Breaking news: how headlines and returns affect investor buying
behavior in a small stock market
Studentnr: s2167476 Name: Tjarko Gorter Study Program: MSc. Finance
Supervisor: Dr. A. Plantinga
Abstract
This paper tests the effects of attention grabbing events, such as one day abnormal stock returns and newspaper headlines and its frequency, on abnormal trading volume. By looking at 98 companies quoted on the Euronext Amsterdam index over the period 2010-2015, this paper tries to find whether the relation is valid in the Netherlands. I conclude that both newspaper headlines and abnormal stock returns are positively and significantly related to the aggregate level of abnormal trading volume. However, frequency has a small negative relation to trading. These results confirm the existence of the attention grabbing bias in small stock markets.
1. Introduction
‘When my shoeshine boy starts offering stock tips, it's time to get out of the market’
This is a wisdom often attributed to Joseph P. Kennedy, Sr., and describes unsophisticated investors’ tendency to enter the market at the top, and exit it at the low. This investment strategy, motivated by greed at first and fear secondly, makes them very unsuccessful investors. However, it leaves researchers with the question; what drivers are investors’ stock picks based upon? That is why this paper will try to unveil one of the motives for trading volume by questioning, what is the effect of attention grabbing events on the trading behavior of investors?
This paper employs data over the past five years (2010-2015) on all stocks presently and historically quoted on the Euronext Amsterdam stock exchange in the Netherlands. I thereby make a contribution to the current literature in several ways. First of all, this paper will link stock trading volume with several measures of attention grabbing events for five years starting 2010 and thereby serves as a post financial crisis update of the literature. This is in contrast to papers by for instance Barber and Odean (2008), Tetlock (2007) and Seasholes and Wu (2007), who all focus on the period prior to the financial crisis. Secondly, this paper will show in detail the effects of the relation between attention grabbing events and trading volume from the perspective of a geographically small country with a small capitalization stock market such as the Netherlands. Other authors, such as Li, Shi, Chen, and Kargbo (2014), Takeda and Wakao (2014), and Aouadi, Mohamed, and Teulon (2013), focused on much larger stock markets in China, Japan, and France respectively. The third contribution is how investors are biased in their choices for investment opportunities towards whatever has been at their attention.
The large amount of research on trading volume and attention grabbing events require a structured approach. Therefore, I will discuss separately the most important results derived from the literature on fundamental determinants of trading volume and those derived from the behavioral determinants of trading volume. Lastly, a section on the home country bias is included.
1.1. Fundamental determinants of trading volume
Behavioral finance is not the first to conduct research on the determinants of trading behavior. There have been many researchers investigating the fundamental drivers of trading volume. First of all, Ajinkya, Atiase, and Gift (1991) and Atiase, Ajinkya, Dontoh, and Gift (2011) express that fundamental motives for trading can drive volume. They research the effect of financial analyst earnings forecast on the trading volume of stocks. Ajinkya, Atiase, and Gift (1991) find that, first of all, the volume of trading is positively related to the dispersion in analyst forecasts on earnings. This means that as the absolute value between the forecasted earnings by the analyst increases, so does the aggregate level of trading. Furthermore, They find that revisions of these estimates also play a central role in a company’s trading volume. Here they note that “trading volume is positively related to the magnitude of the monthly revision in the mean of analysts’ EPS forecasts.”
Similarly, Atiase, Ajinkya, Dontoh, and Gift (2011) find that differences in prior beliefs of analysts on a firm’s earnings potential, along with the differential interpretation of information disclosure, and the effect of this disclosure on estimate divergence/convergence have an effect on trading volume. They also highlight that “trading volume models that exclude or fail to control for any of these determinants are misspecified with biased estimate coefficient.”
Related research by Athanassakos, Ackert, Naydenova, and Tafkov (2010), Abdallah, Abdallah, and Saad (2011) and Dodd, Louca, and Paudyal (2015) focusses on the determinants of trading volume in cross listed firms. These researchers attempt to identify the drivers of volume by looking at the differences between the same company being quoted on multiple exchanges allowing for a direct comparison. Athanassakos, Ackert, Naydenova, and Tafkov (2010) conclude that the determinants of trading volume are twofold. On the one hand, investor demand is driven by two visibility characteristics, Institutional interest, and analysts following. On the other hand, investor demand is driven by three differential risk characteristics, firm size, firm beta, and cross listed assets in foreign market. Although the last of these characteristics does not apply to this paper, the other pose relevant variables for investor demand.
accounting standards used by the firm, its stock visibility, and how long it has been listed. Although the overlap with the previous studies is great, it does include alternative measure for visibility and fundamental trading motives.
Lastly, Abdallah, Abdallah, and Saad (2011) also research the determinants of trading volume through cross listings. Their contribution to this paper is mainly through the determination of the research design they employ. Here, they mention that there are several factors that need to be controlled for in an event study on trading volume. These are a firm’s size, market volume, return, forecast error by analysts, whether the firm is in a developed market or not and whether it has raised capital. The Abdallah, Abdallah, and Saad (2011) paper will be discussed in more detail in Section 3.1 of this paper around the development of the econometric model.
Summarizing the results from the previous paper and others, a review of event studies on trading volume by Yadav (1992) finds five factors that drive trading volume: (i) change in individual expectations in relation to price changes, (ii) clientele adjustments related to risk, tax, etc., (iii) information asymmetry, (iv)liquidity considerations, and (v) market microstructure.
Now that the traditional determinants of trading volume have been summarized, it is possible to advance to more contemporary determinants of trading volume. As mentioned earlier, behavioral finance offers additional insights on the drivers of individual stock trading volume.
1.2. Behavioral determinants of trading volume
Ever since prospect theory was developed, much research has been conducted, most notably by Kahneman and Tversky (1973; 1979), on the irrationality of economic participants. Economists have discovered that people make irrational decisions based on limited time, limited attention, and limited information. This gives rise to man’s reliance on heuristics1 as a form of decision making. These
heuristics, or mental shortcuts, determine our daily decision making processes and therefore also our financial decisions. One important bias theorized by Kahneman and Tversky (1973) is the availability heuristic and the related recency and salience bias.
The availability heuristic entails the human shortcut to focus on what can be easily brought to mind. For instance, because it has recently been brought to one’s attention. This is why the availability heuristic is critical to the theme of this paper. Considering it might inspire investors to invest in those stocks that they could have encountered in newspaper articles or is being brought to one’s attention
1 Gilovich, T., Griffin, D., Kahneman, D., Heuristics and biases: the psychology of intuitive judgement. Cambridge
as a result of higher or lower than normal returns. This is the rationale behind the causal relation between attention grabbing events about a certain stock and that stock’s volume.
The above explained rationale, however, is not new in this paper. Many researchers have written about causal relations between media content and stock prices (see, e.g., Li, Wang, Li, Liu, Gong, and Chen, 2014; Pinnuck, 2014; Dellavigna and Pollet, 2009; Ferguson, Philip, Lam, and Guo, 2015; Hu, Dong, Liu, and Yao, 2013), media content and trading volume (see, e.g., Seasholes and Wu, 2007; Yu and Hsieh, 2010; Li, Shi, Chen, and Kargbo, 2014; Fang, Peress, and Zheng, 2014; Barber and Odean, 2008; Tetlock, 2007; Nowak and Anderson, 2014; Lillo, Miccichè, Tumminello, Piilo, and Mantegna, 2014; Cervellati, Ferretti, and Pattitoni, 2014; Yuan, 2015; Scheufele, Haas, and Brosius, 2011), and other related correlations such as stock splits and trading volume (see Huang, Liano, and Pan, 2013), the effect of attention on journalists’ stock recommendations (see Kerl and Walter, 2008) and Google search intensity and trading behavior(see Aouadi, Mohamed, and Teulon, 2013; Takeda and Wakao, 2014). It is especially this second strand of papers that is worthwhile to mention. However, the others also contain information relevant to this paper. First, noteworthy results from the research between media content and stock returns shall be discussed.
Although not directly related, it is still worth the effort to look into the research around the relation between media attention and stock returns. This is because the justification for this relation often comes from the same attention grabbing bias that is researched in this paper. This is especially visible in the paper by Pinnuck (2014). Pinnuck (2014) highlights that news that is characterized as stale (not new) does indeed invoke a price reaction when it is published again. This, Pinnuck (2014) mentions, is mostly attributable to individual investor trading behavior thereby relating it to pressures originating from the attention grabbing effect of these newspapers. A clear indication that individuals use news articles in the New York Times and Wall Street Journal as a filtering mechanism for investing opportunities. Additionally, Li, Wang, Li, Liu, Gong, and Chen (2014) look specifically at whether news can explain the abnormal returns of stocks in a modern setting characterized by social media. They conclude that stock markets are sensitive to public information in an era of social media and that it can influence individual investor trading activity.
importantly, they also show an effect on trading volume noting that at the announcement volume is abnormally high for friday as well as non-friday announcements and remains high for a period of time. Lastly, Ferguson, Philip, Lam, and Guo (2015) note that the tone as well as volume of news articles predict returns for stocks for the coming period and that this attention grabbing relation is mostly observable in larger firms’ stock. The relation between news articles and price reactions in these articles also applies to the research in this paper as the price reaction by stocks is a result of larger than usual trading volume.
Looking at the more directly applicable research on the relation between news articles and stock trading volume, I will start with the most influential one to this paper. Barber and Odean (2008) were one of the first who looked at the relation between attention grabbing events and buy sell imbalances (BSI) on a large scale. They were the first to detect a significant relation between attention grabbing events and individual investor buying. However, they found no such relation for institutional investors and actually conclude they behave opposite to individual investors. In addition to this early research is Seasholes and Wu (2007), who find that stocks that hit record high values (as a substitute for news) are characterized by high returns but more importantly by high volume and news coverage. Additionally, they note that these events lead investors to buy stocks they did not previously own. This directly supports the attention grabbing hypothesis. The prime interest in these two papers is that both were able to detect a higher trading volume from individual investors who did not previously own that stock, making the purchase of it most likely a result of their behavioral bias towards attention grabbing stocks.
Li, Shi, Chen, and Kargbo (2014) use the same approach as Barber and Odean (2008) in detecting a relation between attention grabbing events such as news, extreme returns, and high volume and the so called BSI. They find that especially small investors are prone to the attention bias and larger investors or institutions actually do not appear to exhibit the same behavior. They also note that the robust majority of investors in China is actually part of the small investors group, indicating that individual investors are a large mover of the Chinese stock market in particular, but likely also in general.
and companies are sensitive to endogenous factors such as returns and volatility, although they focus more on the former factors than the latter. Cervellati, Ferretti, and Pattitoni (2014) also find the exact same result as predicted by the attention grabbing hypothesis. They research whether buy and sell recommendations in a weekly column, which holds no additional information, influences the trading behavior of individuals. They observe an adverse amount of trading on the publication date which can only be explained by the salience bias.
Especially the results by Seasholes and Wu (2007) and Cervellati, Ferretti, and Pattitoni (2014) indicate the power of the salience bias and the validity of the attention grabbing bias hypothesized by Barber and Odean (2008). This is because Seasholes and Wu (2014) indicate that attention grabbing events lead people to buy stocks they did not previously own, and Cervellati, Ferretti, and Pattitoni (2014) find that news which is actually not new, leads to higher trading volume. Furthermore, by using market wide trading volume instead of detailed transaction records, Cervellati, Ferretti, and Pattitoni (2014), among others, indicate the quality that can be achieved by using broader trading indicators rather than a detailed BSI used by Barber and Odean (2008).
Around the same time as Barber and Odean (2008), Tetlock (2007) discovered that especially the content was important to the trading volume reaction. Tetlock (2007) finds that unusually high or low pessimism predicts high market trading volume and that “news media can predict movements in broad indicators of stock market activity.” Yuan (2015) also relates attention grabbing events to adverse amounts of trading. However, his research is based on a market wide selling reaction to Dow Jones record levels and front page articles about the stock market. These two papers offer additional insights into the trading behavior of individual investors around attention grabbing events in that they indicate the importance of content, which will be accounted for to some degree in this paper as well.
Although the effect of news on the trading behavior of individuals as well as institutions seems obvious by now, it is not universally accepted. For instance, by studying airline stocks only, Nowak and Anderson (2014) find that although trading spikes around macro announcements, the relation does not seem to hold when it comes to firm specific announcements. Thus, they question whether the relation is truly applicable for this specific industry. In Germany rather than in the US, Taiwan, and China, Scheufele, Haas, and Brosius (2011) find that individual investors do not have a significant enough effect on trading volume and prices. However, this effect is measured over a two month period instead of only one trading day as with the other papers analyzed in this Section.
Focusing on a smaller and less developed market, Yu and Hsieh (2010) find that in the Taiwan stock exchange individual investors as well as institutional investors use attention grabbing events (extreme one day returns) to base their investments on. This research, from a post-crisis era, is in conflict with what was previously known on the relation between attention grabbing events and trading volume and warrants the question on the validity of the relation in post-crisis stock markets.
Considering the research that is not directly related, but holds interesting thoughts for discussion, attention is directed to Huang, Liano, and Pan (2013). By studying the relation of stock splits and stock split announcements with liquidity they research a similar causal effect. Here stock splits announcements represent the news(worthy) factor and liquidity is a direct measure of trading volume as they mention in their paper. They find that the announcements of stock splits have significant positive effects on a stock’s liquidity and that the liquidity improvement is short lived and only visible around the announcement of the stock split. Therefore, Huang, Liano, and Pan mention that “our results appear to be more consistent with the signaling hypothesis and/or the attention grabbing hypothesis, than with the improved liquidity hypothesis.”
Although not a measure of news, but possibly a more direct measure of attention, is the relation between Google search intensity and trading. This measures the (relative) popularity of a company as a search term on Google and is therefore more direct in that it confirms that the company searched for is at the attention of the searcher. Aouadi, Mohamed, and Teulon (2013) and Takeda and Wakao (2014) note that a publicly quoted company that is searched for on Google is thus just as much at ones attention as a newspaper article and therefore wish to look at its relation with trading behavior. Both Takeda and Wakao (2014) and Aouadi, Mohamed, and Teulon (2013) find that search activity on Google is strongly related to trading volumes in Japan and France respectively. The importance of this research is especially in the fact that Google search intensity is a more direct measure of investor attention. However, issues with this measure are twofold. First of all, Google measures its search intensity on a monthly basis, which does not allow direct comparison of attention (grabbing) and trading volume. Furthermore, searches are all aggregated, which means that a teenager searching for Heineken beer will also be included in the statistics while I wish to limit the attention grabbing variable to investors.
recommendations on whatever grabs his or her attention and “thereby, journalists intensify the well documented attention bias of individual investors.”
1.3. Home country bias
Important to the development of a strongly supported hypothesis on small stock market countries, as this paper is attempting to do, it is also vital to discuss the mechanisms that drive the differences between large and small stock markets. The reason why there should be a difference in the effect of attention grabbing events on trading volume is found in another bias inherent in humans. This bias is known as the home country bias and expresses that people in general are biased towards investing in companies that are from their own country instead of investing in foreign firms.
The home country bias was tested by French and Poterba (1991) and revealed some striking evidence on people’s bias towards the familiar. They found that U.S. investors were almost 93.8% invested in U.S. stocks, Japanese investors for 98.1% in Japanese stocks and United Kingdom (U.K.) investors, although more diversified, still 82% in U.K. stocks. This exposes the overwhelming tendency of investors to invest in domestic securities. However, that is not all, if these investors were to diversify internationally, according to Solnik (1974), they would be able to achieve the same return with lower risk. Furthermore, French and Poterba explain that rational explanations for this bias such as capital movements restrictions, differential trading costs, and differential tax rates do not pose a severe enough hurdle anymore to explain the allocations of capital above. Coval and Moskowitz (1999) reason that home bias is the result of people having, or can feel that they have an informational advantage on firms that are in geographical vicinity instead of being domestic.
1.4. Research objective and hypotheses
Although the impact of news in general has already been tested widely, there hasn’t been any notable research on the fact whether this relation is just as relevant within a country with a much smaller stock market. Yu and Hsieh (2010) find that the Taiwan stock market is also subjected to the attention grabbing bias. However, the focus in their paper is only on extreme returns in the period prior and during the financial crisis. Therefore, this paper adds to previous research by investigating the relation in a post-crisis era and also in a much smaller stock market setting. The smaller stock market implies a much smaller investment universe (assuming all investors are home biased). This smaller investment universe should make it easier for investors to filter through the possibilities available in the stock market. Therefore the expectation is that the impact of attention grabbing events is less pronounced in a smaller country but still present.
Therefore, central to this paper will be the question: Are investors in the Netherlands influenced by attention grabbing events such as newspaper articles and previous day stock returns? Specifically, I wish to research the following hypotheses when it comes to attention grabbing events: H1: Individual stocks quoted on the Euronext Amsterdam stock exchange show higher volume of
trading as a result of news article headlines.
H2: Individual stocks quoted on the Euronext Amsterdam stock exchange show higher volume of
trading as a result of increased frequency of news headlines.
H3: Individual stocks quoted on the Euronext Amsterdam stock exchange show higher volume of
trading as a result of previous day stock returns on the same individual stock.
The remainder of the paper will be organized as follows. Section 2 details the empirical model and the rationale for all individual components. Section 3 comments on the origin and quality of the data. Section 4 reports on the results derived from the data and Section 5 will conclude the paper.
2. Research methods
Abnormal trading indexit = αi+β1Newsit +β2Frequencyit +β3Lagged returnit
+β4Sizeit +β5Market volume indext +β6Returnit
+β7Strategic holdingsit +β8Betait +β9Capitalit +β10Ageit (1)
+β11Earnings reportit +∑ βi(Dayi)
n
i=1 +∑ βi(Monthi) n
i=1
Where abnormal trading index is defined as the relative level of trading volume in company i at date t weighed by the relative level of trading in the market. News is a dummy variable which is equal to one at date t and t+1, if company i is mentioned in a headline at date t and zero otherwise. Frequency is the sum of the ‘news’ variable for the previous 21 trading days for company i at date t. Lagged return is the absolute previous trading day return for company i at date t. Size is the market value of all free float shares of company i at date t, measured in millions of euros. Market volume index is the relative level of trading volume averaged for all companies listed at date t. Return is the daily return for company i at date t. Strategic holdings is the relative amount of company i’s free float shares held by strategic investors at date t. Beta is the previous five year, monthly, measure for company i’s stock beta at date t. Capital is a dummy variable equal to one if company i issues shares at date t and zero otherwise. Age is the number of years company i has been quoted at the Euronext Amsterdam exchange at date t. Earnings report is a dummy variable equal to one if company i issues an earnings report on date t and zero otherwise. Day is the sum of five dummy variables for Monday, Tuesday, Wednesday, Thursday, and Friday, which are equal to one if the trading day t is on a Monday, Tuesday, Wednesday, Thursday and Friday respectively. Month is the sum of 12 dummy variables for January, February, March, April, May, June, July, August, September, October, November, and December, which are equal to one if the trading day t is in January, February, March, April, May, June, July, August, September, October, November, and December respectively.
Abnormal trading index is defined as the relative level of trading volume in company i at date t weighed by the relative level of trading in the market. In mathematical terms, it is calculated using the following equation.
Abnormal trading indexit = Vit
Vmt * Vm
Vi (2)
where Vit equals company i’s volume at date t in units, Vmt is the aggregate volume in the market at
time t and Vm and Vi are measures of average volume in the market and company i respectively over
the previous 40 trading days. This formula weighs the relative level of volume of company i at date t (Vit/Vi) with the relative volume of trades in the entire market at date t (Vmt/Vm). This way, when the
days, following the example of Harris and Gurrel (1986). The average level of market volume is on an equally weighted basis to prevent large corporations from having an adverse effect on this variable.
In contrast to other papers, I do not employ the BSI, a measure introduced by Barber and Odean (2008). Instead, in Eq. (2), a measure of relative volume as suggested by Yadav (1992) is used. The intuition for the exclusion of the BSI comes from Barber and Odean’s (2008) paper. They motivate that the BSI is used solely as a way to test the effects of attention grabbing events on different categories of investors, and does not have more informational value than a market wide stock trading volume. This is because the buy and sell volume of all the investors in the world is by definition equal. High stock market volume therefore means that a stock has high demand just as much as with using the BSI.
Now, the composition and rationale for all of the endogenous variables employed in the model will be discussed. The first three variables, news, frequency, and lagged returns, are the variables being tested by the model, all other variables are controls.
2.1. News
The news variable is a dummy variable, which is equal to one for date t and t+1 if company i is mentioned in a headline on date t and zero otherwise. News tests the hypothesis that newspaper headlines attract investor attention and consequently lead to abnormally high trading volume. It is therefore expected that β1 in Eq. (1) is positive. I decided to use the occurrence of company i’s name
in the headline because of its direct impact on investors. Generally, when investors read the newspaper, they tend to focus on the headlines in the paper first. If this headline catches their attention they will continue reading the article. Thus it is very unlikely that the mention of a company’s name only once throughout a large newspaper article will catch anyone’s attention. In this fashion, the likelihood that news is an attention grabbing event is enlarged, and thus reliable results can be inferred from the data on this variable. Accounting for lagged reactions by investors, the news variable dummy will also be equal to one for the day following the headline. Any news article headlines appearing on a non-trading day were attributed to the first following trading day.
2.2. Frequency
Frequency is equal to the sum of the news variable for the previous 21 trading days for company i at date t. Frequency tests the hypothesis that the frequency with which a company is in the news on a monthly basis increases the abnormal trading index of that company. It is therefore expected that β2 in Eq. (1) is positive. Besides simply looking at the occurrence of news for a company,
the news variable is summed for the previous 21 trading days. Considering this paper focusses on attention grabbing effects, it tries to test the effect of short term events on abnormal trading volume. Therefore, a short term frequency variable is preferred over a long term frequency variable. Monthly frequency is then preferred over shorter intervals because shorter intervals were too highly correlated with other variables, including the news variable, which will be discussed in Section 3.3. The decision was made to sum the values over a period of 21 trading days because this is about the average amount of trading days in one month over the period 2010-2015.
2.3. Lagged return
Lagged return is equal to the absolute previous trading day return for company i at date t. Lagged return tests the existence of a relation between a stock’s previous trading day percentage return, whether this is positive or negative, and its subsequent abnormal trading index. It is therefore expected that β3 in Eq. (1) is positive. I use the return of stock i on date t-1 following Barber and Odean
(2008), Tetlock (2007), Yu and Hsieh (2010), Cervellati, Ferretti, and Pattitoni (2014), Takeda and Wakao (2014) and Li, Shi, Chen, and Kargbo (2014). The motivation to use lagged returns is threefold. Firstly, previous research by Abdallah, Abdallah, and Saad (2011) indicates that a company’s return on date t is actually more appropriate as a control variable. This will be discussed in Section 3.6. further. Secondly, individual investors generally do not encounter the high returns until markets have closed and they are not able to buy the stock until next trading day. Thirdly, the use of lagged returns eliminates a potential endogeneity problem. The motivation to use the absolute value of the lagged returns is derived from, among others, Barber and Odean (2008). They find that both positive outliers (top decile of returns) as well as negative outliers (bottom decile of returns) lead to higher trading volume.
2.4. Size
2.5. Market volume index
Market volume index is equal to the relative level of trading volume averaged for all companies listed at date t. It is constructed as follows; all values for abnormal trading index at date t, for all constituents that are traded, are averaged at day t to indicate that day’s market volume index. This control variables is the second control variable derived from the Abdallah, Abdallah and Saad (2011) model. They motivate that it is positively related to a firm’s trading volume as suggested by previous research by Yadav (1992).
2.6. Return
Return is equal to the daily return for company i at date t. Returns are measured based on continuously compounding using the natural logarithm. Return is also a control variables used by Abdallah, Abdallah and Saad (2011). They motivate the use of this variable by referring to Datar, Naik, and Radcliffe (1998), who show that stock returns are negatively related to share turnover. Also, they mention that Yadav (1992) refers to stock returns as the arrival, or change, of information about a firm. In this respect, the return variable can be seen as including the effects of information which is not accounted for by any other variable in the model. Also, daily returns might be a motivator for unsophisticated investors as it is a signaling device of a good company and might induce trading volume.
2.7. Strategic holdings
Strategic holdings is equal to the relative amount of company i’s free float shares held by strategic investors at date t. It is calculated by totaling the percentage of shares held by strategic block holders which have reported to own more than 5% of total free float shares in company i. Athanassakos, Ackert, Naydenova, and Tafkov (2010) use institutional holding levels as a measure of the visibility of a firm. They note that Merton (1987) finds that firms whose stock is not held by institutional investors in large amounts are likely neglected by investors. This paper expands the use of institutional investors to strategic investors. This type of investor often holds large percentages of a firm’s stock which limits their ability to be traded in the market and thereby withdrawing possible trading volume from the market.
2.8. Beta
mention that beta is used by (international) investors as a proxy for correlation with other assets and is thus a measure for diversification by investors and a determinant of trading volume. This relation is also supported by Halling, Pagano, Randl, and Zechner (2008).
2.9. Capital and earnings report
Capital is a dummy variable equal to one if company i issues shares at date t and zero otherwise, and earnings report is a dummy variable equal to one if company i issues an earnings report on date t and zero otherwise. There are several events in finance that inherently influence the liquidity of stocks as they contain new information that influences the stock directly. Firstly, when a firm raises capital on the stock market by issuing a Seasoned Equity Offering (SEO) or Initial Public Offering (IPO). This variable (Capital) is the last control variable adopted from Abdallah, Abdallah and Saad (2011). They find a positive relation between the issuances of capital and trading volume. Secondly, when a company issues a quarterly earnings report. This relation is detected by, among others, Morse (1981), and Ahmed, Schneible JR., and Stevens (2003). Morse (1981) finds that on the day prior to the announcement and the day itself, excess trading volume is observed. Ahmed, Schneible JR., and Stevens (2003) observe higher trading volume reactions unrelated to price increases. Following this research, I include a variable for capital issuances and earnings report announcement days.
2.10. Age
Age is equal to the number of years company i has been quoted at the Euronext Amsterdam exchange at date t. It is measured from the day of inception of the Euronext Amsterdam exchange or since the first day of quotation. In addition to a firm’s size, the seniority of a company’s stock is an excellent measure of visibility. This follows the logic of Dodd, Louca, and Paudyal (2015) who incorporate this variable in their model as well.
2.11. Seasonal controls
influenced by seasonality. Trading levels are not constant over the year and through the week. For instance, Pettengil and Jordan (1988) find that relative trading volume has similar seasonal patterns as returns. Therefore, seasonal controls are introduced for the all trading days of the week and months of the year.
3. Data
Most data originate from Thomson Reuters’ Datastream, a database with all relevant company data on stocks worldwide. Using Datastream ensures the exclusion of the survivorship bias. This means that companies that are no longer quoted on the Euronext Amsterdam stock exchange are still included in the research. This database returns a total of 382 shares for the Euronext Amsterdam exchange. After filtering out the stocks quoted outside the period 2010-2015, a total of 125 quoted companies remain. Using the entire Euronext Amsterdam to construct the data meant the inclusion of many firms for which trading activity was significantly low. Therefore, I introduce a cut-off value of 95% for trading volume data. Any company listed in the Euronext Amsterdam in the period 2010-2015 for which trading volume was zero more than 5% of the time was excluded from the analysis. This decision was made to exclude noise originating from small companies that have unusually high abnormal trading indexes on each day of positive trading volume. After this revision, 98 companies remain to be researched in this paper. Capital was constructed using information from behr.nl, a local stock watchers website that holds information on SEOs. Information on news and frequency was derived from LexisNexis Academic.
LexisNexis database is used because it holds all newspaper articles published by De Telegraaf in the required period. De Telegraaf was selected for two reasons. In accordance with Tetlock (2007) who uses the Wall Street Journal for its high circulation, De Telegraaf is the Dutch (financial) newspaper with the highest circulation. Additionally, as this paper attempts to measure the trading reaction of primarily unsophisticated investors, I decided that De Telegraaf would be a better measure than Het Financieele Dagblad, which is mostly read by well-educated and sophisticated individuals. The rest of Section 4 will highlight several descriptive statistics, followed by detailed summaries of several variables and their implications on the research. Concluding, the correlations of all variables will be discussed.
3.1 Descriptive statistics
50% of the observations have a positive value (one). From the mean of these dummy variables, the percentage of total observations that is not zero can be derived. For instance, for the news variable a mean value of 0.074 amounts to 7.4% positive values for news. Capital has a mean value of zero, which implies that there are almost no positive values for this dummy. The data on the dummy variables do not require any specific action.
The remaining variables are of more interest as they might represent abnormal distributions. For instance, the abnormal trading index has a minimum of zero, which confirms that there are companies in the sample for which there are days with no trading volume and a maximum of about 58.930. More interesting are the mean and the median which are one and 0.796 respectively. A value of one for abnormal trading index is to be expected. As one might recall, the abnormal trading index was constructed by dividing the relative turnover of company i by the relative turnover of the market. Thus, taking the mean of all those values implies dividing the mean of the market by the mean of the market, which has to be equal to one. The median reveals that there are more days with lower abnormal trading index than higher than the market. The difference between the mean and median indicate a potential distribution issue and will be investigated further in the next section.
For frequency the median reveals that more than half of the values are zero and that on average there is around 0.831 headline for all companies in the Euronext Amsterdam in a 21 day trading period. The lagged return, which is an absolute value, has an expected minimum of zero (no return) and a maximum of 609.807%. This maximum is the same value found under return which is constructed with more or less the same values. The average value of lagged return, which equals 1.538%, is probably influenced by outliers. This can be concluded from the lower median of 0.922%. The daily returns are also characterized by unusual data. The negative return of -412.296% seems counter-intuitive, however, using logarithmic scaling, a return larger than -100% is possible. The median value reveals that the midpoint of all returns is no return, and the mean highlights that on average returns are slightly negative over the period 2010-2015. All these variables warrant more attention concerning their distributions.
Royal Imtech. The mean beta is a little lower than one, which is close to its median of 0.898. Lastly, age reveals that the youngest company quoted on the Euronext Amsterdam was 0.003 years old, or exactly one day. The highest value for age is 42.724 years, which is shared by several companies that have been quoted on the Amsterdam stock exchange since its inception until the end of the tested period such as Shell, Unilever, Heineken et cetera.
3.2. Transformed variables
Table 1
Description of variables.
The table provides summary statistics of the dependent and independent variables. The table includes 98 stocks over 1,281 trading days between September 23, 2010 and September 23, 2015. Strategic holdings has 103,185 observations due to missing values. Other variables have the maximum number of 103,492
observations. Abnormal trading index is the relative level of trading volume in company i at date t weighed by the relative level of trading in the market. News is a dummy variable which is equal to one at date t and t+1, if company i is mentioned in a headline at date t and zero otherwise. Frequency is the sum of the ‘News’ variable for the previous 21 trading days for company i at date t. Lagged return is the absolute previous trading day return for company i at date t. Size is the market value of all free float shares of company i at date t, measured in millions of euros. Market volume index is the relative level of trading volume averaged for all companies listed at date t. Return is the daily return for company i at date t. Strategic holdings is the relative amount of company i’s free float shares held by strategic investors at date t. Beta is the previous five year, monthly, measure for company i’s stock beta at date t. Capital is a dummy variable equal to one if company i issues shares at date t and zero otherwise. Age is the number of years company i has been quoted at the Euronext Amsterdam exchange at date t. Earnings report is a dummy variable equal to one if company i issues an earnings report on date t and zero otherwise. Monday, Tuesday, Wednesday, Thursday, and Friday are dummy variables, which are equal to one if the trading day t is on a Monday, Tuesday, Wednesday, Thursday and Friday respectively. January, February, March, April, May, June, July, August, September, October, November, and December are dummy variables, which are equal to one if the trading day t is in January, February, March, April, May, June, July, August, September, October, November, and December respectively. SD is short for standard deviation.
Variable Mean Median Minimum Maximum SD
Abnormal trading index 1.000 0.796 0.000 58.930 1.247
News 0.074 0.000 0.000 1.000 0.074
Frequency 0.831 0.000 0.000 17.000 1.528
Lagged return 1.538% 0.922% 0.000% 609.807% 3.464
Size 5,585.480 934.645 0.062 122,303.700 13,443.810
Market volume index 1.072 1.015 0.186 2.851 0.296
3.3. Correlation
Table 2 shows the correlation matrix including key variables. It also includes an additional weekly frequency measure, calculated as the last five trading days sum for news, following the construction of monthly frequency. Table 2 shows that the correlation of the variables are, with the exception of the frequency variables, not high. As can be observed, the range of the correlation matrix is [-0.227; 0.369], which does not pose an issue. However, both weekly and monthly
frequency are highly correlated with news and each other. Therefore, it was decided to use the least correlated variable in the estimation of the model, monthly frequency. For the full correlation matrix, also including all time control variables, see Table A.1 in the Appendix.
Table 2
Correlation matrix
The table below indicates the correlations between the most relevant variables included in the model plus an additional parameter for weekly news frequency, calculated in a similar fashion as the monthly frequency.Abnormal trading index is the relative level of trading volume in company i at date t weighed by the relative level of trading in the market. News is a dummy variable which is equal to one at date t and t+1, if company i is mentioned in a headline at date t and zero otherwise. Frequency is the sum of the ‘News’ variable for the previous 21 trading days for company i at date t. Lagged return is the absolute previous trading day return for company i at date t. Size is the market value of all free float shares of company i at date t, measured in millions of euros. Market volume index is the relative level of trading volume averaged for all companies listed at date t. Return is the daily return for company i at date t. Strategic holdings is the relative amount of company i’s free float shares held by strategic investors at date t. Beta is the previous five year, monthly, measure for company i’s stock beta at date t. Capital is a dummy variable equal to one if company i issues shares at date t and zero otherwise. Age is the number of years company i has been quoted at the Euronext Amsterdam exchange at date t. Earnings report is a dummy variable equal to one if company i issues an earnings report on date t and zero otherwise. Monday, Tuesday, Wednesday, Thursday, and Friday are dummy variables, which are equal to one if the trading day t is on a Monday, Tuesday, Wednesday, Thursday and Friday respectively. January, February, March, April, May, June, July, August, September, October, November, and December are dummy variables, which are equal to one if the trading day t is in January, February, March, April, May, June, July, August, September, October, November, and December respectively.
Abnormal trading index News Weekly frequency Monthly frequency Lagged return Size Market volume index Return Strategic holdings Beta Capital Age
News 0.093
Week frequency 0.085 0.684
Month frequency 0.042 0.463 0.682
Lagged return 0.152 0.100 0.078 0.041
Size 0.087 0.188 0.256 0.369 -0.147
Market volume index -0.049 0.013 0.014 -0.000 0.047 -0.002
4. Results
This section discusses the empirical results derived from the data and the model trying to measure whether attention grabbing events and frequency have a positive effect on the aggregate market level trading volume in the Netherlands. General regressions as well as more sophisticated techniques are employed to accurately measure their effect. I will also discuss several diagnostics and their implications and implement robustness checks on the results. On a general note, considering the model is of a log-log scale, some explanation is warranted. All variables that have been transformed in Section 4.2. should be read as follows. For any 1% increase in said variable the value of abnormal trading index rises by (1.01coefficient-1)*100%. For all other variables, this means that an increase of
exactly one unit leads to an increase in abnormal trading index of the magnitude (ecoefficient-1)*100%.
These effects will be discussed in more detail in the following sections. 4.1. Panel event study
Since the data are distributed over a total of 98 firms and 1281 trading days, it is more logical to employ a panel event study than a normal event study with pooled data. A Hausman specification test as well as a Breusch-Pagan Lagrange-multiplier test for random effects reveal that the proper model for the data is fixed effects. Table 3 indicates the estimation results on cross-section as well as a time-fixed effect model. In the cross-section fixed effect model, control variables friday and december are omitted. In the time-fixed effect model, all time control variable and market volume index are omitted. Both times this is to prevent potential problems from multi-collinearity. A Chow test reveals that there is no evidence for a difference in coefficients between the cross-section fixed effects and time-fixed effects. Therefore, in the remainder of this paper, cross-section fixed effects will be employed as this regression is less influenced by the effects of multi-collinearity and includes several time sensitive variables.
Table 3 below shows that the two fixed effects models are not very different from each other. There is only one variable that changes by sign, which is Beta. All other coefficients are necessarily the same and do not change by more than one basis point. This, together with the fact that model two omits market volume index and model one includes several time dummies already, leads to the conclusion that either one is as good as the other. Therefore, models three and four were also estimated using cross-section fixed effects. The explanatory power of both models is also more or less equal. Time-fixed effect explains roughly 0.5% more of the variance in the data, which is low. More important than the adjusted R2 is the fact that almost all variables are significant at a level lower than
Table 3
Abnormal trading index following attention grabbing events.
Regression (1) is a cross-section fixed effects model and regression (2) is a time-fixed effects model. Both model (1) and model (2) include all three attention grabbing events. Regression (3) excludes lagged returns and regression (4) excludes news (both cross-section fixed effects). Regression (5) excludes all control variables. Dependent variable, abnormal trading index, is the relative level of trading volume in company i at date t weighed by the relative level of trading in the market. News is a dummy variable which is equal to one at date t and t+1, if company i is mentioned in a headline at date t and zero otherwise. Frequency is the sum of the ‘News’ variable for the previous 21 trading days for company i at date t. Lagged return is the absolute previous trading day return for company i at date t. Size is the market value of all free float shares of company i at date t, measured in millions of euros. Market volume index is the relative level of trading volume averaged for all companies listed at date t. Return is the daily return for company i at date t. Strategic holdings is the relative amount of company i’s free float shares held by strategic investors at date t. Beta is the previous five year, monthly, measure for company
i’s stock beta at date t. Capital is a dummy variable equal to one if company i issues shares at date t and zero
otherwise. Age is the number of years company i has been quoted at the Euronext Amsterdam exchange at date
t. Earnings report is a dummy variable equal to one if company i issues an earnings report on date t and zero
otherwise. Monday, Tuesday, Wednesday, Thursday, and Friday are dummy variables, which are equal to one if the trading day t is on a Monday, Tuesday, Wednesday, Thursday and Friday respectively. January, February, March, April, May, June, July, August, September, October, November, and December are dummy variables, which are equal to one if the trading day t is in January, February, March, April, May, June, July, August, September, October, November, and December respectively. Standard error are included in parentheses. Levels of significance are indicated by *(10%), **(5%), and ***(1%).
Regression (1) (2) (3) (4) (5) News 0.096*** 0.095*** 0.118*** 0.097*** (0.004) (0.004) (0.004) (0.004) Month frequency -0.004*** -0.010*** -0.004*** 0.003*** -0.005*** (0.001) (0.001) (0.001) (0.001) (0.001) Lagged return 0.102*** 0.109*** 0.108*** 0.100*** (0.002) (0.002) (0.002) (0.002) Size 0.010*** 0.015*** -0.000 0.011*** (0.002) (0.001) (0.002) (0.002)
Market volume index -0.087*** -0.077*** -0.086***
Focusing on the most striking results, one can see that both news and lagged return, in accordance with the first and last hypothesis, have positive and significant coefficients. In both models, the news dummy indicates that news leads to a minimum of (e0.095-1)*100% =9.97% increase in
abnormal trading index. On the other hand, for every 100% increase (doubling the variable) in the lagged return of company i, its abnormal trading index rises by a minimum of 7.18%. Lastly, in contrast to the second hypothesis, the monthly frequency of news is negatively related with abnormal trading index. Every additional day that company i is in the news in the last month leads to a drop in abnormal trading index of around -0.4% in model one and -1.00% in model two.
There are several reasons for the unexpected negative relation between abnormal trading index and monthly frequency. Over the measured period of 21 trading days it is likely that all trading that is motivated by the news has already taken place and therefore, no higher trading volume is to be expected. Also, investors might simply be tired of news around a certain high interest stock and not pay as much attention to it as they might have before. Lastly, news that occurs in high frequency is often of the same nature. For instance, Royal Imtech received considerable news attention around its bankruptcy several months ago. Although this does result in a high level for news frequency, it is mostly concerned with a single topic.
Hiding lagged returns from the regression, results in a high increase of the effect of news on trading. Now, in the event of news, abnormal trading index increases by 12.52%. Other coefficients do not change significantly other than the constant term, which increases from 0.419 to 0.581. Repeating the regression with news hidden, yields the effect of monthly frequency positive. This is probably caused by frequency describing the effect of news on its own. The coefficient for lagged return is almost identical to the previous regressions and therefore a 100% increase (relatively) in lagged returns results in an increase in abnormal trading index of about 7.77%. In regression four, the constant term reverts to the original value of around 0.419.
abnormal trading index on that day. An earnings report is much more influential on the abnormal trading index, since it leads to an increase of about 58.09% to 59.68% in the event of such a report. Lastly, strategic holdings has no impact whatsoever and is seldom significant, together with all the time control variables, which can be observed in Table A.2. in the Appendix.
Lastly, the fifth regression includes only the variables news, monthly frequency and lagged returns. Regression five shows that, not controlling for other variables, does not lead to significantly different result. News still leads to an increase in abnormal trading index of 10.19%. Every additional news event on a monthly basis leads to an decrease of around -0.50% and lagged return positively influence abnormal trading index by 7.18% for every doubling of the absolute return of a stock. Logically, the intercept of the model increases to 0.542, as it captures some of the effects previously measured by the control variables.
4.2. Diagnostic tests
Following the results of the empirical model, it is also important to test the quality of the model itself. Diagnostic test help conclude the relevance of the results and their implications. Optimally, a model adheres to five assumptions for BLUE (Best Linear Unbiased Estimator)2.
1. The average value of the errors is zero 2. Assumption of homoscedasticity 3. Assumption of no autocorrelation
4. The regressors are not correlated with the error term 5. The errors are normally distributed
Table 4 below shows a detailed description of the error terms of the cross-sectional fixed effects model as described under Table 3 regression (1). Table 4. shows values for the skewness and kurtosis which are not consistent with a normal distribution. Carrying out a Jarque-Bera test confirms that the errors are far from normally distributed. However, for sufficiently large sample sizes, violation of the normality assumption of the errors is virtually inconsequential according to Brooks (2014)2. Thus,
with a sample size of 103,185, there is reason to doubt the impact of this violation. The second assumption is also troublesome, and likely violated as a result of the high number of observations. Because simple heteroscedasticity test are not easily implemented on fixed effect models, a simple plot with residual versus fitted values should suffice. This plot, which can be seen in Fig. 2 in the
Appendix, shows that the errors are not constant over the course of the plot and therefore, variance of the error terms is not constant throughout the regression. This violation is the only violation that might impact the quality of the results. To limit the effects of this violation, several robustness checks are employed in the following section.
There is no evidence to support autocorrelation in the data after conducting a Wooldridge test3. Thus this assumption remains intact in the model with reasonable certainty. In the Appendix,
under Table A.1, the correlations can be found of all variables including the error term. That correlation matrix show that there is no correlation whatsoever between any of the variables and the error term, and does not violate assumption 4. In short, diagnostic test show that, although the model is not the Best Linear Unbiased Estimator, there is no reason to doubt the quality of the results in light of a robust estimation.
Table 4
Detailed description of cross-section fixed effect errors.
Observations 103,185 Mean 2.01E-12 Standard deviation 0.321 Variance 0.103 Skewness 2.107 Kurtosis 13.048 4.3. Robustness checks
Several techniques were employed to confirm the robustness of the initial results. First, two separate regressions were run on two halves of the data. Secondly, an identical regression was run as before under the conditions of robust standard errors. The results of these regressions are given in Table 5. In general, the table reveals that all three regression show that the magnitude and significance of the hypothesized variables are indeed valid. News’ coefficient has a range of [0.093; 0.098], which is around the same value found in the model excluding robustness checks. For lagged return, the range is [0.099; 0.106] and thereby also almost identical to the non-robust estimations. Although monthly frequency does not have a large coefficient, robust regressions indicate that the effect of frequency on abnormal trading index is still around -0.5% for every additional news event. In short, there is no reason to doubt the magnitude and significance of these three variables in this paper.
Table 5
Robust estimation of abnormal trading index following attention grabbing events.
Regression (1) tests the first 50% of the data, regression (2) includes the second half. Regression (3) expresses the entire data set with robust standard errors. Dependent variable, abnormal trading index is the relative level of trading volume in company i at date t weighed by the relative level of trading in the market. News is a dummy variable which is equal to one at date t and t+1, if company i is mentioned in a headline at date t and zero otherwise. Frequency is the sum of the ‘News’ variable for the previous 21 trading days for company i at date t. Lagged return is the absolute previous trading day return for company i at date t. Size is the market value of all free float shares of company i at date t, measured in millions of euros. Market volume index is the relative level of trading volume averaged for all companies listed at date t. Return is the daily return for company i at date t. Strategic holdings is the relative amount of company i’s free float shares held by strategic investors at date t. Beta is the previous five year, monthly, measure for company i’s stock beta at date t. Capital is a dummy variable equal to one if company i issues shares at date t and zero otherwise. Age is the number of years company i has been quoted at the Euronext Amsterdam exchange at date t. Earnings report is a dummy variable equal to one if company i issues an earnings report on date t and zero otherwise. Monday, Tuesday, Wednesday, Thursday, and Friday are dummy variables, which are equal to one if the trading day t is on a Monday, Tuesday, Wednesday, Thursday and Friday respectively. January, February, March, April, May, June, July, August, September, October, November, and December are dummy variables, which are equal to one if the trading day t is in January, February, March, April, May, June, July, August, September, October, November, and December respectively. Standard error are included in parentheses. Levels of significance are indicated by *(10%), **(5%), and *** (1%).
Regression (1) (2) (3) News 0.093*** 0.098*** 0.096*** (0.006) (0.006) (0.011) Month frequency -0.004** -0.004*** -0.004** (0.002) (0.001) (0.002) Lagged return 0.106*** 0.099*** 0.102*** (0.003) (0.003) (0.006) Size 0.013*** 0.008*** 0.010 (0.004) (0.003) (0.006)
Market volume index -0.086*** -0.089*** -0.087***
5. Conclusion
This paper started with the expectation that investors are far from optimizing beings. This expectation would expose itself, among others, in the fact that investors base their investment decisions on events that grab their attention such as news and high positive or negative returns. This logic is derived from theory, mostly supported by Barber and Odean (2008), and comes from these investors’ inability to assess the quality of all stocks in the stock market. This paper attempted to research the presence of such an attention bias in stock market behavior in the Netherlands over the period 2010-2015.
The research conducted in this paper reveals several important relations between attention grabbing events and abnormal trading volume in the Netherlands. As was already detected in larger stock markets and countries, like the U.S., Japan, and France, a small country with a small stock market also exhibits a positive relation between attention grabbing events and abnormal trading volume. This conclusion corresponds with the hypothesized relations in Section 2.4. of this paper. The results show that for every day a company is mentioned in a headline, that company’s abnormal trading volume increases by approximately 10%. For lagged returns, every doubling of the absolute lagged returns of a company leads to an increase in abnormal trading volume of around 7.5%. However, unlike the attention grabbing events, the frequency of news events on a monthly basis does not seem to entice investors to buy that stock. For every additional occurrence in a headline, abnormal trading volume for a company drops by approximately 0.4%. In an attempt to explain this, I point towards the lagged nature of this variable and the saturation of investor demand.
The results of this paper were of sufficient quality to survive several robustness checks. The results of the hypothesized variables do not show changes when accounting for robustness. The effect of news remains around 10%, lagged returns around 7.5% and frequency around -0.4%. Diagnostics also reveal no results that could invalidate the conclusions derived from the data. Therefore, this research is a valuable addition to the existing papers on attention grabbing events and trading behavior. In contrast to previous research, I show that attention grabbing events lead to higher than normally expected trading volume in a small country, in this case the Netherlands. Also compared to the previous papers examined, this paper highlights the validity of the attention grabbing hypothesis in the 2010-2015 post crisis period.
coverage for all companies included in the analysis meant that it was not possible to research other effects of news, such as the type of news, the tone or the importance of the news. Thus, the results of this paper are limited to an aggregated relation between attention grabbing events and market trading volume.
Unfortunately, it was not within the scope of this paper to investigate trading or attention on an individual level. This individual level would entail researching the detailed trading behavior and effects of certain events on the attention of a person who acts upon the stock market. If applied, future research will then capture, more accurately, the underlying effects this paper attempts to measure. That is, instead of using market wide trading volume, it should focus on all separate trades executed in the market by a person. Also as a substitute for general attention grabbing events, it should researched whether a stock is truly at one’s attention. Other researchers have either implemented the first or the second, but none have attempted the combination. Additionally, future papers can focus on other aspects of these attention grabbing events, such as including a frequency factor of lagged returns, and finding a better measure for the frequency of either of these events. Also future research might focus on drawing a direct parallel between a large stock market and a small stock market and test the difference in effects between the two. It was not in the scope of this paper to do so, but it is worthwhile to investigate whether home bias influences small stock market behavior.
Appendix
A.1. Correlations Table A.1.
Correlation matrix
The table below indicates the correlations between the most relevant variables included in the model plus an additional parameter for weekly news frequency, calculated in a similar fashion as the monthly frequency. Abnormal trading index is the relative level of trading volume in company i at date t weighed by the relative level of trading in the market. News is a dummy variable which is equal to one at date t and t+1, if company i is mentioned in a headline at date t and zero otherwise. Frequency is the sum of the ‘News’ variable for the previous 21 trading days for company i at date t. Lagged return is the absolute previous trading day return for company i at date t. Size is the market value of all free float shares of company i at date t, measured in millions of euros. Market volume index is the relative level of trading volume averaged for all companies listed at date t. Return is the daily return for company i at date t. Strategic holdings is the relative amount of company i’s free float shares held by strategic investors at date t. Beta is the previous five year, monthly, measure for company i’s stock beta at date t. Capital is a dummy variable equal to one if company i issues shares at date t and zero otherwise. Age is the number of years company i has been quoted at the Euronext Amsterdam exchange at date t. Earnings report is a dummy variable equal to one if company i issues an earnings report on date t and zero otherwise. Monday, Tuesday, Wednesday, Thursday, and Friday are dummy variables, which are equal to one if the trading day t is on a Monday, Tuesday, Wednesday, Thursday and Friday respectively. January, February, March, April, May, June, July, August, September, October, November, and December are dummy variables, which are equal to one if the trading day t is in January, February, March, April, May, June, July, August, September, October, November, and December respectively.
Abnormal trading index News Weekly frequency Monthly frequency Lagged return Size Market volume index Return Strategic holdings Beta Capital Age
News 0.093
Week frequency 0.085 0.684
Month frequency 0.042 0.463 0.682
Lagged return 0.152 0.100 0.078 0.041
Size 0.087 0.188 0.256 0.369 -0.147
Market volume index -0.049 0.013 0.014 0.000 0.047 -0.002
Table A.1. (continued)
Abnormal trading index News Weekly frequency Monthly frequency Lagged return Size Market volume index Return Strategic holdings Beta Capital Age
March 0.000 0.014 0.024 0.051 -0.005 0.008 0.033 0.007 0.003 0.004 0.001 -0.001 April 0.010 0.004 0.002 0.000 0.002 0.008 -0.098 -0.006 0.002 0.005 0.001 0.003 May 0.001 -0.011 -0.009 0.008 -0.007 0.007 -0.090 -0.005 -0.007 0.005 -0.002 0.003 June 0.005 -0.014 -0.020 -0.032 -0.002 0.006 -0.039 -0.010 -0.008 0.003 0.001 0.004 July 0.005 0.001 -0.004 -0.020 -0.003 0.008 -0.077 -0.002 -0.008 -0.001 0.005 0.002 August -0.004 0.003 0.007 0.021 0.054 -0.001 0.087 -0.019 -0.014 -0.003 -0.002 0.002 September -0.003 -0.017 -0.025 -0.027 0.013 -0.003 -0.034 -0.003 0.009 -0.003 -0.004 0.002 October -0.007 0.005 0.004 -0.029 0.005 -0.010 0.094 0.008 0.005 -0.007 0.000 -0.006 November -0.007 0.021 0.033 0.046 0.010 -0.014 0.028 -0.005 0.004 -0.003 -0.002 -0.005 December -0.002 -0.102 -0.012 0.008 -0.026 -0.008 -0.203 0.014 0.008 0.000 -0.002 -0.003 Residuals 0.960 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Table A.1. (continued)
Earnings report Monday Tuesday Wednesday Thursday Friday January February March April May June July
Table A.1. (continued)
Earnings report Monday Tuesday Wednesday Thursday Friday January February March April May June July August September October November December
September -0.033 0.003 0.002 -0.005 0.003 -0.003 -0.092 -0.089 -0.092 -0.087 -0.091 -0.091 -0.094 -0.093
October 0.004 0.000 -0.002 0.005 -0.003 0.000 -0.094 -0.091 -0.094 -0.089 -0.093 -0.092 -0.096 -0.095 -0.094
November 0.006 -0.002 0.004 -0.033 -0.003 0.005 -0.092 -0.089 -0.092 -0.087 -0.091 -0.091 -0.094 -0.093 -0.092 -0.094
December -0.032 0.009 -0.001 -0.006 -0.005 0.003 -0.091 -0.088 -0.091 -0.086 -0.089 -0.089 -0.092 -0.091 -0.090 -0.092 -0.090
