are 25,087 and 12,485, respectively. The Moody’s bond rating change data are hand-collected from Moody’s investor service. Because of time constraints in manually recording the data, this data set spans a much shorter period, January 1997 to October 1998. However, with 410 announcements, (all from NYSE and AMEX) the sample size is large enough to provide an event study analysis. Center for Research in Security Prices (CRSP) daily data for all companies in NYSE or AMEX from 1986 to 2000 are combined with these four samples to obtain trading volume. Only observations that have at least 40 days of trading volume before and at least 10 days of trading volume after announcements are included in the sample. To control for firm-specific characteristics and increase the power of the tests, I match the companies’ earnings and takeover data in the robustness check section. After I select the companies existing in both earnings announcement sample and acquisition announcement sample, there remain 55,747 earnings announcements and 19,888 acquisition announcements. For companies included in both the earnings and the target announcement data sets, there are 51,130 earnings announcements and 8,430 target announcements. Since the Moody’s sample is too small, I do not match this data with other data sets. The number of observations after each filter is given in Table I. Insert Table I here Since there are various measures of trading volume, I choose one commonly accepted standardized measure. Trading volume is affected by the number of outstanding shares, so this research uses turnover instead. Turnover is defined as trading volume divided by 9
outstanding shares.6 Table II provides summary statistics of daily return, turnover, and log(turnover) from companies existing in the data set for at least 1 year between 1985 and 2000. The turnover distribution has a very fat tail and an extreme positive skewness, as shown in Table II. Insert Table II here Compared with the cross-sectional average of skewness and kurtosis of returns (1.049 and 23.407 respectively from 1986 to 2000), those of turnover are 8.596 and 159.439 respectively. The extreme skewness and kurtosis clearly show a break from normality, an assumption required for the statistical procedures of this research. Therefore, as Ajinkya and Jain (1989) argue, log turnover is appropriate in this analysis.7 Once a logarithm is applied, the skewness and the kurtosis decrease to −0.098 and 0.701. The measure of trading volume in this study is, therefore, log(turnover) as shown in Equation 1. Ã
Log T urnover(τi,t ) = Log
T rading V olumei,t Outstandingi,t
!
Abnormal T urnover(ξi,t ) = τi,t − τ¯i
Pt=−11
where, τ¯t =
(1)
(2)
t=−40 τi,t
30
The difference between log turnover during the test period and the estimation period is our measure of abnormal trading volume near announcements, as in Equation 2.
B. Empirical results Using the variables in the previous section, I obtain the cross-sectional average of abnormal turnover (ξi,t ) from t=-10 to 10 in Table III. The level of abnormal turnover before 6 7
See Lo and Wang (2000) for a systematic description of different measures of trading volume. In the robustness check section of this paper, the ordinary turnover is used to provide similar results.
10
earnings announcements is much smaller than that before other announcements. Trading volume before earnings announcements actually decreases significantly in the period from t=-10 to t=-3 with less than a 1% p-value, as Table III shows. On average, I find a decrease of around 1 to 3% of daily trading volume prior to each earnings announcement. Pt=−3
For a summary measure, I construct the average of abnormal turnover (
t=−10 ξi,t /8)
in
the period from t=-10 to t=-3 and observe that the negative turnover is large (more than 15% cumulatively) with t-statistic of −8.96.8 This implies that low turnover in any day is significant, but more importantly, a continuous streak of low turnover days in this period is extraordinary. Insert Table III here I compare the results from scheduled earnings announcements with the results from the three other types of unscheduled announcements. This comparison will indicate that timing information is critical in trading volume dynamics before announcements. The time series patterns of turnover before unscheduled announcements exhibits clear differences from those before earnings announcements. I observe significantly positive abnormal trading volume prior to these three types of unscheduled announcements instead of negative abnormal trading volume.9 When comparing acquisition, target, and Moody’s announcements to earnings announcements, the decreased trading volume before earnings announcements becomes even more conspicuous. Insert Figure 1 here To summarize, I provide a plot of cumulative abnormal percentage turnover in the period from t=-15 to t=15. Cumulative abnormal trading volume is drawn using average 8 The number of days (8) used in this summary measure does not affect the relative size of turnover before earnings announcements compared to the turnover before other announcements. 9 For an explanation of this positive trading volume, see Sanders and Zdanowicz (1992); Jarrell and Poulsen (1989b), etc.
11
log turnover in t=-45 to t=-16 as a benchmark (Fig. 1). In this plot, I intentionally use different estimation and testing periods to assure the robustness of the original results. The turnover before an earnings announcement decreases more than 15% cumulatively for around 12 consecutive days (Fig. 1(a)). However, for acquisition (Fig. 1(b)), target (Fig. 1(c)), and Moody’s announcements (Fig. 1(d)), I do not observe a decrease in turnover; instead, turnover increases considerably. To confirm that this measure of the abnormal turnover is unbiased, I randomly select days and companies and do the same procedure as for the announcements. I carefully select days such that the estimation and the event window do not include any of four announcements. The average abnormal turnover for these benchmark points is very near zero for the period (Fig. 1(e)). The flat line in Fig. 1(e) confirms that the measure of abnormal turnover is unbiased. Because the four different announcements (earnings, acquisition, target, and Moody’s announcements) do not represent all corporate events, I implement a supplemental analysis with randomly selected “events,” which are simply days with high absolute return. Large price changes are often associated with information releases, so these randomly selected “events” represent other announcements. The results of this analysis reinforce the unique nature of trading volume decreasing prior to earnings announcements. The following is the procedure to choose high absolute return “event”. First, I form 5 strata according to the size of companies in the NYSE and AMEX. 10,000 observations with a minimum absolute return are randomly selected from each strata. For example, I select a random sample of 10,000 data points in which absolute return is larger than 1% and market capitalization falls in the bottom 20%. I repeat this procedure for the other 4 size groups. The number of observations from all 5 strata totals 50,000. Without size-stratified sampling, the randomly chosen sample would be biased to include smaller and more volatile
12
companies.10 The size-stratified sampling ensures that the sampling is not biased. With the 1% minimum absolute return filter, the average of absolute return from the random sample is 2.82% on t=0. Any earnings, acquisition, target, and Moody’s announcements are excluded around those randomly selected days (i.e., in t=-40 to t=+10) before the random selection procedure. Finally, with the same procedure used to produce Table III, I obtain the average of abnormal turnover from t=-10 to t=-3. Insert Figure 2 here As shown in Fig. 2, none of the average abnormal trading volumes in t=-10 to -3 across different values of absolute return on t=0 have statistically significant negative values. The minimum absolute returns from left to right are 0.2%, 0.4%, 0.6%, 0.8%, 1.0%, 2.0%, 3.0%, 4.0%, 5.0%, 7.0%, and 10.0% for the 11 bars in Fig. 2. Note that around 2.66% (the mean of absolute returns on earnings announcement days11 ), only positive abnormal trading volume is observed. For example, with average absolute return on t=0 of 2.64%, the abnormal turnover between t=-10 and t=-3 is around 0.50% with t-statistic of 2.20. Therefore, these results show that abnormally high trading volume is observed before large price change “events.” This provides additional evidence that the decreasing trading volume before earnings announcements is unique and that timing information before earnings announcements has an important role in trading volume dynamics.
C. Robustness check For the preceding analyses, the estimation window is the period from t=-40 to t=-11 (Table III). The first robustness check uses estimation windows of different lengths. I attach a summary table in Panel A of Table IV using a longer estimation window from 10
Without size stratification, a sample contains 43.03%, 19.33%, 14.69%, 12.54%, and 10.41% of points from the smallest to largest quintiles of firms, though in the earnings announcement sample, the percentages are 18.34%, 20.51%, 20.57%, 20.12%, and 20.46%. 11 The average of absolute return on acquisition announcement days is 2.43%, that on target announcement days 5.53%, and that on Moody’s announcement days 2.72%
13
t=-55 to t=-11.12 In this table, the main result is even stronger, with cumulative abnormal turnover decreasing more than 18% over 8 days. The results are robust to tests with various estimation windows. A second robustness check uses raw turnover instead of log turnover. In Panel B of Table IV, I use median abnormal turnover and a bootstrapped distribution for inference. Using the median value of raw turnover between t=-40 and t=-11 as a benchmark, abnormal turnover in a testing period from t=-10 to t=+10 is estimated. I report the cross-sectional median of the abnormal turnover. For the inference, I use a non-parametric bootstrapped distribution.13 Once again, trading volume decreases significantly only before earnings announcements. Insert Table IV here In an event study, abnormality of a variable (e.g., return or volume) is usually measured in one of two ways. The first is using a fixed mean from an estimation period and the second is using a market factor model. To investigate the robustness across methods of defining abnormal turnover, I use an one-factor market model. From the estimation period (t=-40 to t=-11), the coefficient of a value-weighted log turnover index is estimated and applied to obtain abnormal turnover in the testing period. The reported values are the differences between realized log turnover and estimated normal log turnover, as in Equation 3. As in the other robustness checks, a statistically significant decrease in turnover is observed only before earnings announcements; the cumulative decrease is over 30% between t=-10 and t=-3.
Abnormal T unrover(ξi,t ) = τi,t − α ˆ i − βˆi τm,t 12
(3)
For brevity, I do not report the results from different estimation windows, e.g., with the estimation window of one full year of trading days, the results are similar. 13 Distribution for the cross-sectional median is constructed by generating 1,000 random samples with the same observation number as in corresponding announcement data. See Efron and Tibshirani (1983) for a non-parametric bootstrap procedure.
14
To check the stability of this specific pattern through time, I provide subperiod results in Panel D of Table IV.14 Again, I observe decreasing turnover only before earnings announcements. In the earliest of three sub-periods (1986-1990), the significance level is 5%. In the other two subperiods, abnormal turnover decreases by more than 15% and is much more significant. Since I use several databases, there might be unobservable company characteristics specific to each one. To control for firm-specific characteristics of each database, I create two matched samples, one matching earnings and acquisition companies and another matching earnings and target companies. For example, the earnings and acquisition matched sample includes only companies that exist in both the I/B/E/S and the SDC acquisition databases. The results from the matched samples show an even stronger pattern of decreasing trading volume before earnings announcements (Panel E of Table IV). From the unmatched I/B/E/S database, I showed that cumulative abnormal turnover from t=-10 to t=-3 is −15.06%. For the matched sample, the results strengthen slightly to −15.32% (acquisition matched) or −15.10% (target matched). As verified with this series of robustness checks, we can safely conclude that trading volume does decrease significantly prior to a scheduled earnings announcements.
III
Further empirical analysis
In this section, Hypotheses 2, 3, and 4 will be tested. First, testing Hypothesis 2 will provide the cross-sectional relation between trading volume and information asymmetry. Next, testing Hypothesis 3 will verify the theoretical predictions of George et al. (1994) and reconfirm the main result of this paper. Finally, testing Hypothesis 4 will reveal price sensitivity to order flow. 14
Because of the limited number of Moody’s announcements, I cannot use these data in Panels D and E.
15
A. Information asymmetry and trading volume before announcements I now turn to analyzing the cause of decreasing trading volume prior to scheduled earnings announcements. The intuition from most theoretical models predicts that greater information asymmetry results in less trading. I regress abnormal trading volume on information asymmetry proxies and control variables. Abnormally low trading volume is strongest between t=-10 and t=-3, so I define the dependent variable as the cumulative abnormal trading volume over this period. Though I obtain a similar relation between trading volume and information asymmetry proxies using different periods (e.g., from t=-10 to t=-2), the analysis of this period most clearly illustrates this relation. In a later section, analyses on trading volume from t=-2 to t=+2 and from t=+3 to t=+10 will complement the analysis in this section. To account for clustered data and obtain a more conservative econometric result, I implement Fama/MacBeth (1973) type regressions. Since earnings announcements are quarterly, each cross-sectional Fama/MacBeth type regression is done by quarter. However, the numbers of observations in different quarters are not similar enough, so the usual procedure of using simple time series averages of the coefficients will provide inefficient estimators. Therefore, we use weighted averages of the coefficients in quarterly cross-sectional regressions (Table V). The weight used is the number of observations in each quarter. The regression specification for each quarterly cross-sectional regression is:
ξi,q = αq + βq · Inf oAsymi,q + γq · Risk0i,q + δq · P rcChgi,q + εi,q
(4)
The term ξi,q is the average daily abnormal trading volume between t=-10 and t=3 at quarter q for company i, Inf oAsymi,q is each proxy for information asymmetry at quarter q for company i, Risk0i,q is a control variable for risk on the announcement day, and P rcChgi,q is another control variable defined as the absolute value of the average return between t=-10 and t=-3.
16
Since I cannot directly measure information asymmetry, I use several proxies, including size, analyst coverage, average bid-ask spread, and industry dummies. Though there is no standard measure of information asymmetry, these variables are widely used in the finance literature and have intuitive economic relations with information asymmetry.15 In this study, size is defined as the logarithm of the market capitalization of each company, analyst coverage as the number of analysts for each company in the I/B/E/S data, average bid-ask spread as the average of percentage bid-ask spreads between t=-70 and t=-41, and industry dummies as the 13 different categories in Lewellen (1999). Due to the nature of each company’s business, earnings announcements from companies in different industries release different amounts of new information. For example, compare earnings announcements from a clothing company and a petroleum company. The performance of the oil company heavily depends on the market price of oil, which is readily available as public information. The performance of the clothing company is driven mostly by retail sales, which are more difficult to predict or estimate with public information. Therefore, the price impact of an oil company’s earnings announcement is typically much smaller than for a clothing company. In particular, the average of absolute returns on earnings announcement days in the oil industry is 1.32%, the lowest level among the 13 industry groups. Risk during announcements and contemporaneous price change should be included as control variables. Given various empirical results about increasing volatility around earnings announcements (e.g., Donders and Vorst (1996)), changing risk can be a driving force behind trading volume before corporate announcements. In a hypothetical world with no information asymmetry, investors will trade securities with allocation or liquidity motives. If investors with only allocational motives know that there will be a change of risk in the future, they will have an incentive to trade. For example, if a utility company acquires a high risk bio-tech company, this increased risk can induce demand from risk seekers and supply 15
For example, Llorente et al. (2002), Hong, et. al. (1998), and Venkatesh and Chiang (1986).
17
from risk averters. In this example, trading volume before a risk-changing announcement can increase. Since no risk measures perfectly capture all risk that a stock exhibits, I use several measures of risk in this paper. The first one is the price change on an announcement day. In principle, this is an ideal measure of risk caused by the announcement, but it has an endogeneity problem. Specifically, higher turnover prior to an announcement could lead to a smaller price effect on the day of the announcement. To accommodate this endogeneity problem, I use the price change from the announcement prior to the one in question. Especially in the case of earnings announcements, this measure is a good proxy for the price change on the current earnings announcement because of small variations between the performance of adjacent quarters.16 Finally, I investigate the change in beta between pre-announcement and postannouncement periods to accommodate changing systematic risk. The change in beta is measured by the percentage change from beta in the pre-announcement period (i.e., t=-1 to -70) to beta in the post-announcement period (i.e.,t=+1 to +70). Since non-synchronous trading in daily data affects the estimation of beta, I use the estimation method in Dimson (1979). Table V reports the results from the analysis with the price change from the previous announcement. The results with two other control variables are similar and thus not reported here. Contemporaneous price change is perhaps the most important control variable to be included in the regression specification. As shown in many previous studies (e.g., Wang (1994)), trading volume has high correlation with contemporaneous price volatility. To control for this relation, I include the price change in the period between t=-10 and t=-3. Only before earnings announcements can I notice the expected relation between proxies of information asymmetry and trading volume (Table V). The coefficients of log capitalization across different specifications are statistically significant. With two control variables, 16
The correlation between the absolute price change of this quarter’s earnings announcement and the previous quarter’s is 23.9%.
18
the coefficient for size is 1.89 with a t-statistic of 6.37. If we accept the argument that a smaller company exhibits more information asymmetry than a larger company, information asymmetry surely affects the trading behavior before an earnings announcement. Even though there is a larger price change before acquisition announcements and a similar price change before target announcements, size does not seem related to trading volume before those announcements. Insert Table V here Since analysts post their forecasts in various public and private media outlets, individual investors can obtain this information more easily if there are more analysts. If investors cannot easily find information or the cost of obtaining information is high, they will consider themselves uninformed. As uninformed investors, they should consider adverse selection costs and trade less. As shown in Table V, trading volume before earnings announcements is positively related with the number of analysts covering the company. However, there is almost no relation between the number of analysts and trading volume before an acquisition announcement or a target announcement. This suggests that uninformed investors know about their disadvantage before scheduled announcements but not before unscheduled announcements. Market makers are also considered to be uninformed investors with respect to corporate insider information (e.g., Kyle (1985), Venkatesh and Chiang (1986), etc.). Since the bid-ask spread is one method by which market makers protect themselves adverse selection costs, it has been used as a proxy for information asymmetry (e.g., Llorente et al. (2002)). Table V also shows the result from the regression using average bid-ask spread as an information proxy measure. The coefficient has a statistically significant value in the earnings announcement sample. However, unlike the coefficients for other proxies, the coefficient of average bid-ask spread is statistically significant even in the acquisition and target samples. We can interpret this significance as follows. As Stoll (1989) argues, the bid-ask spread 19
has three components: order processing costs, inventory holding costs, and adverse selection costs. Therefore, the relation between abnormal trading volume and components other than adverse selection costs can also drive the negative relation. For example, high trading before acquisition and target announcements can be negatively related to the bid-ask spread cross-sectionally. In this case, even though trading volume is not related to information asymmetry, we can still observe a negative relation (Panels B and C of Table V). I show the effect of information asymmetry on trading volume using industry dummy variables in Table V. Prior to earnings announcements from firms in raw material industries (such as petroleum, logging, or mining), all traders have access to publicly available raw materials prices. Uninformed investors do not have to worry about the information advantage of informed traders and can stay in the market before scheduled earnings announcement. Again, statistically significant values of these coefficients are observed mostly in scheduled earnings announcements. In Panel D of Table V, I apply the same investigation as in Panels A, B, and C, to the high absolute return “events” mentioned before.17 Overall, the relation between information asymmetry and volume is much smaller and less significant for these “events” than for earnings announcements. For example, the coefficient for size (Log Cap) from this data is 1.00, around half of the coefficient from the earnings announcement data. Since the randomly selected data may include scheduled announcements other than earnings (e.g., court orders), I expected this similarity. However, it is clear that for the controlled unscheduled announcements, none of the information asymmetry proxies (except for the bid-ask spread) has significant t-statistics (see Panels B and C). Insert Table VI here To test the statistical differences of coefficients in the regressions for scheduled earnings 17 I use the 50,000 randomly selected high absolute return days used to draw the fourth bar in Fig. 2. These have the average price change (2.64%) closest to that of earnings announcements (2.66%).
20
announcements and unscheduled announcements, I implement the Wald test. Table VI reports the results. Most information asymmetry proxies in the unscheduled announcements sample are statistically different from the proxies in the scheduled announcements sample. This provides evidence that information asymmetry does not influence trading volume before these unscheduled announcements in the same way as it does before scheduled announcements. Without timing information, uninformed investors do not seem to realize the existence of information asymmetry and thus not optimize their trading.
B. Information asymmetry and trading volume on/after announcements In the previous section, I tested whether a decrease in trading volume before scheduled earnings announcements results from ex ante information asymmetry. In this section, I analyze trading volume on and after earnings announcements (as in Hypothesis 3). George et al. (1994) argue that the increase in trading volume on announcements can be explained by the lower adverse selection costs remaining after announcements. Though George et al. do not directly state that trading volume should decrease before announcements (when adverse selection costs are still high), such a pattern would be a reasonable extension of their argument. I verified this extension in previous sections. As a more direct test for their theoretical prediction, I test whether the relation between trading volume and information asymmetry reverses on/after announcements. Any trade demand that is delayed because of information asymmetry before an announcement should be submitted on or after the announcement. Evidence from this test reassures us of the results in the previous sections. Insert Table VII here The regressions in this section have the same specifications as in the last section, except that the dependent variable is now abnormal turnover on and after announcements (see Table VII). In Panel A the dependent variable is abnormal turnover from t=-2 to t=+2, and 21
in Panel B the dependent variable is the abnormal turnover from t=+3 to t= +10. As in Hypothesis 3, the coefficients for information asymmetry proxies change sign or have much smaller magnitude in the regressions. In Table VIII, I also implement the Wald test to see how statistically different the coefficients are. Most coefficients for information asymmetry proxies are significantly different. The Wald statistics are much larger than 3.84, which is the critical value of 95% in χ2 -distribution. Therefore, I generally confirm the prediction of George et al. (1994) and Hypothesis 3. In summary, the magnitude of ex ante information asymmetry appears to drive the magnitude of turnover patterns before and after scheduled earnings announcements.
C. Price sensitivity of market makers According to predictions from Kyle (1985), if more informed trading is expected then market makers should increase price sensitivity to order flow to protect themselves. Therefore, by investigating price sensitivity near announcements, I can make inferences about information asymmetry. As discussed in the previous sections, uninformed traders behave differently depending on the availability of timing information. Market makers have more information than uninformed investors (they know order flow, order books, etc.). An analysis of their behavior should indicate how their information set affects their behavior around announcements, with or without timing information. I use the Trade and Quote (TAQ) database for these tests. First, 1,100 observations are randomly selected from the initial earnings, acquisition, target, and high price change data sets.18 After matching with the TAQ database, 1,068, 1,025, 961, and 719 observations remain and are analyzed. Table IX compares price sensitivity to order flow across announcement types. The price sensitivity is measured by regressing return on signed order flow.19 18 The average absolute return on t=0 in high price change data is about the same as that of earnings announcement days, i.e., 2.64% 19 The Lee and Ready (1991) tick test is used to determine the sign of each trade order.
22
Since the increase in trading before an announcement is also related to ex ante information asymmetry and informed/uninformed investors’ trading in the market, the companies in the data set are sorted according to the total number of trades in the testing period. After obtaining price sensitivities from estimation and testing periods, I construct a measure of the difference between these periods. Insert Table IX here I find that market makers raise price sensitivity before all three types of announcements and high absolute return days (see Table IX). Without any restriction on the number of trades in the testing period, the differences in price sensitivities between the testing and estimation period are 0.255, 0.211, 0.425, and 0.623 from earnings announcements, acquisition announcements, target announcements, and high absolute return days, respectively. All of these differences are statistically significant. Therefore, market makers increase price sensitivity before all announcements, whether scheduled or not. This result warrants further investigation. First of all, the market maker increases price sensitivity before scheduled announcements as hypothesized. This phenomenon can be explained with the same argument as before, namely that market makers know informed traders are in the market and increase price sensitivity to protect against adverse selection costs. More interestingly, I also observe higher price sensitivity before unscheduled announcements and high absolute return days. I conjecture the following explanation. Market makers have a special type of “non-inside but private” information in their order books. The order book may somehow reflect timing information or informed limit orders. If market makers can extract timing information or detect levels of information-based trading, they can optimize their behavior (increase price sensitivity) regardless of the existence of timing information. This increased price sensitivity would not be observed by uninformed investors since order-by-order information is not instantly available. 23
Based on the results in this section, market makers generally behave in an appropriate way to protect themselves. They can find and react to abnormal movements in order flow near announcements even without public timing information. Therefore, I conclude that decreased trading volume before earnings announcements is consistent with market makers’ price setting behavior.
IV
Other implications
On the basis of the tests in this paper, I show that timing information alerts uninformed investors to evaluate adverse selection costs and optimize when to trade. Without timing information, uninformed investors do not seem to react to potential adverse selection costs. Therefore, another interpretation of these results concerns the overconfidence of uninformed investors. As shown in takeover (especially, target) announcements, uninformed investors maintain participation in the market when they observe abnormally high trading volume. Although they know that informed investors exist, uninformed investors do not consider possible adverse selection costs and join the flood of trading before takeovers. This phenomenon may be an example of the behavioral finance argument that investors are overconfident. Another point to reflect upon is the behavior of market makers. Market makers are also uninformed about the timing of announcements. However, they react consistently whether or not there is timing information. I reason that the private observation of the order book and order flow can help them recognize impending information issuances and protect themselves by increasing the price sensitivity. Therefore, it would also be interesting to see if the same trading volume pattern exists where the order book and order flow are public information, such as Paris exchange.20 The results in this paper contribute to understanding 20
I appreciate Carl Hopman for providing the information about the Paris exchange.
24
potential effects of information disclosure in the stock market and optimal transparency of dealer-specific information.
V
Conclusion
Using I/B/E/S earnings announcements, I find that cumulative trading volume decreases by more than 15% before scheduled earnings announcements. Before all other investigated announcements (SDC takeover, SDC target, and Moody’s bond rating change), cumulative trading volume increases. I also show that cross sectional trading volume is negatively correlated with levels of information asymmetry before scheduled announcements and positively correlated with information asymmetry after scheduled announcements. Again, this pattern holds only for scheduled announcements. All the results for scheduled announcements are consistent with theories of asymmetric information, in which DLTs decrease trading volume when they perceive adverse selection. However, I find that market makers act appropriately by increasing price sensitivity before all announcements. This implies that market makers might extract timing information from their order books. On the other hand, liquidity traders do not seem to correctly read information embedded in prices before unscheduled announcements. This suggests that uninformed liquidity traders cannot behave according to theories of information asymmetry when they do not have timing information. These results leave two open questions. First, what is the relation between timing information and return/volatility? Second, how do market makers determine optimal price sensitivity before different types of announcements? Studies of these questions yield further advances in our understanding of trading under information asymmetry.
25
References Admati, Anat, and Paul Pfleiderer, 1988, A theory of intraday patterns: Volume and price variability, Review of Financial Studies 1, 3-40. Ajinkya, Bipin B., and Prem C. Jain, 1989, The behavior of daily stock market trading volume, Journal of Accounting and Economics 11, 331-359. Atiase, Rowland K., and Linda Smith Bamber, 1994, Trading volume reactions to annual accounting earnings announcements, Journal of Accounting and Economics 17, 309-329 Bamber, Linda Smith, 1987, Unexpected earnings, firm size, and trading volume around quarterly earnings announcements, Accounting Review 62, 510-532. Bamber, Linda Smith, and Youngsoon S. Cheon, 1995, Differential price and volume reactions to accounting earnings announcements, The Accounting Review 70, 417-441. Black, Fischer, 1986, Noise, Journal of Finance 41, 529-543. Chordia, Tarun, Roll, Richard, and Subrahmanyam, Avanidhar, 2001, Market liquidity and trading activity, Journal of Finance 56, 501-530. Dimson, Elroy., 1979, Risk measurement when shares are subject to infrequent trading, Journal of Financial Economics 7, 197-226. Donders, Monique W.M., and Ton C.F. Vorst, 1996, The impact of firm specific news on implied volatilities, Journal of Banking and Finance 20, 1447-1461. Easley, David, Soeren Hvidkjaer, and Maureen O’Hara, 2002, Is information a determinant of asset returns? Journal of Finance 57, 2185-2221. Efron, Bradley, and Tibshirani, Robert J., 1983, An introduction to the Bootstrap, Chapman and Hall, New York, New York. Fama, Eugene, and James MacBeth, 1973, Risk, return, and equilibrium: Empirical tests, Journal of Political Economy 71, 607-636. Foster, F. Douglas, and S. Viswanathan, 1990, A theory of the interday variations in volume, variance, and trading costs in securities markets, Review of Financial Studies 3, 593-624. Foster, George, Chris Olsen, and Terry Shevlin, 1984, Earnings releases, anomalies, and the behavior of security returns, The Accounting Review 59, 574-603. George, Thomas J., Gautam Kaul, and M. Nimalendran, 1994, Trading volume and transaction costs in specialist markets, Journal of Finance 49, 1489-1505. Grossman, Sanford, and Joseph E. Stiglitz, 1980, On the impossibility of informationally efficient markets, American Economic Review 70, 393-408. Hand, John R. M., Robert W. Holthausen, and Richard W. Leftwich, 1992, The effect of bond rating agency announcement on bond and stock prices, Journal of Finance 47, 733-752. Hong, Harrison, Terence Lim, and Jeremy C. Stein, 2000, Bad news travels slowly: Size, analyst coverage and the profitability of momentum strategies, Journal of Finance 55, 265-295. Jarrell, Gregg A., and Annette B. Poulsen, 1989a, Returns to acquiring firms in tender offers: Evidence from three decades, Financial Management 18, 12-19.
26
Jarrell, Gregg A., and Annette B. Poulsen, 1989b, Stock trading before the announcement of tender offers: Insider trading or market anticipation? Journal of Law, Economics and Organizations 8, 225-248. Jensen, Michael C., and Richard Ruback, 1983, The market for corporate control: The scientific evidence, Journal of Financial Economics 11, 5-50. Karpoff, Jonathan M., 1987, The relation between price changes and trading volume: A Survey, Journal of Financial and Quantitative Analysis 22, 109-126. Kyle, Albert S., 1985, Continuous auctions and insider trading, Econometrica 53, 1315-1336. Lee, Charles M.C., and Mark J. Ready, 1991, Inferring trade direction from intraday data, Journal of Finance 46, 733-746. Lewellen, Jonathan W., 1999, The time-series relations among expected return, risk, and book-tomarket, Journal of Financial Economics 54, 5-43. Lo, Andrew W., Jiang Wang, 2000, Trading volume: definitions, data analysis, and implications of portfolio theory, Review of Financial Studies 13, 257-300. Llorente, Guillermo, Roni Michaely, Gideon Saar, and Jiang Wang, 2002, Dynamic volume-return relation of individual stocks, Review of Financial Studies 15, 1005-1047. Milgrom, Paul, and Nancy Stokey, 1982, Information, trade and common knowledge, Journal of Economic Theory 26, 17-27. Sanders, Ralph W. Jr., and John S. Zdanowicz, 1992, Target firm abnormal returns and trading volume around the initiation of change in control transactions, Journal of Financial and Quantitative Analysis 27, 109-129. Stoll, Hans R., 1989, Inferring the components of the bid-ask spread: Theory and empirical tests, Journal of Finance 44, 115-134. Venkatesh, P.C., and R. Chiang, 1986, Information asymmetry and the dealer’s bid-ask spread: A case study of earnings and dividend announcements, Journal of Finance 41, 1089-1102. Wang, Jiang, 1994, A model of competitive stock trading volume, Journal of Political Finance 102, 127-168.
27
Figure 1. Cumulative abnormal log turnover from t=-15 to t=15 Acquisition announcements 180
150
150 Cumulative abnormal log(turnover)
Cumulative abnormal log(turnover)
Earnings announcements 180
120
90
60
30
120
90
60
30
0
0
-30
-30 -15
-10
-5
0
5
10
15
-15
-10
-5
Relative Days
(a)
150
150
120
90
60
30
10
15
90
60
30
0
-30
-30 0
15
120
0
-5
10
Moodys announcements 180
Cumulative abnormal log(turnover)
Cumulative abnormal log(turnover)
Target announcements
-10
5
(b)
180
-15
0 Relative Days
5
10
15
-15
-10
-5
Relative Days
0
5
Relative Days
(c)
(d)
Random Days 180
Cumulative abnormal log(turnover)
150
120
90
60
30
0
-30 -15
-10
-5
0
5
10
15
Relative Days
(e)
For each announcement, average log turnover is determined from t=-45 days to t=-16 days, where turnover is daily volume divided by shares outstanding. These plots show cumulative abnormal turnover from t=-15 to +15, i.e. cumulative excess over the average of the preceding month. Plot (e) selects random days as t=0 and illustrates that the measure, cumulative abnormal turnover, is unbiased. For further explanation, see Section II.
28
Figure 2. Abnormal log turnover from t=-10 to t=-3 according to absolute return on t=0
Abnormal log(turnover) in (-10, -3) 10.00
9.20 (5.34)
9.00
8.00 6.42 (6.35)
Percentage change
7.00
6.00
5.12 (7.25)
5.00 3.63 (7.32)
4.00 3.00 (12.48) 3.00
2.13 (11.61)
2.00
1.00
0.13 (0.71)
0.10 (1.02)
2.10
2.20
0.63 (3.70)
0.50 (2.22)
2.42
2.64
0.61 (4.43)
0.00 2.82
4.04
5.27
6.58
7.75
10.52
14.71
Average absolute return (%) on t=0
Each bar represents the average abnormal log turnover from t=-10 to -3, with average absolute return on the x-axis. For example, the first bar is the average abnormal log turnover from t=-10 to -3 given a minimum absolute return of 0.2% on t=0. “2.10” is the average absolute return on t=0 in percent, “0.13” is the point estimate, and “0.71” is the t-statistic of the estimate. 50,000 random days are chosen for each bar of the graph, based on a minimum absolute return filter (0.2%, 0.4%, 0.6%, 0.8%, 1.0%, 2.0%, 3.0%, 4.0%, 5.0%, 7.0%, and 10.0% from the left to the right). 10,000 points are chosen for each size quintile of NYSE and AMEX companies over 1986 to 2000 given the respective filter. No earnings, acquisition, or target announcements exist in the 50,000 random observations from t=-40 to t=+10. The abnormal turnover is the average of the differences between log turnover from t=-10 to t=-3 and average log turnover estimated from t=-40 to t=-11.
29
Table I: Data sets and applied filter This table reports the number of observations in each sample after filters are applied. “# of Obs. Filter” means that at least 40 trading days before and 10 days after the announcement are required to be included. The third and fourth rows indicate the number of observations for companies existing in two announcement sets. For example, “Same Companies in Earnings and Target” means that the companies have at least one earnings announcement and one target announcement. Earningsa
Acquisitionb
Targetb
Moody’sc
Before # of Obs. Filter
65,912
25,087
12,485
410
After # of Obs. Filter
64,285
22,930
11,255
320
Same Companies in Earnings and Acquisition Same Companies in Earnings and Target
55,747
19,888
N/A
N/A
51,130
N/A
8,430
N/A
NYSE and AMEX
a The earnings announcement data between 1986 and 2000 are extracted from the I/B/E/S database. The companies had to be in the NYSE and the AMEX. b The acquisition and the target announcement data between 1986 and 2000 are from the SDC database. c The Moody’s bond rating change data between January 1997 and October 1998 are hand-collected from Moody’s Bond Survey from Moody’s Investors Service.
30
Table II: Summary statistics This table reports summary statistics of daily percentage return and percentage turnover from stocks existing in NYSE or AMEX for at least 1 year (252 days) in each period. The summary statistics are the averages of estimates (mean, standard deviation, skewness, and kurtosis) for each company. The daily turnover is defined as daily trading volume divided by outstanding shares. Period
Mean
Standard Deviation
Skewness
Kurtosis
# of firms
Daily Return(%) 1986-2000
0.056
3.290
1.049
23.407
5,982
1986-1990 1991-1995 1996-2000
0.040 0.075 0.045
3.068 3.038 3.046
0.620 0.685 0.904
18.875 12.215 17.209
3,031 3,823 4,406
Daily Turnover(%) 1986-2000
0.389
0.865
8.596
159.439
5,982
1986-1990 1991-1995 1996-2000
0.260 0.272 0.447
0.439 0.447 0.964
7.166 7.259 7.317
101.495 108.934 108.898
3,031 3,823 4,406
Log Daily Turnover 1986-2000
−2.007
1.075
−0.098
0.701
5,906
1986-1990 1991-1995 1996-2000
−2.221 −2.160 −1.825
1.053 1.033 0.975
−0.026 −0.091 −0.022
0.590 0.630 0.826
2,972 3,731 4,404
31
Table III: Daily abnormal turnover around different events This table contains the daily abnormal turnover around each announcement. The abnormal turnover (from the companies in the NYSE and the AMEX between 1986 and 2000) is reported as the difference between log turnover and average log turnover estimated from t=-40 to t=-11, where turnover is trading volume divided by shares outstanding. The t-statistics are given in parentheses. Average(-10,-3) is the average abnormal turnover from t=-10 to t=-3. “Difference of averages from earnings” is the difference in Average (-10,-3) values from the earnings Average(-10,3). T-values are estimated with the assumption that the abnormal turnover in each announcement has a different variance. Announcements No. of Obs.
Earning 64,285
Acquisition 22,930
Target 11,255
Moody’s 330
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
-1.032( -3.048) -1.852( -5.367) -2.616( -7.526) -2.645( -7.573) -2.283( -6.502) -1.418( -4.083) -1.831( -5.222) -1.381( -3.910) 0.863( 2.435) 12.593( 34.627) 41.141(108.997) 37.636(100.763) 22.422( 61.530) 16.836( 46.624) 12.584( 34.918) 11.127( 30.726) 8.282( 22.848) 6.535( 18.003) 4.761( 13.058) 4.777( 13.180) 4.432( 12.066)
2.044( 3.578) 2.633( 4.524) 2.239( 3.843) 2.937( 4.963) 3.538( 5.933) 2.359( 3.974) 4.073( 6.749) 4.149( 6.839) 5.819( 9.535) 12.386(19.852) 33.022(49.506) 28.896(43.922) 19.270(30.746) 14.697(23.650) 10.589(16.976) 9.322(15.035) 8.566(13.996) 7.226(11.642) 5.437( 8.809) 4.476( 7.259) 4.921( 7.978)
9.342( 9.941) 8.942( 9.363) 11.727(11.979) 14.954(14.885) 16.345(16.367) 16.243(16.165) 17.955(17.759) 21.516(21.092) 26.733(25.304) 38.659(34.851) 98.625(75.857) 90.169(71.989) 62.520(54.358) 49.347(44.044) 38.723(35.956) 35.012(32.120) 30.179(28.603) 25.887(24.447) 23.424(22.183) 21.681(20.475) 19.733(18.853)
2.842( 0.670) -1.929(-0.443) -0.449(-0.100) 2.907( 0.696) 3.657( 0.847) -0.529(-0.115) 2.050( 0.428) 9.114( 1.730) 13.693( 2.630) 6.546( 1.138) 5.401( 0.939) 10.685( 1.880) 8.698( 1.542) 8.808( 1.632) 4.895( 0.941) 2.233( 0.480) 3.565( 0.710) -1.765(-0.356) -1.005(-0.204) 2.531( 0.521) -7.742(-1.555)
-1.882(-8.955)
2.997(10.231)
14.628(9.497)
2.208(1.820)
N/A
4.879(13.534)
16.510(10.620)
4.090(3.323)
Average(-10,-3) Difference of averages from earnings
32
33
3.238? 6.033? 12.598? 33.290? 29.099? 19.524? 8.427? Acquisition 22,817 1.177? 2.801? 9.201? 26.733? 24.433? 13.962? 5.908?
−2.251‡ 0.506 12.197? 40.834? 37.435? 22.258? 8.392? Earnings 64,014 −2.492‡ −0.236 11.666? 46.212? 41.153? 22.337? 7.571?
(-10, -3) -2 -1 0 1 2 (3, 10)
Announcement No. of Obs.
(-10, -3) -2 -1 0 1 2 (3, 10)
Panel B: Using median abnormal turnover
Acquisition 22,797
Earnings 65,447
Announcement No. of Obs.
Panel A: Using larger estimation window(t=-55 to -11)
10.697? 21.460? 33.547? 116.091? 103.052? 63.526? 27.570?
Target 11,162
16.115? 28.179? 40.137? 100.281? 91.718? 63.999? 31.993?
Target 11,220
−1.482 6.885 −1.173 −2.868 10.164 5.645 −0.200
Moody’s 330
5.170◦ 17.339? 8.877◦ 8.756◦ 14.073? 11.095∗ 4.562
Moody’s 328
This table reports alternative measures of abnormal log turnover. Panel A employs a larger estimation window of 45 days, Panel B uses the difference between raw and median turnover, Panel C uses the residuals from a one factor market model, Panel D uses subperiod analysis, and Panel E uses matched samples. The rows of (-10,-3) and (3,10) report the summary measures of average of daily abnormal trading volume in those 8 days. Turnover is trading volume divided by shares outstanding. In panel A, C, D, and E, ¦, †, and ‡ mean respectively 10%, 5%, and 1% in the left tail of t-distribution and ◦, ∗, and ? mean respectively 10%, 5%, and 1% in the right tail of the t-distribution. In panel B, ¦, †, ‡, ◦, ∗, and ? are from a bootstrapped distribution.
Table IV: Robustness check
34
1991-95 18,961 −2.807‡ −0.533 8.337? 48.136? 42.777? 23.940? 8.168?
1986-90 12,974 −0.821‡ 7.237? 25.557? 31.024? 20.489? 11.487? 6.512?
Subperiod No. of Obs.
(-10, -3) -2 -1 0 1 2 (3, 10)
5.458? 10.009? 14.899? 38.952? 33.651? 21.620? 8.807?
1986-90 4,565 2.063? 6.133? 8.148? 34.209? 28.486? 18.277? 6.302?
1991-95 6,031
Acquisition
Panel E: From the same companies
−1.765‡ −0.876† 9.889? 41.098? 41.499? 25.918? 9.823?
1996-00 32,350
Acquisition 19,889 2.908? 5.334? 11.749? 31.228? 27.916? 18.134? 7.844?
Earnings 55,747 −1.915‡ 0.723∗ 12.985? 41.702? 37.972? 22.759? 8.712?
Announcements No. of Obs.
(-10, -3) -2 -1 0 1 2 (3, 10)
Companies in both earnings and acquisition
Earnings
Announcement
2.265? 4.187? 10.431? 29.784? 26.025? 16.729? 7.306?
−4.186‡ −2.333‡ 9.076? 36.989? 33.600? 19.875? 7.272?
(-10, -3) -2 -1 0 1 2 (3, 10) Panel D: Using subperiod
Acquisition 22,420
Earnings 64,840
Announcement No. of Obs. 12.676? 23.261? 35.452? 93.230? 85.358? 59.024? 29.656?
Target 10,829
21.247? 36.589? 42.441? 113.681? 105.052? 72.479? 34.527?
1986-90 3,726
11.258? 21.478? 23.985? 80.930? 72.592? 47.007? 20.892?
1991-95 2,891
Target
−1.887‡ 0.609◦ 12.949? 41.459? 37.749? 22.482? 8.826?
Earnings 51,130
12.228? 22.553? 34.813? 86.962? 76.786? 51.983? 24.835?
Target 8,430
11.413? 22.091? 44.768? 97.560? 89.171? 64.184? 33.250?
1996-00 4,638
2.028 13.186? 10.800∗ 10.650∗ 18.012? 10.681∗ 1.137
Moody’s 323
Companies in both earnings and target
2.542? 4.115? 13.529? 30.247? 27.337? 18.887? 8.819?
1996-00 12,334
Panel C: Using one factor market model
Table IV: (continued)
35
Table V: Regression analysis
17.50 (2.70) −19.59 (−3.12) 3.81 (2.91) −10.20 (−9.29) 4.37 (2.91) −2.91 (−2.28) 3.17 (3.45) −10.97 (−12.18)
−15.82 (−2.40) −53.04 (−8.81) −3.73 (−3.72) −16.66 (−20.94) −0.92 (−0.85) −7.15 (−7.61) −2.02 (−2.31) −14.09 (−18.46)
Constant
−0.71 (−2.39) 0.41 (1.46)
0.68 (2.15) 1.89 (6.37)
Log Cap
−0.12 (−1.61) −0.03 (−0.49)
0.30 (3.55) 0.42 (5.15)
Number of Analysts
Agriculture Fishing Mining
Logging Paper Printing
0.93 (0.77) −0.37 (−0.31)
1.28 (1.33) 2.32 (2.41)
−37.04 (−0.89) −395.35 (−8.93) 0.14 (0.09) −1.38 (−0.84)
0.06 (0.04) 0.10 (0.09)
Panel B: Acquisition announcements
−21.35 (−0.91) −325.48 (−13.44)
Panel A: Earnings announcements
Average Spread [-70,-41]
Independent Variables
0.51 (0.20) 2.27 (1.19)
3.51 (2.04) 5.18 (2.94)
Petroleum
−67.05 (−3.91)
−31.19 (−1.85)
−76.11 (−4.12)
−62.32 (−3.68)
−60.57 (−7.05)
−22.57 (−2.64)
−60.27 (−6.85)
−52.99 (−6.40)
Labsret0
2402.08 (22.32)
2611.30 (22.40)
2382.19 (19.87)
2405.58 (22.08)
2228.95 (20.26)
2415.75 (21.03)
2234.39 (19.87)
2295.68 (20.73)
Absolute Return [-10,-3]
10.39
−0.04
11.77
0.34
9.48
0.23
10.71
0.42
7.87
0.12
8.56
0.15
8.18
0.37
8.57
0.61
¯2 R
This table reports the results of regressions of abnormal trading volume before announcements on proxies of ex ante information asymmetry. All results are from Fama/MacBeth (1973) type regressions. The coefficients are time series averages of the coefficients from cross-sectional regressions using each quarter data. The time series averages of the coefficients are obtained by weighting the number of observations used in the cross-sectional regressions. The dependent variable is defined as the difference between average log turnover from t=-10 to t=-3 and average log turnover from t=-11 to t=-40. Among independent variables, “Labsret0” is the absolute return on the last announcement day. All the ¯ 2 means the average of the adjusted R-squares coefficients have been multiplied by a factor of 100, and t-statistics are given in parentheses. R in each cross-sectional regression.
36
2.67 (1.42) −17.48 (−9.18) 1.22 (1.69) −10.90 (−11.98) 2.10 (2.68) −5.05 (−6.14) 0.21 (0.30) −10.41 (−12.86)
52.75 (7.49) −14.02 (−1.59) 14.32 (8.74) −9.06 (−4.55) 14.31 (7.36) 2.72 (1.44) 14.78 (12.73) −6.38 (−4.95)
Constant
−0.36 (−1.44) 1.00 (4.25)
−1.93 (−5.31) 0.36 (0.85)
Log Cap
6.79 (0.15) −430.01 (−7.62)
Average Spread [-70,-41]
Logging Paper Printing
−4.15 (−1.38) −8.38 (−2.84)
2.65 (0.83) 1.82 (0.57)
Panel C: Target announcements
Agriculture Fishing Mining
−6.19 (−1.61) −1.87 (−0.57)
Petroleum
5.81 (0.40)
28.23 (1.82)
3.01 (0.15)
10.27 (0.67)
Labsret0
−0.05 (−1.16) 0.05 (1.13) −56.25 (−2.57) −335.72 (−13.31) 2.47 (1.75) 0.73 (0.55)
1.34 (0.88) 2.75 (1.82)
3.91 (1.97) 5.33 (2.83)
−126.59 (−9.48)
−17.09 (−1.38)
−113.55 (−6.71)
−105.26 (−8.53)
Panel D: Random high price change days (Average absolute return on t=0 is 2.64%)
−0.45 (−3.80) −0.09 (−0.64)
Number of Analysts
Independent Variables
Table V: (continued)
2105.89 (23.36)
2321.96 (24.14)
2217.26 (19.87)
2152.56 (23.46)
2518.67 (18.03)
2741.52 (18.85)
2741.43 (19.87)
2516.06 (17.52)
Absolute Return [-10,-3]
8.78
0.08
9.91
0.28
8.67
0.02
9.06
0.32
14.87
−0.18
16.58
0.72
14.88
0.24
15.24
0.62
¯2 R
Table VI: Wald test: Comparison of coefficients before different announcements This table reports the Wald test to compare the coefficients of information asymmetry proxies for earnings announcements to those for other announcements and random days in univariate Fama/Macbeth type regressions of Table V. The 95% significance level for χ2 with degree of freedom = 1 is 3.84.
Log Cap
Number of Analysts
Average Spread [-70,-41]
Agriculture Fishing Mining
Logging Paper Printing
Petroleum
Acquisition
10.24
13.78
0.11
0.17
0.51
0.92
Target
29.27
26.49
0.30
2.47
0.17
5.28
Absolute return = 2.64%
6.66
13.55
1.18
0.68
0.00
0.02
37
38
Table VII: Regression analysis
−0.74 (−1.68) 0.42 (1.01)
−1.66 (−3.72) 0.09 (0.25)
56.83 (5.99) 0.28 (0.03) 23.24 (15.33) 1.46 (1.09) 21.50 (13.32) 10.54 (7.51) 23.50 (18.17) 2.42 (1.98)
23.89 (2.52) −12.38 (−1.39) 9.91 (5.73) −3.25 (−1.93) 14.38 (7.90) 7.03 (4.20) 8.89 (6.13) −3.51 (−2.41)
Log Cap
Constant
−0.18 (−1.65) −0.04 (−0.39)
−0.03 (−0.32) 0.11 (1.15)
Number of Analysts
Agriculture Fishing Mining
Logging Paper Printing Petroleum
−6.51 (−4.15) −6.63 (−4.39)
−0.52 (−0.43) 1.69 (1.53)
−9.41 (−3.37) −4.60 (−1.73)
−203.61 (−5.94) −513.39 (−14.72) −2.82 (−1.66) −3.66 (−2.12)
1.56 (1.33) 2.34 (2.04)
−5.33 (−2.09) −4.11 (−1.71)
Panel B: After announcements [+3,+10]
72.10 (1.88) −394.81 (−13.44)
Panel A: On earnings announcements [-2,+2]
Average Spread [-70,-41]
Independent Variables
−54.37 (−4.52)
10.05 (0.84)
−55.69 (−4.69)
−53.17 (−4.45)
24.19 (2.06)
73.70 (6.55)
23.78 (2.06)
23.52 (2.10)
Labsret0
2285.39 (18.59)
2552.90 (20.78)
2260.16 (18.83)
2299.84 (19.20)
2038.62 (22.61)
2151.06 (23.08)
2025.31 (22.32)
2034.69 (23.07)
Absolute Return [-10,-3]
6.73
0.25
8.02
0.52
6.75
0.43
7.22
0.89
12.88
0.23
13.53
0.29
12.87
0.27
13.21
0.86
¯2 R
This table reports the results of regressions of abnormal trading volume during and after earnings announcements on proxies of ex ante information asymmetry. All results are from Fama/MacBeth (1973) type regressions. The coefficients are time series averages of the coefficients from cross-sectional regressions using each quarter data. The time series averages of the coefficients are obtained by weighting the number of observations used in the cross-sectional regressions. The dependent variables are defined as the difference between average log turnover from t=-2 to t=+2 and average log turnover from t=-11 to t=-40 for Panel A, and the difference between average log turnover from t=+3 to t=+10 and average log turnover from t=-11 to t=-40. Among independent variables, “Labsret0” is the absolute return on the last announcement day. All the coefficients have been multiplied by a factor of 100, and t-statistics are given in parentheses.
Table VIII: Wald test: Comparison of coefficients (“Before” vs. “Before” vs. “After”)
“On” and
This table reports the Wald test to compare the coefficients of information asymmetry proxies for earnings announcements with different dependent variables. In panel A, it compares the univariate Fama/MacBeth type coefficients of information asymmetry proxies for log turnover prior to earnings announcements (t=[-10,3]) and those on earnings announcements (t=[-2,+2]). In panel B, it compares the univariate Fama/MacBeth type coefficients of information asymmetry proxies for log turnover prior to earnings announcements (t=[10,-3]) and those on earnings announcements (t=[+3,+10]). The 95% significance level for χ2 with degree of freedom = 1 is 3.84.
Log Cap
Number of Analysts
Average Spread [-70,-41]
Agriculture Fishing Mining
Logging Paper Printing
Petroleum
Panel A: Comparison with “on earnings announcements [-2,+2]” 18.30
6.11
4.34
14.07
1.36
15.52
Panel B: Comparison with “after earnings announcements [+3,+10]” 6.89
12.03
19.26
3.21
39
0.03
8.26
Table IX: Abnormal price sensitivity before announcements The price sensitivity (λ) is estimated from the corresponding period, t=-40 to -11 or t=-10 to -3, in random samples from the initial data sets of earnings, acquisition, and target announcements. The numbers of observations in the initial data sets are 65,912 earnings announcements, 25,087 acquisition announcements, and 12,485 target announcements. From 1,100 randomly selected observations in each announcement data set, after matching with the Trade and Quote (TAQ) data set and the applied number of observation filters, the final data sets have 1,068 earnings, 1,025 acquisition, and 961 target observations. For Panel D, 719 observations from the data set used in Panel D of Table V are analyzed. To estimate price sensitivity, I regress return (in basis points) on signed order flow (in hundreds, signed by the technique in Lee and Ready (1991)). Average number of trading per day in (-10,-3)
(0,10]
(10,100]
(100,∞)
Total
0.017 (1.19) 390
0.255 (6.96) 1,068
0.052 (1.59) 413
0.211 (5.12) 1,025
0.036 (3.46) 314
0.425 (6.22) 961
0.011 (0.55) 147
0.623 (4.10) 719
Panel A: Earnings announcements λ(−10,−3) − λ(−40,−11) T-statistics No. of Obs.
1.018 (4.47) 140
0.228 (6.27) 538
Panel B: Acquisition announcements λ(−10,−3) − λ(−40,−11) T-statistics No. of Obs.
0.879 (3.26) 126
0.173 (4.25) 486
Panel C: Target announcements λ(−10,−3) − λ(−40,−11) T-statistics No. of Obs.
1.915 (5.04) 156
0.199 (5.06) 491
Panel D: Random high price change days (Average absolute return on t=0 is 2.64%) λ(−10,−3) − λ(−40,−11) T-statistics No. of Obs.
1.867 (3.45) 194
0.222 (3.64) 378
40
