Earning Announcements, Limited Attention
and Managers: How Are They Connected?
University of Amsterdam Msc Business Economics Track: Finance Master Thesis Sebastiaan Höring 10868712 06-2015 Supervisor: dr. S. Arping
Statement of Originality
This document is written by Student Sebastiaan Höring who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract:
The market responds to earning announcements, but the market reaction to these announcements differs for different days. Friday announcements are already characterized as disclosures of news that receive less immediate reaction than announcements on other days. In this thesis I will show that Monday announcements contain similar characteristics. Negative earning surprises, defined as when the forecast EPS is lower than the actual EPS, are also clustered on Mondays. The immediate response to earning announcement is lower on Mondays compared to the rest of the week. This will be compensated after some time. Managers know this and might exploit this knowledge in order to keep the stock price high. Other incentives can also explain why there is a strategy behind announcing negative surprises on certain days. Furthermore I show that negative announcements are also clustered in days when there are more negative announcements.
Table of Contents
I. LITERATURE REVIEW ... 8
II. DATA AND METHODOLOGY ... 11
III. BASIC STATISTICS ... 14
IV. RESULTS ... 17
A.MARKET RESPONSIVENESS ... 17
B.BUY-AND-HOLD REGRESSIONS ... 20
B.1.IMMEDIATE RESPONSE ... 21 B.2.DELAYED RESPONSE ... 24 C.CLUSTERING ... 25 C.1.CLUSTERING BY WEEKDAY ... 25 C.2.CLUSTERING BY CALENDAR DAY ... 30 V. ROBUSTNESS ... 34 VI. CONCLUSION ... 36 BIBLIOGRAPHY ... 39 APPENDIX ... 42
In an early paper Cutler, Poterba and Summers (1989) found that the arrival of news is an important driver of changes in stock prices. It is not necessarily the case that “big news” creates stronger market reaction than news on a much lower scale. To go even further, they found that there is a relatively small market response to major political- and world events. Previous studies show that share prices react to all kinds of announcements, such as corporate control announcements, macro economic conditions and regulatory policy changes.
In an efficient financial market, stock prices should immediately adjust to earning announcements. The presence of liquidity traders prevents this framework from working accordingly. The importance of small-return predictability should not be underestimated as it may yield significant returns (Xu, 2004).
Small return predictability is a powerful tool that can be manipulated by market participants to achieve certain goals. Investor inattention may cause underreaction to announcements. Managers know this and can use this fact to satisfy their own desire. Dellavigna and Pollet (2009) found that Friday earning announcements led to a lower response compared to other weekdays. They made a distinction between these two groups and found significant results indicating a delayed response to Friday announcements. The economic rationale for performing such a study is that people get distracted on Fridays, because weekend is near.
Investor inattention has been an interesting topic for some time already. Merton (1993) introduced his model in which investors receive some information on some stocks. He found that stocks that have been traded less than other stocks sell at a discount. Underreaction to information could explain why distraction increases the drift in stock prices. Hong and Stein (1999) argue that this drift is caused by limited cognitive abilities.
This drift can also be caused on purpose, when managers deliberately keep information to themselves for some time. Doyle and Magilke (2012) found that managers announce negative news on purpose on days in which investor inattention is most likely. Strategically disclosing information related to earning announcements is interesting for managers for several different reasons. Amongst others, gaining private internal benefits (Niessner, 2015), anticipating on negative announcements earlier in
the week (Bagnoli, Clement & Watts, 2005) and pleasing the CEO, because his reputation closely depends on the value of the stocks of the company (Bergstresser & Phillipon, 2006).
This thesis will analyze whether there is a post-weekend effect causing investors to underreact to Monday announcements. On Mondays investors may be distracted from the arrival of news because people like to socialize and tell their colleagues what they did over the weekend, or because they have to go through their mail and have other important tasks. In the sample I have used there are about twice as much announcements on Monday as there are on Friday, while in another study Hirshleifer, Lim and Teoh (2009) found that market reaction becomes weaker when there are more earning announcements, because investors have limited capacity of processing this information.
The aim of this research is to go beyond the perception that inattention to earning announcements is only subject to Friday announcement. I will analyze what the effect is of earning announcements on Monday to stock prices. I will analyze this effect using different approaches; the first one is comparing Monday to all the other days combined in one group and the other is comparing all the individual days of the week to each other.
The central puzzle this thesis is trying answer, is whether Monday announcements contain characteristics of stocks that receive less immediate reaction, and if so, whether managers exploit this knowledge to their interest in the sense that they cluster negative earning announcements on these days?
Previous studies have solely focused on Friday announcements, whereas I expect that investors also pay less immediate attention to Monday announcements. Since I have found results indicating that managers indeed announce negative news relatively often on days that investor inattention is more likely, I expect that managers cluster negative earning announcements on Mondays and Fridays as opposed to other weekdays.
Based on the fact that more announcements call for more distraction, there is another clustering effect going on. Managers can also cluster negative announcements
in days when there are more announcements as to hide information among many other announcements.
This research is a valuable contribution to the existing literature because it provides a broader view of the phenomenon of investor inattention theory. Previous studies were right for assigning Friday as the day where inattention is very likely, however I will show that this concept should not be underestimated for Monday announcement. Furthermore I show that announcements are clustered by both weekdays as well as amount of announcements in one day.
Announcements right after the weekend on Monday will be compared to announcements on the rest of the week. I have specifically chosen to include the recent global financial crisis in my sample period, because there are more relevant announcements during difficult times for investors to consider, which may enhance overall distraction. So the sample period for this research is January 2007 to December 2012.
In the first section of this thesis I will go through theories of established literature on investor inattention and managerial incentives to cluster. The general thought related to these topics is that investor inattention causes managers to announce negative earning announcement more often on Friday to satisfy their own interest. Then section II goes through the necessary steps for obtaining the data I have used in order to do a proper analysis.
In section III I will provide some basic statistics and show the differences in the immediate stock response and delayed response between Mondays relative to other days. I show that the likelihood of an announcement on Monday to be a negative earning announcement is higher than a positive one, even though there are more positive surprises in absolute numbers.
Section IV contains the main results of this research. In the first part of this section I will show that negative earning announcement are announced on days when investor inattention is most likely. The second part of this section discusses clustering by weekdays and calendar days. There are two factors causing earning surprises on Mondays and Fridays to be negative, namely that inattention is higher on these days and that, to some extent, more (negative) announcements attracts other managers to
do the same. In section V I will provide robustness checks and provide ideas for others how to further develop this thesis. I will finalize with a conclusion in section VI, where I provide a summary of the main findings in order to answer the main research question.
I.
Literature review
Earning announcements are required to be announced at least four times a year, namely three times in the end of the quarter reports (10-Q’s) and one time in the end of the fiscal year report (10-K). The SEC requires companies to file the 10-Q’s no later than 45 calendar days after the end of the quarter and the 10-K no later than 90 calendar days after the end of the fiscal year.
The reason that earning announcement theory is an interesting field to study is that we gain knowledge of what kind of announcements we can expect on which days, why these announcements are made and who can benefit from it. Information flows are fundamental for an efficient market. Since the fact that the SEC regulates the conditions earning announcement disclosure have to fulfill, the government acknowledge the importance this matter. The SEC adopted the Regulation Fair Disclosure in 2000 in order to stop selective disclosure and to structure the process of earning announcement disclosure (Gomes, Gorton & Madureira, 2007). They also found that this implementation had unintended consequences for small firms, which now faced higher cost of capital.
The SEC also sets requirements for earnings management with respect to accounting standards. So firms not only time the disclosure date, they can also manipulate the actual content of the announcement. Dechow, Sloan and Sweeney (1996) found that the most common motivation for manipulating the content is to gain increased access to capital at lower cost. However they did not find evidence of increased private benefits for the managers themselves.
Analyzing earning announcements and what causes distraction has been a relative widely invested theme in finance. Friday has often been stated in literature as the day when investor inattention is likely (Michaely, Rubin & Vedrashko (2014) &
Damadoran (1989)). The first traces of after announcement price drift date back to Patell and Wolfson (1982). While the incentives for managers to announce negative earning announcements after the market closes might not be entirely clear, Patell and Wolfson were the first researchers that identified this effect empirically. They argued that this strategic was to disseminate information over time.
The drift in stock prices, after earning announcements are made, is caused by the size of the surprise (Foster, Olsen & Shevlin, 1984). Foster et al. provided evidence relating to the size of the surprise; if the surprise is more negative, then the abnormal returns become more negative in the period after the announcement vice versa. This idea is captured in the remainder of this thesis, because I have used a variable categorizing the magnitude of the surprise, I will elaborate on this in section II. In this early research abnormal returns were used to estimate the drift in the stock prices.
To the extent that managers are able to choose the announcement date strategically, Penman (1987) examined seasonality in announcements. Penman found that firms are more likely to announce positive news in the first two weeks of the second and forth calendar quarters. Economic clarification for this phenomenon is that firms prefer to announce good news first, causing bad news to be announced in later weeks of these quarters.
It is hard to prove that managers strategically disclose information on days when investors do not pay much attention to earning announcements. Niessner (2015) tried to investigate whether investors have incentives to announce negative earning announcements on Fridays. As a reason for the strategy to announce negative announcements on Friday, he argued, that underreaction to this news arrival left time for the managers to sell shares of their own stock. This way they could benefit from distracted investors. Firms are obliged to disclose important information within four business days after they found out about this information by the SEC. He used the filings in the 8-K to investigate whether firms could have a strategy for announcing good and bad news on certain days.
Another reason why managers have a strategy behind earning announcement disclosure is pleasing the CEO in order to increase to stock price because his
reputation closely depends on the value of the stocks of the company (Bergstresser & Phillipon, 2006). This in turn might be accompanied by an increase of their own benefits in term of increased salaries or bonuses. However this has not yet been backed up by empirical research.
Yermack (1997) found that CEO’s sometimes receive stock options of the company and this is often in times before these companies announce positive earning announcements. Managers themselves do not benefit from their news announcements in the sense that they use their private knowledge to buy shares or options of the company. Givoly and Palmon (1985) studied this phenomenon. They found no evidence of insider trading based on positive announcements of the company. Private benefits from insider trading are not related to disclosure of certain events of the company.
A more recent research showed that insider trading is also difficult since especially CEOs are being followed closely and that makes them more cautious (Fidrmuc, Goergen & Renneboog, 2006). Also directors are not allowed to trade before the announcement date. The punishment for the firms when managers or directors do act based on their private information raises the cost of capital of the firm significantly. This should refrain them from violating the SEC laws.
Skinner (1997) examined earning announcement theory from another angle and researched whether announcing bad news early might actually be positive for the firm. He found results indicating that bad earning announcements best be disclosed early as to ensure that they do not resolve into stockholder litigation. He also found that the magnitude of the earning announcements affects the decision on when to announce certain events. The more negative the event was the sooner managers were willing to make it public.
In his earlier work (Skinner, 1994) he found that firms are not always voluntarily disclosing information related to earning announcements. Negative news has a stronger effect on the stock price as positive news, which is why negative news more often announced as soon as possible in order to limit the negative consequences of withholding information. Also managers announce negative earning surprises
early because they want to keep good relations with their investment community, this way they show their best interests.
Magilke and Doyle (2006) formed a hypothesis that announcing negative earning announcements on Fridays could diminish the market penalty caused by limited attention. Another strategy is announcing after the market closes because managers desire more time for the market to anticipate on these announcements. The exact benefits for the managers themselves from this strategy are not clear.
The focus of this thesis is not to answer the phenomenon of why managers choose to disclose information on a certain time or day, but whether it happens on both Fridays and Mondays in my sample period. Throughout history this effect has already been confirmed for Fridays, however no one has looked at whether this effect also (partly) holds for Mondays. Therefor I will provide evidence that there are signs of clustering negative earning announcements on both Fridays and Mondays, and even on days when there are more announcements, because more news leads to more distraction (Hirshleifer, Lim & Teoh, 2009).
DellaVigna and Pollet (2009) provided evidence that negative earning announcements are more likely to be announced on Fridays. The immediate market reaction to Friday announcements is lower compared to other days. This initial underreaction will be compensated with a higher delayed market reaction. This confirmed their inattention hypothesis; it is indeed the case that investors do not pay as much attention to earning announcements relative to other days of the week.
Throughout this research I will use some of the proxies and the approach as in DellaVigna and Pollet (2009) and Niessner (2015) to examine whether Monday announcements are more likely to be negative compared to other days. This is not the only clustering phenomenon this thesis is trying to identify. I will also investigate whether there are more clustering trends in this sample period.
II. Data and Methodology
This thesis uses US companies in the sample period from 2007 to 2012. The global financial crisis of 2007-2008 is involved in this research, because this might
yield interesting results and it creates extra distraction for investors, which can result in different behavior and results then prior research predicts. I have used US listed companies because data for these companies are readily available and there are sufficient listed companies in the US to do a proper research.
The empirical findings in this paper are built upon an event study. I will go in detail through the necessary steps I have taken to come to my final sample, but first I will sketch an image of what kind of data is needed for this research. Since I am doing an event study, accompanied by OLS regressions, I will need stock prices of some set of firms. In this case US listed companies. I want to know whether investors react differently to different announcements dates: Monday announcements versus announcements on the rest of the week. From here on I will need abnormal returns in different event windows.
The other important instrument for this research, besides the day of the week, is the surprise effect of an announcement. If the “expected” earnings per share are higher than the actual earnings per share, investors are negatively surprised and vice versa. It is very likely that investors do not react in the exact same way to equal positive as to negative surprises. This is the reason I decided to distinguish between positive and negative announcements.
Earnings data are provided by the I/B/E/S database from the WRDS website. Then the first step is to obtain forecasted earnings per share on a quarterly basis, where only consensus data are kept for which at least 3 analysts have submitted forecasts in order to have reliable forecasts. If there are multiple forecasts for a given firm and quarter only the most recent forecast for a given analyst is kept. I deleted older forecasts for a given firm and quarter of the same analyst because the newest forecast contains most of his wisdom and knowledge, which should be the best possible forecast subject to his capabilities. The consensus forecast can then be described as the median forecast for a given quarter and firm. I will eliminate weekend announcements made on Saturday and Sunday, because there are too little of them to analyze them separately and have significant results.
The next step is to obtain actual earnings per share, which are also obtained from the I/B/E/S database; I have deleted announcements that occurred during
weekends. Furthermore I have deleted both penny stocks as well as stocks with a market value of equity below five million dollars.
The next step is to get stock prices and returns from the CRSP database. Stock prices are merged on the consensus forecast dataset based on ticker and a variable that represents 5 working days prior to the announcement date, because this way I eliminate the effect of insider knowledge. Kothari (2001) has defined the earning surprise as the difference between the earning announcement and the consensus earnings forecast divided by the stock price five days prior to the announcement in quarter (t) of one share of a stock of the company in question (C)1.
𝑆𝑢𝑟𝑝𝑟𝑖𝑠𝑒!,! = 𝐸𝑃𝑆!,!− 𝐸𝑃𝑆!,!
𝑃𝑟𝑖𝑐𝑒!,! (1)
Once the earning surprises are known, I created 11 quantiles with surprises ordered from low to high, where quantile 6 contain announcements with no surprise. Quantile 1 through 5 contains negative surprises and quantile 7 through 11 positive surprises. Then I calculated the mean cumulative abnormal returns for each stock by quantile as the difference between the buy-and-hold returns of the individual stock minus the buy-and-hold returns of the market. More formally stated in a formula:
𝐵𝐻𝐴𝑅!,!!,! = 𝐵𝐻𝑅!,!!,!− 𝐵𝐻𝑅𝑀!,!!,! (2)
Where 𝐵𝐻𝐴𝑅!,!!,! stands for the mean buy-and-hold abnormal return in quarter (t) for company (C) from day (p), which is either 0 or 2, to (P) days after the announcement, which is either 1 or 70. Throughout this thesis negative announcements may be interchanged with negative earning surprise. Do not confuse negative earning announcement for negative announcement, as there is an important
1 EPS measures the actual earnings per share while 𝐸𝑃𝑆
!,! measures the
difference between the two. A negative earning announcement may yield a positive surprise, when the predicted EPS is lower than the actual EPS. I define a negative announcement as when the predicted EPS is higher than the actual earnings per share, and thus the consequence is a negative surprise in formula (1). In short, a negative announcement means a negative surprise throughout this thesis.
Finally I made several regressions based on data explained in this part of this thesis. Regressions are explained in the results part themselves, because this helps the reader to understand the text and results better.
III. Basic Statistics
First I will provide some basic statistics based on the sample I have collected and described in the previous section. This is interesting since these tables provide descriptions of the dataset, which are fundamental for understanding the result part. Previous studies focused on Friday as the optimal day to announce negative earning surprises, which basically excludes the idea of clustering among days with more announcements. Hiding news could be an option since the fact that more information is more difficult to process, hence underreaction. If managers would like to announce negative earning surprises on days when investors have less immediate attention to information, it also means that announcements on the other days should contain mostly positive announcements. The statistics presented are very much alike the statistics Dellavigna and Pollet (2009) found. This indicates that not much has changed with respect to the distribution of earning announcements over the week. They examined whether Friday announcements had a different impact on investor attention then on other weekdays.
I will both provide information on the distribution of announcements over the week as the distribution of negative versus positive earning surprises in Table 1. First of all I present the amount of observations, then you can find this amount in a percentage of total. Finally I will provide the percentage of negative earning surprises of total announcements per day.
In Table I it can be easily seen that most of the earning announcements occur on Tuesday, Wednesday and Thursday, these days count for 82% of total announcements. On Monday there are relatively little announcements compared to these days, however Mondays have twice as much announcements as Friday has. The amount of negative surprises as a percentage of total is highest for Friday announcements, which is probably the reason why other researchers have focused on this day. The gap between Friday announcements and Monday announcements with respect to the percentage of negative earning surprises is about the same as the gap between Monday announcements and announcements made on the rest of the week.
Table I: Distribution of Earning Announcements
In Table I the distribution of earning announcements over the week is provided. The sample period is January 2007 – January 2012. Quarterly earning announcements are gathered from the I/B/E/S database and are merged with stock prices from 5 working days prior to the announcement gathered from the CRSP database. Finally weekend announcements and penny stocks are deleted from the sample.
Distribution of Earning Announcement by Day of the Week Monday Tuesday Wednesday Thursday Friday Total Observations 7140 13604 15955 19860 3441 60000 Percentage 11,90% 22,67% 26,59% 33,10% 5,74% 100% % Negative
announcements 0,39 0,36 0,35 0,36 0,42 So most of the earning announcements are made on Tuesday through Thursday. Friday is the greatest exception in this case; only 6 percent of total announcements in this sample are made on Friday. Another interesting question is whether announcements are made on purpose on a certain day. Niessner (2015) has found results indicating that managers strategically disclose the date of the announcement in order to gain private benefits. Niessner used the 8-K files (current reports) companies must file by major events. Ever since 2002 the SEC required firms to announce more events and in a shorter time period. Nowadays firms have 4 working days before they have to publicly announce certain events. This means that announcements can be disclosed as the manager thinks is in the best interest of the
firm (of course within 4 working days). He also found that negative announcements are usually made when investors are more distracted. So if the hypothesis of this thesis is correct, then most negative announcements should be made on Mondays and Fridays.
Since we have defined the earning surprise, we can categorize the announcements from very negative surprises (indicated as low), to very positive surprises (indicated as high), divided over 11 quantiles. Quantile 1 through 5 contain negative surprises; quantile 6 is when there is no surprise, while quantile 7 through 11 contain positive surprises.
Table II presents the average earning surprise within each quantile for both the Monday and the non-Monday announcements in percentages. Quantile 2 through 11 are almost alike for Mondays and non-Mondays, however the average earning surprise between Mondays and non-Mondays in quantile 1 differs substantially. Furthermore it can be seen that there are more positive surprises relative to negative earning surprises.
However even though there are more positive surprises relative to negative surprises, the average earning surprise of the entire sample is negative (-0,18%). Furthermore we are mostly interested in the quantile(s) for which the gap is greatest between Mondays and the rest of the week. Finally we can see in table II that the difference for positive surprises between Mondays and other days is not so large as the same difference for negative surprises. This is the reason why I will use these quantiles as the base for my empirical analysis.
Table II: Average Surprise by Quantile
Table II shows the average surprise by quantile on Mondays and non-Mondays in percentages. Quantile 1 to 5 contain negative earning surprises, quantile 6 contain announcements with no earning surprise, and quantile 7 through 11 contain positive earning surprises. In this sample there are relative more positive earning surprises than negative earning surprises.
Average Surprise by Earning Surprise Quantile Mondays Non-‐Mondays
Quantile Average N Average N Difference Low surprise 1 -‐6,28 583 -‐4,68 3805 -‐1,598 2 -‐0,71 605 -‐0,70 3784 -‐0,008 3 -‐0,32 538 -‐0,32 3849 0,001 4 -‐0,15 533 -‐0,15 3855 -‐0,001 5 -‐0,05 513 -‐0,05 3874 -‐0,002 Zero surprise 6 0,00 579 0,00 4245 0,000 7 0,05 703 0,05 5950 0,002 8 0,11 752 0,11 5890 0,001 9 0,21 701 0,21 5948 0,001 10 0,39 774 0,39 5873 0,005 High surprise 11 1,68 859 1,62 5787 0,059
IV. Results
A. Market ResponsivenessThis subsection measures whether investors react differently to announcements on different days. The immediate response to earning surprises by the different quantiles is projected in Figure 1a. It is clear to see that on Monday investors are initially distracted as they underreact to earning announcements compared to other days, mostly for very negative earning surprises. The mean cumulative abnormal return in the event window from the day of the announcement to one day after the announcement (the immediate response) for Mondays is not as smooth as it is for non-Mondays. For quantiles 5 through 10 the market response is almost equal, however for other quantiles investors underreact to market announcements on Mondays.
Then if investors are distracted initially, they should respond more later on. Moreover to compensate for this initial underreacting for quantiles 1 through 4 and 11
and 12, investors will have to respond even more in later stages. I will show this in the Figure 1b.
The mean cumulative abnormal return for announcement with no earnings surprise (quantile 6) should be very close to zero, given that no new company related information was revealed. Both in Figure 1a and b the mean cumulative abnormal return is indeed around zero in quantile 6. Furthermore if it is indeed the case that there is a delayed market response to the earning announcement, Figure 1b should be (close to) the inverse of Figure 1a.
Figure 1a and b: Immediate and Delayed Stock Responsiveness
Figure 1a and b show the average surprise by quantile on Mondays and non-Mondays. In Figure 1a the market reaction in the short term is measured by quantile, while in Figure 1b the same effect is provided for the long term. Sample period is January 2007 – December 2012. CRSP stock prices are linked to I/B/E/S quarterly earning announcements. Earnings surprise is calculated as difference between forecast and actual value divided by the stock price of exactly one week before the announcement. These surprises are divided over 11 quantiles where quantile 1 through 5 contain negative earning surprises, quantile 6 is a zero surprise and quantile 7 through 11 contain positive surprises.
(b)
In Figure 1b the mean cumulative abnormal return from 2 working days after the earnings announcement to 70 working days after the earnings announcement by quantile is projected. The results in Figure 1b are more volatile than in Figure 1a for both Mondays and non-Mondays. But by comparing quantile by quantile, Figure 1b shows almost for every quantile the reverse result as in Figure 1a. The market reaction in Figure 1b is the opposite of Figure 1a for all quantiles except for quantile 2 and 11. This indicates that there are signs of a delayed market reaction to the earning announcements. These results indicate that on Mondays investors are initially more distracted or at least do no react as much to earning announcements compared to other weekdays. However investors compensate this inattention later on, when the market reaction to the same earning announcement compensates for the initial underreacting.
Finally there is a difference in the scale of the y-axis between Figure 1a and b. The scale of the market responsiveness between Monday and non-Mondays are smaller in Figure 1a, which is because of the shorter time period used. So for managers to somehow gain private benefits, very negative earning surprises and very
positive surprises are most interesting, because in these quantiles the differences are greater than when the surprise becomes close to zero.
I will also include a table to show whether the relationship described above is statistically significant so that we can actually interpret the results. These findings can be found in table III.
Table III shows us that variation in the mean cumulative abnormal return for the delayed market reaction (regressions 3 and 4) can be less explained by variations in the quantiles. Even though the lines in figure 1a and b are statistically significant, this table indicates that the effect of changes in the earning surprises have a stronger effect on stock prices in the short run. In the next session I will elaborate this relationship and further develop this model.
Table III: Stock Responsiveness to Monday Announcements
Table III shows the average surprise by quantile on Mondays and non-Mondays. This table displays the regression based on figure 1a and b. Regressions 1 and 3 represent Mondays whereas 2 and 4 represent non-Mondays. T-statistics are reported in parentheses (* significant at 5%, **significant at 1%).
Dependent Variable: Mean Cumulative Abnormal Return in Event Time from 0 to 1
Dependent Variable: Mean
Cumulative Abnormal Return in Event Time from 2 to 70 (1) (2) (3) (4) Quantiles 0.001 0.002 0.002 0.002 (9.06)** (35.40)** (17.81)** (33.19)** Constant -0.008 -0.013 -0.019 -0.014 (6.95)** (31.02)** (21.39)** (27.17)** R2 0.01 0.02 0.00 0.00 N 8.146 60.564 180.27 782.701 B. Buy-and-Hold Regressions
In order to measure the immediate and delayed response by means of a regression I have made several regressions impeding the effect of the quantiles with a low earning surprise. In Table IV I have presented 3 sets of regressions for two distinct dependent variables. The dependent variables are as in Figure 1a and b, namely the mean cumulative abnormal return for the period 0 to 1 and 2 to 70, where
zero stands for the day of the announcement. The first three regressions should estimate the immediate response, while the latter three should estimate the delayed response. The specifications for these regressions are as follows:
𝐵𝐻𝐴𝑅!,!!,! = ß! + ß!∗ 𝑀𝑜𝑛!,!+ ß!∗ 𝑄!,! + ß!×(𝑄!,!∗ 𝑀𝑜𝑛!,!) + + 𝜀!,! (3) 𝐵𝐻𝐴𝑅!,!!,! = ß! + ß!∗ 𝑀𝑜𝑛!,!+ ß!∗ 𝑄11!,!+ ß!×(𝑄11!,!∗ 𝑀𝑜𝑛!,!) + 𝜀!,! (4) 𝐵𝐻𝐴𝑅!,!!,! = ß! + ß!∗ 𝑀𝑜𝑛!,!+ ß!∗ 𝑄𝑇𝑜𝑝2!,! + ß!×(𝑄𝑇𝑜𝑝2!,!∗ 𝑀𝑜𝑛!,!) + 𝜀!,!
(5)
Where 𝐵𝐻𝐴𝑅!,!!,! stands for the mean buy-and-hold abnormal return in quarter (t) for company (C) from day (p), which is either 0 or 2, to (P) days after the announcement, which is either 1 or 70. Specification (3) contains announcements for all quantiles, while specification (4) contains announcements for the bottom and top quantile, whereas specification (5) contains announcements for the top two and bottom two quantiles. Mont,c is a dummy variable indicating whether the earning
announcement is made on a Monday (𝑀𝑜𝑛!,! = 1). Qt,c is a variable indicating in
what quantile the earning surprise for company (C) is at quarter (t). Q11t,c is a dummy
variable indicating whether the earning surprise of company (C) in quarter t is in the top quantile or not. The same logic applies to Qtop2; in this case it is about whether the earning announcement is in quantile 10 or 11. Standard errors are clustered by date because this adjusts for correlation of the buy-and-hold returns of different companies on the same day. The most interesting variables for determining whether there are differences in market reaction to announcements between Mondays and non-Mondays are ß1 and ß3.
B.1. Immediate response
For the immediate response regressions we have a different dependent variable then for the delayed response regressions. The immediate response regressions require the mean cumulative abnormal return to be measured in the event window of the day of the earning announcement to one day after the announcement.
In order to measure the immediate response and especially the underreaction to Monday announcements we have to look at ß3 from formula (4). Note that formula
means the formula in the text while regressions refer to the output in the tables. If we look at the second regression (2) in Table III we can calculate the initial underreaction by taking the difference between Quantile 11 and Quantile11*Monday, which is 0.008/0.021=38%. This difference is statistically significant, but seems quite large. For regressions (1) and (3) this underreaction is respectively 50% and 36%.
From the first regression we can see that the mean cumulative abnormal return is a linear function of the different quantiles. This is in line with our earlier findings in Figure 1a. For Mondays in each quantile the market reaction is less than for any other day (-0,001). The interesting regressions are (2) and (3) actually, since these regressions actually compare positive surprises to negative surprises. Regression (2) compares the bottom to the top earning surprises, which means the most positive earning surprises against the most negative surprises. Regression (2) tells us that if an announcement is made on Monday the market reaction becomes close to zero, and thus there is less immediate reaction. Moreover if an announcement is positive news, the buy-and-hold returns become a relatively small positive value (-0.01+0.021), because a small positive value gets added to the negative constant.
The third regression (3) basically yields the same results as regression (2). The difference with respect to regression (2) is that constant is closer to zero, however if a certain announcement is in the top two quantiles and is made on Monday, the total effect is the same for both regressions. If an announcement is negative, the total effect for regression (2) is slightly more negative than the same characteristics in regression (3), a difference of (–0.001).
Table IV: Stock Responsiveness to Monday Announcements
Table IV contains ordinary least squares regressions on the stock price response to earning announcements. Sample period is January 2007 – December 2012. CRSP stock prices are linked to I/B/E/S quarterly earning announcements. Earnings surprise is calculated as difference between forecast and actual value divided by the stock price of exactly one week before the announcement. Observations are reported at company level. Dependent variable is the mean cumulative abnormal return for different event windows as explained in the Table. Monday is a dummy variable indicating whether the earning announcement is made on a Monday. Quantiles is a variable indicating in what quantile the earning
surprise for company C at quarter t is in. Quantile11is a dummy variable indicating whether the earning
surprise of company c in quarter t is in the top quantile or not. Quantile 10 & 11 indicates whether the earning announcement is in quantile 10 or 11. Regressions (1) and (4) are based on the entire sample, regression (2) and (5) on the top and bottom quantile and finally regression (3) and (6) on the bottom and top 2 quantiles Standard Errors are adjusted for clusters and heteroskedasticity. T-statistics are reported in parentheses (* significant at 10%, ** significant at 5%, ***significant at 1%).
Dependent Variable: Mean Cumulative Abnormal Return in
Event Time from 0 to 1
Dependent Variable: Mean Cumulative Abnormal Return in
Event Time from 2 to 70
(1) (2) (3) (4) (5) (6) Constant -0.013 -0.01 -0.009 -0.014 -0.029 -0.019 (-24.62)** (-9.68)** (-12.98)** (10.39)*** (10.70)*** (10.85)*** Monday 0.006 0.005 0.005 -0.005 -0.006 0.000 (4.59)*** (2.25)** (3.51)*** (3.90)*** (2.17)** -0.11 Quantiles 0.002 0.002 (32.19)*** (32.19)*** Quantiles* Monday -0.001 0.000 (4.57)*** -0.84 Quantile 11 0.021 0.039 (19.14)*** (15.09)*** Quantile11 *Monday -0.008 0.008 (3.11)*** (2.15)** Quantile 10&11 0.019 0.027 (25.58)*** (16.72)*** Quantile 10&11* Monday -0.007 -0.004 (4.45)*** (1.66)* R2 0.03 0.05 0.04 0.00 0.01 0.00 N 68,710 12,276 24,682 943,041 166,727 337,035
B.2. Delayed Response
In the case of the delayed response regressions, the dependent variable is the mean cumulative abnormal return measured in the event window of two days after the earning announcement to 70 days after the announcement. If we look at regression (4) we see that the function has become statistically insignificant. Therefor this regression will not be interpreted. In order to see whether there is a delayed response we look at the results and then we see that the output tells us two different stories. In case of regression (5) the delayed response to Monday announcements is approximately 21% (0,008.0,039). Whereas regression (6) tells us that in the top two quantiles with respect to the bottom two quantiles there is evidence of underreaction of about 15%.
The fact that regression 6 seems to indicate that there is no delayed response comes from the fact that in Figure 1b quantile 10 is not the opposite of the same quantile in Figure 1a. This indicates that for these earning surprises there is no evidence of a delayed response whereas the effect does hold for the top quantile. Also then R2 equals zero for this regression, so the independent variables do not explain changes in the dependent variable very well for this regression. Even though the R2 for regression (5) is not much higher, this regression better explains variations
in the delayed stock response.
In regression (5) we observe that if an announcement is made on Monday and is positive news, the effect of ß3 in the long run is the inverse of the short run effect.
Which means that a positive announcement gets a higher market return and a negative announcement a higher (negative) market return. This indicates that there is indeed a delayed market reaction to Monday announcements.
So the evidence of an initial underreaction to earning announcements placed on Mondays and an increased delayed reaction to the same announcements is not as clear as expected, in the sense that the R2 is relatively low, however the data tells us that are signs of underreaction and of a delayed response. The next step is to analyze whether managers use this knowledge for some reason. I will do this in the next section by looking whether negative announcements (negative surprises) are clustered at a specific date or day. Finally you can see in Table IV that the dataset used for the
different regressions is different by looking at the total number of observations in each regression. Regression (1) and (4) contain all the observations for different dependent variables. Because the first three regressions are based on the immediate return, the amount of observations is way smaller than in the latter three regressions.
C. Clustering
C.1. Clustering by Weekday
For this section we analyze whether clustering occurs by weekday. This can be measured by taking all the separate days in my sample period and count the amount of announcements made in one day. Figure 2 shows that the likelihood of a positive announcement increases with the number of announcements in one day. This means that in this sample, managers announced negative earning announcements more often on days when there are few other announcements, perhaps because investors are distracted on these days. From table I we have seen that on Fridays and Mondays the amount of announcements is relatively low compared to other days, so the chance that on these days an earning announcement is negative is higher than for other days. Following the line of reasoning of Niessner (2015) and DellaVigna and Pollet (2009), on days when investors are distracted there should not only be small amounts of announcements, but also most of them should be negative earning announcements. This would mean that managers have incentives to use the fact that investors have limited attention to earning announcements on Mondays and therefore announce negative surprises more often on these days.
Hirshleifer, Lim and Teoh (2009) suggested that when there are more announcements in one day, this could also cause distraction and thus possibly underreaction to these earning announcements. Managers could cluster negative announcements also on these days. We will go through this framework later on.
Figure 2: Mean Surprise by Amount of Announcements by Day
In Figure 2 the amount of announcements in one day are plotted against the mean surprise per announcement date. First categorize the dataset by announcement date, calculate the mean for every date and then the mean surprise of total sample by number of in one day.
First in Figure 3 I will present a graph where you can find the amount of announcements in one day per weekday. Each bullet in this figure represents a single day. In Figure 3 Monday is represented by 1, Tuesday by 2, Wednesday by 3, Thursday by 4 and Friday by 5. Thursday is the day that the most announcements take place in one day. This is in line with our findings in Table I, which stated that the most announcements over the entire sample period took place at Thursday. Now we found that also the most announcements in one day take place on Thursday, increasing the likeliness of positive announcements on these days.
There are relatively little announcements in one day on both Monday and Friday. This combined with Figure 2 suggests that it is very likely that announcements on Monday and Friday are negative earning announcements, since negative announcements are most likely to be announced on days when there are little
other announcements (Figure 2) or when investor inattention is highest (Niessner, 2015). This is actually the contrary of what Hirshleifer, Lim and Teoh (2009) predicted. This does not mean that their results are false; it only means that their findings do not apply to this case.
Figure 3: Amount of Announcements in One Day by Weekday
In Figure 3 the amount of announcements in one day are plotted against the day of the week. Monday is represented by 1, Tuesday by 2, Wednesday by 3, Thursday by 4 and Friday by 5.
Niessner (2015) came up with a regression that analyzed whether there is a higher chance on a certain day that an announcement is a negative announcement, in his case applied to Friday announcements. He found that on average managers are 2,5% more likely to announce negative earning surprises on Friday. The following regression can be used to examine this relationship:
I will use the formula to analyze whether Monday announcements contain relatively more negative earning announcements with respect to other days. First I will provide a table containing these results with several different sets of fixed effects. The findings can be found in Table V.
Table V: Negative Announcements on Mondays
Table V contains ordinary least squares regressions on whether it is more likely that managers disclose information on Mondays. Observations are reported at company level. The Negative Announcement dummy equals 1 for negative announcement surprises zero otherwise. The dependent variable is a dummy variable that equals 1 for announcements made on a Monday, zero otherwise. I have used different sets of fixed effect. Intercepts are not relevant. Standard errors are clustered and t-statistics are reported in parentheses (* significant at 5%, ** significant at 1%).
In column (1) I found that on average managers are more likely to report negative announcements by 0,9% relative to non-negative announcements. Since the R2 does not differ between columns (1), (2) and (3), I will apply the regression as in column (1) to all other weekdays in order to examine whether it is indeed the case that managers announce relatively more negative earning announcements on Mondays and Fridays than on other weekdays. I will use the regression of column (1) because economically it makes the most sense. Even though the fourth regression differs substantially from the others with respect to its main coefficient, column (4) has the lowest R2 so we will neglect this one as well for further investigation.
Dependent Variable: Dummy Variable for Whether Information Was Disclosed on a Monday
(1) (2) (3) (4)
Negative Announcement 0.009 0.009 0.009 0.011
(3.36)* (3.01)** (3.41)** (3.13)**
R2 0.42 0.42 0.42 0.00
N 59,998 59,998 59,998 59,998
Year FE Yes No Yes No
Month FE Yes Yes No No
Day of Announcement FE Yes Yes Yes No
Table VI: Negative Announcement on Rest of the Week
Table VI contains ordinary least squares regressions on whether it is more likely that managers disclose information per weekday (as can be seen in each column). Observations are reported at company level. The Negative Announcement dummy equals 1 for negative announcement surprises zero otherwise. The dependent variable is a dummy variable that equals 1 for announcements made on a specific weekday, zero otherwise. I have used different sets of fixed effect. Intercepts are not relevant. Standard errors are clustered and t-statistics are reported in parentheses (* significant at 5%, ** significant at 1%).
Dependent Variable: Dummy Variable Day of Disclosure
(Tuesday) (Wednesday) (Thursday) (Friday)
Negative Announcement -0.004 -0.011 -0.004 0.009
(0.86) (2.04)* (0.87) (4.03)**
R2 0.23 0.00 0.41 0.20
N 59,998 59,998 59,998 59,998
Year FE Yes Yes Yes Yes
Month FE Yes Yes Yes Yes
Day of Announcement FE Yes Yes Yes Yes
Although not all results in Table VI are significant (Tuesday and Wednesdays should not be interpreted), the results indicate that it is indeed true that announcements on Friday are more likely to contain negative earning announcements relative to positive earning announcements. The opposite is indicated for the other days, since the coefficient on the variable Negative Announcement is negative for all these days. Keeping in mind that these coefficients should not be interpreted, as they are statistically insignificant, the results provide wisdom to some extent.
The conclusion of this subsection is that clustering by weekday is definitely present in this sample period. Managers are more likely to announce negative surprises on Mondays and Fridays. The data also provided evidence of the opposite effect, namely that positive surprises are more likely to be announced on days when there are more other announcements, as is the case on Tuesdays, Wednesdays and Thursdays. The next subsection will focus on calendar dates instead of weekdays, this way I would like to exclude that
C.2. Clustering by Calendar Day
In this section I will analyze whether managers cluster negative earning announcements in days where many other announcements are made. So in principle this means that managers deliberately hide their negative earning announcement amongst other announcements in order to decrease their own exposure. This would mean that there is a possibility for managers to have the opposite strategy of the weekday effect, because if this holds, managers also have incentives to announce negative surprises on Tuesday through Thursday.
I have used the following model to predict the results:
𝑀𝑆𝑢𝑟𝑝𝑟𝑖𝑠𝑒!,! = 𝛼 + ß! 𝐷1! + ß!𝐷2! + 𝐹𝑖𝑥𝑒𝑑 𝐸𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜀!,! (7)
The results are provided in Table VII. In formula (7) “MSurprise” is the dependent variable, which indicates the average surprise by calendar day2. In formula (1) you can find the way I have estimated the earning surprise. The intercept is captured by 𝛼. D1 is the dummy variable indicating to which group the amount of announcements belong. The groups are proportionally distributed according to their size; group 1 is the group where the lowest amounts of announcements are placed, while group 5 is the group with the highest amounts of announcements. D2 is also a dummy variable indicating which day of the week the announcement is made. Finally I have used size-, year- and month fixed effects to adjust for differences caused not by the main variables.
Since the main variables are all dummy variables, we view the results from the perspective of a reference group. Note that regression (2) is basically the same regression as (1), only from a different perspective. The difference between these regressions is that the output should be estimated from Monday’s perspective in regression (1) while in regression (2) Friday is the reference group from which we
2 Note that the sample upon which the dependent variable is estimated is
different for regression (3) and (4) i.e. in this case the sample includes negative surprises only.
estimate our results. The same logic applies to regression (3) and (4). I have included both regressions because this way the results are easier to compare and it is easier to analyze differences between Friday and Monday announcements.
Furthermore I have chosen to include regressions with only negative earning surprises in regression (3) and (4) to analyze whether there is difference in the magnitude of the negative earning surprise when there are more negative earning announcements. This output should be interpreted a bit different from regression (1) and (2). I would like to examine whether investors hide negative announcements amongst many other announcements, this effect is captured in the regression (1) and (2). Then if this holds, I would like to examine whether the surprise becomes even more negative, if there are more other negative announcements on a certain day, which is measured by regression (3) and (4). This sounds like a mathematical consequence of having more negative announcements, however it could be the case that the mean surprise becomes less negative. An example is if there are 10 negative surprises of -0.001 or 20 of -0.009, is does not mean that more negative announcements automatically results in a more negative mean surprise.
In regression (1) and (2) the constant is less negative than is the case for the other two regressions. This makes sense, since in this sample there are both negative as positive (and even zero) earning surprises. With respect to the amount of announcements in one day, the results indicate that for the first four groups (g1 to g4) there are only small differences between the average surprises. However for the largest group of announcements in one day, the surprise becomes more negative than in any group with fewer announcements in one day.
In regression (3) and (4) it is easy to observe that the more (negative) announcements are made in one day, the less gets added to the negative constant. This in turn indicates that the more negative announcements made in one day, the higher the average negative earning surprise. This means that managers allow themselves to hide higher negative earning surprises in days when there are relatively more negative announcements than in other days. So the higher the amount of negative announcements in one day, the more managers allows themselves to announce relatively high negative earning announcements.
Table VII: Negative Surprise and Calendar Dates
Table VII contains ordinary least squares regressions on whether it is more likely that managers disclose information on days when there are lots of other announcements. Observations are reported at company level. The dependent variable for the first two regressions is the mean surprise by number of announcements in one day. The main independent variables (variables g1 to g4) are dummy variables indicating in which group the amount of announcements belong. I have created five groups, where the amount of announcements is proportionally distributed from little to large amounts of announcements in one day over these five groups. The weekdays are dummy variables indicating whether the announcement was made on that specific day. I have adjusted for year-, size-, and month fixed effects. T-statistics are reported in parentheses (* significant at 5%, ** significant at 1%).
Dependent Variable: Mean Surprise by Number of Announcements in 1 Day of All
Announcement
Dependent Variable: Mean Surprise by Number of Announcements in 1 Day of Negative Announcement (1) (2) (3) (4) Group 1 0.0010 0.0010 0.0042 0.0042 (7.13)** (7.13)** (8.75)** (8.75)** Group 2 0.0012 0.0012 0.0031 0.0031 (9.61)** (9.61)** (7.41)** (7.41)** Group 3 0.0014 0.0014 0.0033 0.0033 (12.46)** (12.46)** (8.47)** (8.47)** Group 4 0.0010 0.0010 0.0015 0.0015 (10.03)** (10.03)** (4.41)** (4.41)** Monday 0.0012 0.0022 (8.35)** (4.50)** Tuesday 0.0017 0.0029 0.0043 0.0065 (15.35)** (20.44)** (11.60)** (13.80)** Wednesday 0.0021 0.0034 0.0050 0.0073 (19.40)** (23.52)** (13.61)** (15.37)** Thursday 0.0025 0.0038 0.0063 0.0085 (22.74)** (26.08)** (16.42)** (17.61)** Friday -0.0012 -0.0022 (8.35)** (4.50)** Constant -0.0060 -0.0072 -0.0181 -0.0203 (32.11)** (35.23)** (27.83)** (29.07)** R2 0.13 0.13 0.16 0.16 N 59,998 59,998 21,939 21,939
Year FE Yes Yes Yes Yes
Month FE Yes Yes Yes Yes
Based on Table VII we will have to conclude that there are no clear signs of clustering negative announcements in days when there are many other announcements. For the largest group of announcements in one day (group 5) the surprise is very negative, however the difference of the surprises in groups 1 through 4 do not show a clear pattern.
Even though this section was not intended to show clustering by weekday, this effect can be seen in Table VII as well. The effects of the weekdays are stronger and easier to observe. We observe that Fridays contain the most negative surprises of all weekdays; 0.12 percent lower surprise than on Monday. In turn Monday announcements are lower (or more negative) than announcements made on other days of the week except for Friday announcements. I have included regression (2) and (4) because these regressions best show that Friday announcements are most negative. In regression (2) we see that announcements on all other days are less negative, because a positive amount for each dummy variable gets added to the negative constant, making the overall effect less negative.
I will show the above results in a simple graph to illustrate the points made. I will use the difference between group 5 and group 1 to make the case, because the differences between groups 1 through 4 are relatively small.
Table VIII: Summary of table VII
Mean Surprise by Number of Announcements in 1 Day of All Announcement
Mean Surprise by Number of Announcements in 1 Day of Negative Announcement Only
Group 1 Group 5 Group 1 Group 5
Monday -0.005 -0.006 -0.0139 -0.0181
Tuesday -0.0033 -0.0043 -0.0096 -0.0138
Wednesday -0.0029 -0.0039 -0.0089 -0.0131
Thursday -0.0025 -0.0035 -0.0076 -0.0118
Friday -0.0062 -0.0072 -0.0161 -0.0203
If we look at the results in Table VIII two important features of this Table arise. The first one is that when there are more announcements in one day, the average earning surprise becomes more negative (left hand side of Table VII). Moreover,
when there are relatively more negative announcements, the effect becomes even stronger (right hand side of the Table). This table only shows the difference of the bottom to top group and thus slightly misrepresents the situation, because the middle part is quite volatile as can be seen in Table VII.
The second feature is the clustering by weekday. This part has already been elaborated in part 4 of this thesis, however Table VII shows this phenomenon perfectly. The average earning surprise is the lowest (most negative) for Fridays. This is in line with previous studies mentioned earlier on. Then there is a little gap between Friday and Monday announcements, which confirms my previous statements that investors also have limited attention to Monday announcements, and thus managers cluster negative announcements on Mondays too. Interacting all the terms seems to further improve the results at first sight, but doing this is actually a fallacy. This would only misrepresent the data since we have seen in Figure 4 that not all days are represented in the groups created. For example in Group 5, Friday announcements are not represented since there are never so many announcements in one day on Friday.
The reason that managers might choose Monday over Friday as a resort for their announcement to take place could be the fact that on Monday there are more announcements, hence more distraction. Probably the size of the surprise matters, very negative earning announcements are more likely to be published on Fridays.
V. Robustness
Future research related to this topic could study whether there is a weekend effect by using a large sample. Because of the short time period I have used, I deleted weekend announcements. However this might actually contain important information and this has, to my knowledge, not yet been examined.
Furthermore I think it is interesting to do interviews to better develop the understanding the motives of managers on when to announce what kind of earning announcement. Data does not tell us what the motives are behind choices, they only make us guess on what we think is plausible.
With respect to future research I think that the question of whether there is a delayed response to earning announcements on certain days is not the interesting question any more. Future research should focus on what the effects are of changes in SEC policy and link insider benefits to postponing the date of announcements.
Whether there are actually two clustering effects remains difficult to proof. Interviews could improve research on this topic. Clustering negative announcements on Fridays and Mondays does not necessarily exclude that managers also contemplate to “hide” negative news in days when there are many other announcements. In this sample the average surprise was negative, so it could be interesting to compare another time period to this thesis and see the differences.
Throughout this thesis I put several fixed effects in the regressions in order to see how this affects the output. For example for Table VII I chose to make the groups dummy variables, but I also tried to use just one variable. If this coefficient were to be positive it would simply mean that if the amount of announcements per day increases, the mean surprise would increase vice versa. This however resulted in insignificant results.
I also checked whether involving new variables increased the R2 without affecting the significance. Most of the times when correcting for fixed effects, the variables had very little impact on the dependent variable. Furthermore I adjusted results for clusters, because this could mislead the situation in the sense that more than one analyst provides consensus forecasts for a given firm, it is likely that another analyst provides around the same information. So this additional analyst does not give us as much information as the first one, however his added value is greater than zero. With clusters I have corrected for these problems. Moreover I frequently checked for extreme outliers and deleted them if applicable.
For the regressions in Table VII for example I tried increasing the different groups, however this did not improve the results. Furthermore I tried to exclude the middle groups to see the difference between the bottom and top groups, as in the regressions presented in Table IV. The effect was the same and shall be presented in the Appendix as Table A-1. The main result was again that the average surprise increases (becomes more negative) with the number of announcements in one day.
The sample only included data for the bottom and top group with respect to the amount of announcements per day. This however misrepresents the actual facts, because Fridays and possibly Mondays are not represented in Group 7, which affects the outcome significantly.
Basically robustness checks means adding variables and see what happens and whether it makes sense. In my case it would be ideal to exclude all incentives people have to trade a certain stock besides acting on the new information. Stock prices depend on so many different things that we cannot all take into account. I used fixed effects to account for some basic things, like size of the company. Also time is an important factor, in times of crisis investors react differently on the arrival of news, most of the times they are more cautious. I have tried to take these obvious effects into account as much as possible. However I could not take into account whether the given firm had a new CEO or merged with another company or such things, which would actually affect the stock price significantly and thus mislead the output of my data.
VI. Conclusion
Previous studies were right for focusing on Friday announcements in the sense that inattention to these announcements in the short run is indeed larger than for Mondays. However this phenomenon should not be underestimated for Monday announcements as I have provided evidence that there is indeed slower market anticipation to Monday announcements compared to announcements made on other days of the week. Even though I have not examined the incentives managers have to announce negative earning surprises when inattention is largest, I have examined whether different earning surprises are more likely to be announced on certain days. I also examined whether this effect is larger when there are more announcements in general.
First I presented basic statistics, which showed that the percentage of negative announcements is higher on Mondays and on Fridays compared to the rest of the week. The difference of the surprise between Mondays and the rest of the week were
