Master Thesis
Tone in earnings conference calls:
A study on tone determinants and how certain traders might respond differently to tone in conference calls
Sarah van Kempen – 5624703 MSc. Business Economics
Specialization: Finance Thesis Seminar Finance - group 6
Supervisor – dr. T. Jochem 14th August 2015
2
Statement of Originality
This document is written by Student Sarah van Kempen 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.
3
Abstract – Topic of this research are earnings conference calls and the tone used in these calls. Whereas a lot of research has focused on the way firms may choose to use a certain tone in conference calls to signal some extra information about the firm (Davis et al., 2014; Doran et al., 2012) a more recent study by Davis et al. (2014) investigates whether there might be unintentional influences on the tone used in conference calls. They claim to be one of the first to look into how manager characteristics such as age, gender, education and their own optimism could influence tone used by management (Davis et al., 2014). In my study I build forward on the research by Davis et al. (2014) and try to find other manager characteristics that might influence the tone used. One of these additional measures for optimisim/overconfidence is based on Malmendier and Tate (2008) who claim that holding on to options while these could have been exercised can also be used as an indicator for manager optimisim/overconfidence. Transcripts are downloaded from seekingalpha.com using a web crawler program written by my thesis supervisor dr. Jochem. Tone in conference calls is computed using Excel VBA and Loughran and McDonald (2011) and Henry (2006) wordlists. Next to this I read a research by Blau et al. (2015) that states that it would be important to investigate whether reactions to tone in earnings conference calls might be different for different investors and they use short sellers as more sophisticated investors. They find evidence that short sellers do indeed process tone differently than other investors and that this can influence firms future returns (Blau et al., 2015). Therefore I also look into the possible effect of short seller presence on tone used in conference calls and firms’ returns as suggested by Blau et al. (2015). After putting together the dataset and running the various regressions needed to answer my hypotheses, none of the results turn out to be significant. My dataset provides no evidence that manager characteristics have a significant effect on tone used in conference calls as found by Davis et al. (2014) nor do I find evidence that short sellers process tone in conference calls in combination with the earnings surprise differently from other investors, while Blau et al(2015) did find this result.
Key words: Earnings conference calls, manager characteristics, manager overconfidence, manager optimism, tone, Loughran and McDonald wordlist, Henry wordlist, short sellers, sophisticated traders, Excel VBA
4
Table of contents
1. Introduction 5
2. Literature review 7 3. Data and sample collection 13
4. Research Design 14
4.1 Determining Tone 14 4.2 Managerial and manager characteristic hypothesis 14 4.3 Short seller presence hypotheses 17 5. Descriptive statistics 20
5.1 Dataset construction 20 5.2 Summary statistics 20 5.3 Correlations 22 6. Results 25 6.1 Manager characteristics results 25
6.2 Short seller presence results 28 7. Robustness checks and additional analysis 33
8. Conclusion and discussion 39
Literature 44
Appendix A: Variable definitions 48
Appendix B: Distribution of conference calls per day 50
Appendix C: CAR regressions using Henry wordlist 51
5
1. Introduction
In today’s society information plays a key role and there is a lot to do around information flows. For example, the efficient market hypothesis by Fama (1991) states that in an efficient market prices will reflect all the information currently available in that market. So that when new information becomes publicly available, prices will adjust quickly to this. Nowadays a lot of companies –voluntarily- organize conference calls around for example earnings announcements in which they explain last quarter’s earnings (Chen et al., 2013; Doran et al., 2010). Also, these calls give room for analysts to ask questions and therefore it is an important way of communicating for the firm (Chen et al., 2013; Doran et al., 2010). Furthermore, the acceptance of the Regulation Fair Disclosure by the SEC has made it impossible for firms to display certain types of information to certain pre-selected groups alone and so everyone can access these calls, there is no room for restricted conference calls anymore (Bushee, Matsumoto & Miller, 2004). There is a growing body of literature that focuses on the extra information earnings conference calls might be giving away by examining the tone that is used in these calls by the management (Davis et al., 2014). Tone can either be positive or negative and is measured by counting positive and negative words in financial disclosures (Davis et al., 2014). Question is whether the tone used in the conference calls is explained by current earnings alone or maybe by other factors as well such as, for example, managerial traits (Davis et al., 2014). Other areas of research concerning textual analysis focus on, amongst other things, readability of financial reporting and its effect on stock prices/firm value (Loughran & McDonald, 2014c). So important in the area of research on conference calls is how investors in general are affected by these conference calls, i.e. how they respond to these calls (Frankel, Johnson & Skinner, 1999) and if and how managers/firms could use a certain tone (Davis et al., 2014) or a degree of readability (Loughran & McDonald, 2014c) to influence the calls and what they represent. It is not just the academic world that examines financial documents and the information they hold. At the beginning of this year, for example, Warren Buffett was being criticized for disclosing too little information about his company using only an annual letter to the shareholders and an annual meeting for financial disclosure (Financial Times, 8th February 2015). In this thesis, I want to explore how the tone used in conference calls is explained by economic circumstances the firm is in and possibly other factors such as managerial traits (Davis et al., 2014). When you know what explains the tone used in conference calls, the way you respond to these calls could be different: Tan et al. (2014) for example show –in an experimental setting- that experienced investors react more negatively to positively framed
6
information with low readability while less experienced investors react positively. This also relates to a study by Blau, DeLisle and Price (2015) where they find that short sellers, who they view as sophisticated traders, interpret tone in conference calls different from other investors. Overall contribution to existing literature will be a more extensive look into the specifics of the determinants of tone used in conference calls by including managerial traits and the presence of different investors. My eventual data sample will include conference calls from December 2012 until march 2015. Research question central in this thesis is:
‘Do manager characteristics influence tone used in conference calls and are there differences in reaction to tone by different investors?’
This is the main question in this thesis. Based on studies by amongst others, Blau et al. (2015) an important focus in this thesis will be on how investors use the information that these conference calls might give away by setting a certain tone. Blau et al. (2015) claim that it is highly important to look at how investors respond differently to tone used in conference calls. They claim that earlier literature suggests there is reason to believe that investors with different levels of experience will have different responses to tone used in conference calls and there is need to look at these differences, and they claim to be the first to actually study this (Blau et al., 2015). The reason why looking at this response is important is because according to Blau et al. (2015) having sophisticated traders in the market might give companies less chance to manage stock prices –at least in the long run- by setting an unjustified tone.
Tan et al. (2014), as mentioned earlier, also investigate differences in response for different investors but then with respect to readability of financial transcripts. So there seems to be a new idea in current literature suggesting that we would not so much need to focus on how the tone used –be it by use of positive/negative words or lower/higher readability- in conference calls influences the market as a whole, but more specifically how it affects certain groups and what this means for the firms holding the conference calls (Blau et al., 2015). As Blau et al. (2015) suggest, more sophisticated traders acting on the tone used in conference calls might contribute to market efficiency and will make it harder for companies to cover up lower earnings than expected or to use positive language to conceal less positive upcoming future earnings. In this thesis I will first add some certain characteristics suggested by earlier research to get a more informed measure of tone. After that I will elaborate on the recent research by Blau et al. (2015)
7
and Tan et al. (2014) on how different investors respond differently to the message behind the message and what this might implicate for firms.
The conference call transcripts I use are downloaded from seekingalpha.com. Other variables needed for answering the question of interest will come from I/B/E/S, Execucomp, CRSP and Zephyr. Analysis needed to test hypotheses and answer the research question include amongst other things a regression on tone and testing for abnormal returns. The thesis will be build up as follows: in the next part I will give an overview of existing literature in the areas of interest building up to the hypotheses. After that I will discuss how data is gathered, how the sample is constructed and which methods will be used. Then the results will be summarized and discussed, answers will be given to the hypotheses and the main question which will be followed by a conclusion and discussion with recommendations for further research.
2. Literature review
Prior research has mainly focused on how managers may choose to use a specific tone in financial documents, to signal some extra information about the future performance of the firm to the public (Davis et al., 2014; Doran et al., 2012). So the idea is that, the conference calls around earnings announcements could give away some extra information next to the information explained by current performance by setting a specific tone and this influences market responses (Davis et al., 2014). First of all, I would like to discuss conference calls and their importance, without looking at tone used.
Conference calls have two parts, the first part has a prepared presentation by the manager/CEO and the second part is a question and answer part which gives analysts the opportunity to ask the manager questions (Matsumoto, Pronk & Roelofsen, 2011). Why conference calls can be so interesting is well explained by Frankel et al (1999). They show that these conference calls give away additional information to complement the information that was in the press release foregoing the conference call and that it gives opportunities to market participants, mostly analysts, to get clarifications (Frankel et al., 1999). This is also discussed in a research by Brown, Hillgeist and Lo (2004) where they show that voluntarily organising conference calls reduces asymmetrical information. Like Frankel et al. (1999), Brown et al. (2004) show that the use of conference calls makes investors better informed. Also Brown et al. (2004) suggest that organising conference calls frequently might reduce the cost of equity for firms by lowering asymmetrical information. A more recent research by Matsumoto et al. (2011) shows that management gives away extra information in the earnings calls next to the discussion of the
8
earnings announcement itself. They show a positive relation between firms with disappointing results and managers’ tendency to reveal more information about the future (Matsumoto et al., 2011). Furthermore they claim that the question and answer (Q&A) part of the call, is more informative than the foregoing presentation by the management explaining the past earnings (Matsumoto et al., 2011). This question and answer part in conference calls is interesting because, in contrast to the presentation part, management can’t prepare their exact answers (Chen et al., 2013).
Recently, a lot of studies have focused on the tone used in conference calls and how this influences markets. Huang, Teoh and Zhang (2014) refer to the tone in earnings press releases which is not explained by current performance as a form of ‘tone management’. Their results suggest that managers can misinform their investors by using a specific tone in their press releases (Huang et al, 2014). They show that when there is a positive tone in a press release which is not explained by current firm performance or firm fundamentals, that this positive tone is related to negative future firm performance (Huang et al., 2014). Investors do respond to this positive tone, Huang et al. (2014) find that stock prices increase after the overly optimistic portrayal in the earnings press release. They also show that it are mainly older firms that engage in using tone to mislead investors (Huang et al., 2014). So what this research suggests is that management can –at least in the short run- keep their stock prices up by using an optimistic tone (Huang et al., 2014).
An important article for my research is the study by Davis et al. (2014), they claim to be the first to add an extra dimension to the area of tone management research, something they refer to as ‘managerial style’. This is interesting because they suggest that managers have certain personality traits that in itself affect how optimistic or pessimistic they are. Therefore these managers may have a more optimistic or pessimistic tone than another manager would have had in the same situation (Davis et al., 2014).
What makes earnings press releases and conference calls so interesting is that they are voluntary and therefore do not face the same rules as financial statements such as 10-K’s (Huang et al., 2014). This might bring certain responsibilities with it, as a research by Rogers et al. (2011) shows for example, lawsuits are most frequently based on conference calls. And these charges in turn can lead companies to change their disclosure policy (Rogers & Van Buskirk, 2009). Research by Rogers and van Buskirk (2009) suggests that firms that suffer legal consequences after –voluntary- financial disclosures, will choose to reveal less after that. According to them, firms subject to lawsuits after voluntary disclosures will change their disclosure behaviour and
9
reduce the information they choose to provide to the bare minimum required by legislation (Rogers & van Buskirk, 2009). In a later research Rogers et al. (2011) investigate whether there is a relation between the tone used in conference calls and lawsuits. They find that firms with more optimistic tone have higher chances of getting sued (Rogers et al., 2011). These two studies suggest that managers and companies can’t just choose to set any tone they like in a conference call, using an unjustified tone could be followed by a lawsuit based on rule 10(b)-5. The lawsuits based on Rule 10(b)-5 focus on misleading shareholders or investors, by for example giving a too optimistic view of the companies’ future in financial statements such as 10-k’s or conference calls (Rogers & van Buskirk, 2009).
The Davis et al. (2014) article investigating managerial traits’ influence on tone, considers characteristics such as schooling, gender, age and charitable work next to prior work experience as well. They claim that these factors could contribute to managers being more optimistic (Davis et al., 2014). Currently there are a lot of other studies done in the field of behavioural finance with respect to manager optimism and overconfidence and what could be indicators for this. In a study by Schrand and Zechman (2012) in which they investigate a relation between manager overconfidence, optimism in financial reporting and misreporting, they find variable compensation to be an indicator for more optimistic reporting. They show that for the more optimistic than justified reports, managers are less dependent on variable pay (Schrand & Zechman, 2012). This is remarkable since being more dependent on variable pay is thought of as a reason to portray a more optimistic view: when you get paid for performance you may have more reason wanting to report positive earnings (Schrand & Zechman, 2012). The reason for this optimism while not depending on variable pay would be managers being overconfident (Schrand & Zechman, 2012). So maybe overconfidence could lead to more optimistic tone, Malmendier and Tate (2008) for example show that overconfident managers choose to hold on to their stock options much longer than rational, indicating that those managers believe that the company will do better. Sen and Tumarkin (2014) however, claim that holding on to options is not just determined by managerial optimism but also by other managerial traits and they developed another variable which they claim is a better estimator of managerial overconfidence. They find that, retaining some of the shares when exercising the option is a key characteristic of managerial optimism (Sen & Tumarkin, 2014). Another possible indicator of CEO overconfidence is discussed in Malmendier et al. (2011), who find that CEO’s overconfidence is also shown in the way CEO’s choose to finance their operations. In their research, overconfident CEO’s tend to prefer internal financing (Malmendier et al., 2011).
10
A very recently published paper by Blau et al. (2015) investigates differences in reaction to tone in conference calls for different investors, their focus is on the Q&A part. Their results show that sophisticated traders, who are represented by short sellers in this case, sell stocks short of those firms whose conference calls are overly optimistic in tone and that have a positive earnings surprise (Blau et al., 2015). They suggest that there is a group of investors who process the same tone and information differently, more efficiently, understanding what this information truly signals (Blau et al., 2015). They find that increased short selling around the extremely positive conference calls is an indicator of negative future firm performance (Blau et al., 2015). Blau et al. (2015) claim that the presence of sophisticated investors makes it harder for firms to use a tone that is not explained by current performance in order to increase their stock price. Massa, Zhang and Zhang (2014) label this as the ‘disciplining hypothesis’, which they explain to be the idea that short sellers could possibly influence managers’ or firms’ behaviour in a good way. Massa et al. (2014) find evidence in their study that this ‘disciplining hypothesis’ is at work where short sellers are present.
Another subject frequently studied in literature concerning earnings announcements, is the effect of an earnings surprise on stock prices. As suggested by Fama’s efficient market theory (Fama, 1991), stock prices should adjust to new information as soon as it becomes publicly available. Hirshleifer et al. (2008) however describe a delayed market response to earnings surprises known as ‘post-earnings announcement drift’. Mendelhall (2004) mentions it is also sometimes referred to as the ‘standardized unexpected earnings’ effect. This post-earnings announcement drift describes the delayed response with which stock prices slowly incorporate the new information of the earnings surprise, which could take up to about three quarters after the announcement was made (Collins & Hribar, 2000; Hirshleifer et al., 2008). In their study, Hirsleifer et al. (2008), investigate whether the actions of individual traders have had an impact on the post-earnings announcement drift. They assume that these individual traders will have an underreaction to the earnings surprise because they possibly are less sophisticated traders (Hirsleifer et al., 2008). They refer to this idea as the ‘individual trading hypothesis’ (Hirsleifer et al., 2008). Although this idea is suggested in some research, as pointed out by Hirsleifer et al. (2008), they do not find evidence for this. Implication of post-earnings announcement drift is that it keeps stock prices at unrealistic levels and it is in conflict with the idea of an efficient market, since here any new information should immediately be incorporated into the prices without any delay (Fama, 1991). Mendenhall (2004) finds that one possible explanation of post-earnings announcement drift is that while sophisticated traders –arbitrageurs- are capable to
11
detect the mispricing, it might be too risky or costly to act upon. He finds a positive relation between the size of the post-earnings announcement drift and the riskiness of the stock (Mendenhall, 2004). Therefore, when studying the effect of the presence of short sellers in the event of an positive earnings surprise in combination with an overly optimistic tone, this post earnings announcement drift will be taken into account.
Before looking into why the short seller would qualify as a sophisticated investor as suggested by Blau et al. (2015), I want to discuss what it is that short sellers do. The term short selling refers to selling an amount of shares that you do not own right now (Berk & DeMarzo, 2011). You can do this by borrowing the shares from someone else (Berk & DeMarzo, 2011). So the short seller will get the share price for which it sells today, but since the share was borrowed from someone else, the short seller will eventually have to return the share to the owner by buying back the share and also possibly paying dividends that were paid out during the time the share was borrowed (Berk & DeMarzo, 2011). Reason for selling a stock short would be that the investor expects the price of the stock to go down (Bodie, Kane & Marcus, 2011).
As one of the first to look into short selling and the effects of it, Miller (1977) described that the presence of short sellers puts downward pressure on the price of the underlying stock because the short sales increase the supply of stock. Opinions on whether the presence of short sellers is a good or a bad thing seem to be split. There is a lot of research that shows that short selling would contribute to price efficiency and so more efficient markets (Massa et al., 2014). In September 2008 however, the SEC decided to ban short selling of financial stock in the US because of the potential –unjustified- harm it was doing to companies1. So there is a lot to do around short sales and their effect on stock prices, I will now look at the idea of the short seller as an informed investor.
A study by Boehmer and Wu (2013) for example, shows that the presence of short sellers in stock markets contributes to price discovery as prices are closer to their underlying fundamental value in these markets. Besides that, they conclude that restricting short selling impedes markets because short sellers are indeed informed investors who enhance market efficiency (Boehmer and Wu, 2013). Also Boehmer, Jones and Zhang (2008) find that short sellers are more informed traders and that their effect on stock prices is lasting and thus indicating that they are not speculative traders. Furthermore a study by Engelberg, Reed and Ringgenberg (2012) finds that short sellers’ trades are twice as profitable compared to other investors in case of public
12
news announcements. Which suggests that a short seller is better at processing publicly available information than the average investor is (Engelberg et al., 2012). Dechow, Hutton, Meulbroek and Sloan (2001) find that short sellers target companies which are overpriced -based on fundamentals to price ratios- and therefore can be seen as sophisticated traders. Another study by Drake et al. (2015) investigates how short sellers respond to earnings restatements. Drake et al. (2015) have a conclusion similar to Blau et al. (2015) in the sense that both studies find that short sellers do not anticipate the news that is going to be announced but that they react to these announcements more efficiently than other traders.
Based on the studies discussed above, I find enough reason to believe the short seller can be seen as a sophisticated trader and so I will follow the research by Blau et al. (2015) to test whether short sellers process tone more efficiently than average or less experienced investors do.
Based on the foregoing literature, I formed three hypotheses to answer my research question: (1) Manager characteristics influences tone used in conference calls
(2) Short interest will increase in case of a positive earnings announcement together with a positive tone
(3) The presence of short sellers will mediate the effect of an overly optimistic tone on stock price reaction following a positive earnings surprise
For the first hypothesis I will use the variables by Davis et al. (2014) which they claim are good indicators of manager optimism but I will add extra variables suggested by behavioural finance literature such as factors of overconfidence here as well. I believe that adding variables such as late exercise of stock options (Malmendier et al., 2008; Campbell et al. 2011) and level of variable pay (Schrand & Zechman, 2012) could possibly contribute to a more complete indicator of managerial traits’ effect on tone used in conference calls.
The second and third hypotheses follow the research by Blau et al. (2015), they find that short sellers target firms with positive earnings surprises and a tone that is more optimistic than it should be based on the underlying fundamentals. The underlying rationale behind the third hypothesis is that short sellers are more sophisticated in processing information and will be able to see through a certain tone set by management (Blau et al., 2015). Since they claim to be the first to look into this and only use a data sample of 1300 transcripts in the period 2005-2006 (Blau et al., 2015), I will investigate whether I find a similar result for a more recent dataset. So based on Blau et al (2015), what I expect to see in the third hypothesis is that cumulative
13
abnormal returns following a positive earnings announcement with a positive tone will be lower or maybe even negative.
3. Data and sample collection
Period of interest in my sample will be conference calls around earnings announcements for the period 2004-2014 and will include US firms. Reason for US firms is that for these firms I know the data will be available because of the SEC’s regulation Fair Disclosure. The time period of 2004-2014 is chosen because I would like to include finding a possible change in tone after the crisis. For additional analysis, different time periods can be used then.
Data collection is done by downloading the conference calls transcripts for different firms. These conference calls can be downloaded for free from seekingalpha.com. For this I use a webcrawler program, written by my supervisor Dr. Jochem. Next step is transforming the transcripts from html script to text file, which is done using the same program.
Other data that I will need to collect is information on the managers, who was manager, where and at what time? Following the research by Davis et al. (2014) I will collect this data by using the Execucomp database, this means that my sample will include S&P1500 firms. The Exexucomp database, which is part of the Compustat database, will give detailed information on long and short term compensation, annual compensation and information on who the CEO is and also gender and age information. Furthermore I will need the stock returns from the companies for which the managers work and these stock returns can be downloaded from CRSP. Next to that, information needs to be gathered to control for the part of the tone in the conference calls that is justified by, amongst other things, the firm’s performance and the earnings surprise itself (Davis et al., 2014; Doran et al., 2012). This additional information will be gathered using I/B/E/S and CRSP. Short interest, which is used as an indicator of short seller presence, can be downloaded from the supplemental short interest file provided by Compustat.
14
4. Research Design
4.1 Determining Tone
Finding the tone used by managers will be done by using two word lists. One of the word lists is the Henry word list, which is also used by Price et al. (2012). The other one I will be using is the Loughran and McDonald wordlist (2011), which they developed especially for financial documents and I can download this list from their website. In contrast to Davis et al. (2014) I choose not to use the Diction word list. Current findings by Loughran and McDonald (2014a), who are some of the main researches investigating textual analysis of financial documents and readability of the financial documents, show that in financial reporting the Diction word list is not useful. The use of two wordlists will help to check for robustness of results. So the Loughran and McDonald wordlist can be downloaded from McDonald’s website2. The Henry wordlist is available in her 2006 paper, however can also be found in Price et al. (2012). To establish the tone used in the conference call transcripts, I will use excel VBA to count the positive/negative words using the two wordlists.
Loughran and McDonald updated their wordlist in march 2015, I was able to download the master dictionary. This dictionary contains 2355 negative words and 354 positive words. Also I downloaded the litigious wordlist from this master dictionary and this contains 903 litigious words. The negative and litigious wordlist contain updates until 2014, the positive wordlist does not contain any new words (no words added since 2009). Determining whether the tone in the conference call is positive or negative I follow the Blau et al. (2015) research, who also used the Loughran-McDonald wordlist. They determine tone as a ratio:
𝑇𝑜𝑛𝑒 = 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠−𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠
𝑇𝑜𝑡𝑎𝑙 𝑜𝑓 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑎𝑛𝑑 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠 (1)
According to Blau et al. (2015) this gives a good relative measure where a positive tone will reflect more positive words than negative ones and a negative tone more negative words than positive ones. I also use the Henry wordlist, which I got from Price et al. (2012), this contains 104 positive words and 85 negative words.
4.2 Managerial tone and manager characteristics hypothesis
The first hypothesis is about the determinants of tone used in conference calls, focusing on the influence of managerial overconfidence. This hypothesis will be tested following research by
15
Davis et al. (2014), combined with another variable that was indicated and developed by Malmendier and Tate (2008). However their longholder variable which can be used as indicator for managerial overconfidence is hard to compute with the data that I have and therefore I use a proxy of their variable as developed by Campbell et al. (2011). The longholder variable, following Campbell et al. (2011) is calculated as:
𝐿𝑜𝑛𝑔ℎ𝑜𝑙𝑑𝑒𝑟 = 𝑆𝑡𝑜𝑐𝑘 𝑝𝑟𝑖𝑐𝑒 𝑎𝑡 𝑓𝑖𝑠𝑐𝑎𝑙 𝑦𝑒𝑎𝑟 𝑒𝑛𝑑
𝑆𝑡𝑜𝑐𝑘 𝑝𝑟𝑖𝑐𝑒 𝑎𝑡 𝑓𝑖𝑠𝑐𝑎𝑙 𝑦𝑒𝑎𝑟 𝑒𝑛𝑑− 𝑇𝑜𝑡𝑎𝑙 𝑟𝑒𝑎𝑙𝑖𝑧𝑎𝑏𝑙𝑒 𝑣𝑎𝑙𝑢𝑒 𝑢𝑛𝑒𝑥𝑒𝑟𝑐𝑖𝑠𝑒𝑑 𝑒𝑥𝑒𝑟𝑐𝑖𝑠𝑎𝑏𝑙𝑒 𝑜𝑝𝑡𝑖𝑜𝑛𝑠 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓𝑢𝑛𝑒𝑥𝑒𝑟𝑐𝑖𝑠𝑒𝑑 𝑒𝑥𝑒𝑟𝑐𝑖𝑠𝑎𝑏𝑙𝑒 𝑜𝑝𝑡𝑖𝑜𝑛𝑠
(2)
Campbell et al. (2011) set the cut off point for longholder at more than 70% in the money. I set it at 67%, as Malmendier and Tate (2008) use this cut off point for their 67% in their longholder time variant variable. So the longholder variable is a dummy variable equal to 1, in case the CEO is a longholder in that quarter. The regression testing for manager effects is the following
𝑇𝑜𝑛𝑒𝑖𝑡 = 𝛼 + 𝛽1𝑀𝐵𝐸 + 𝛽2𝐸𝑆 + 𝛽3𝐿𝑂𝑆𝑆𝑖𝑡+ 𝛽4𝑅𝑒𝑡𝑢𝑟𝑛𝑖𝑡 + 𝛽5𝐺𝑟𝑜𝑤𝑡ℎ + 𝛽6𝑅𝑂𝐴𝑖𝑡 + 𝛽7𝑅𝑂𝐴𝑖𝑡+1+ 𝛽8𝑆𝐸𝑂𝑖𝑡+1+ 𝛽9𝑀𝐴𝑖𝑡+1+ 𝐹𝑖𝑟𝑚𝑖+ 𝑌𝑒𝑎𝑟𝑡+ 𝑄𝑢𝑎𝑟𝑡𝑒𝑟𝑘+
𝑀𝑎𝑛𝑎𝑔𝑒𝑟𝑖𝑎𝑙 𝑡𝑟𝑎𝑖𝑡𝑠𝑗 + 𝛽10𝐿𝑜𝑛𝑔ℎ𝑜𝑙𝑑𝑒𝑟 + 𝜀 (3)
Davis et al. (2014) claim that the first nine variables specified here will account for the firms current and future performance, they describe MBE as meeting the market expectations; ES is the earnings surprise; loss is whether the firm faces a loss in this quarter, which will be indicated by net income; and return is this fiscal quarters adjusted return. Furthermore they use ROA as a profit variable and add some predictions about future performance by ROA for next period and the chance of mergers and acquisitions and SEO’s as indicator for strategic reasons behind tone (Davis et al., 2014). I choose only to use current profit and next periods’ instead of next four quarters as done by Davis et al. (2014) since I also look at short run abnormal returns in another hypothesis. Davis et al. (2014) define their MA variable as a dummy where the dummy has the value of 1 when the firm announces a merger or acquisition in the following period. Using information from the Bureau van Dijk Zephyr database, I incorporate this variable into my regression as well since Davis et al. (2014) list this as a possible reason for a more optimistic tone in the call before the announcement. Therefore the expected sign of the coefficient on this variable is positive. Also data on SEO’s is obtained using the Bureau van Dijk Zephyr database. SEO is a dummy variable equal to 1 if the firm increases capital by SEO in the next quarter. The expected sign on the SEO coefficient is also positive since this could also be a strategic reason to be extra positive in the conference call (Davis et al., 2014).
16
The variables reflected by managerial traits are age and gender as specified by Davis et al. (2014). They expect gender -which is a dummy variable- to have an effect on tone (Davis et al., 2014). Based on earlier literature they expect women to be less overconfident than men and so they expect a positive relationship between tone used in the call and being a man (Davis et al., 2014). In their research results Davis et al. (2014) find a small but negative relationship between being a women and tone used in some of their regressions, meaning women would use a less positive tone. Basing my regression on Davis et al. (2014) this is what I expect to see for gender in this study as well. For age, the expectation is that the older one gets the less risk one would like to take and the less overconfident one gets (Davis et al., 2014). The expected effect on tone here would be that the older someone gets, the less likely someone will be to use a positive tone (Davis et al., 2014).
Returns on assets will be calculated as:
𝑅𝑒𝑡𝑢𝑟𝑛 𝑂𝑛 𝐴𝑠𝑠𝑒𝑡𝑠 = 𝑁𝑒𝑡 𝑖𝑛𝑐𝑜𝑚𝑒
𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠 (4)
This is the definition as given by Berk and DeMarzo (2011). Expectation would be that if the return on assets goes up, the tone will be more positive (Davis et al., 2014). Also the earnings surprise needs to be calculated, this surprise is an important variable in both hypotheses. Following Davis et al. (2014), earnings surprise will be measured as:
𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑆𝑢𝑟𝑝𝑟𝑖𝑠𝑒 = 𝐸𝑃𝑆−𝑀𝑒𝑎𝑛 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡𝑒𝑑 𝐸𝑃𝑆
𝑆𝑡𝑜𝑐𝑘 𝑝𝑟𝑖𝑐𝑒 (5)
Expectation is that a positive earnings surprise will have a positive effect on the tone used (Davis et al., 2014). Next to the variables used by Davis et al. (2014) to influence manager optimism such as gender and age (now reflected in managerial traits, as did Davis et al., 2014) I also add the following variable: ‘Variable pay’, a variable following Schrand and Zechman (2012) to explain tone. Variable pay is based on bonus payment to the CEO, which in this case will be cash. Bonus variable is calculated as bonus divided by total compensation. The expectation for the variable pay variable is that lower variable pay leads to a more positive tone, as Schrand and Zechman (2012) find, indicating CEO optimism. However, this variable –as suggested by Sen and Tumarkin (2014)- would be inferior to a share retention variable as developed by Sen and Tumarkin (2014). They created this dummy variable, which equals 1 if the cumulative of shares retained by the CEO over the year while there was the option to exercise exceeds 1% (Sen & Tumarkin, 2014). They have the data on this variable available on
17
Robert Tumarkin’s website3. Unfortunately the dataset only contains information till 2012, therefore I can not use it here. They are however updating their sample, but this will probably not be available in time to be included in this research. Another variable that I add for control and robustness checks is the litigation risk in the industry. The litigation risk dummy is based on Rogers and van Buskirk (2009) who provide information on most frequently sued industries. SIC codes are used to indicate the industry a company is in. If litigation risk is equal to 1, the litigation risk in the industry is high. Expectation for this variable is uncertain, as Rogers and van Buskirk (2011) show that firms with high positive tone get sued more often, Rogers and van Buskirk (2009) show that firms facing litigation risk might be more cautious.
4.3 Short seller presence hypotheses
The second hypothesis looks into the effect of tone on the change in level of short interest. Regression tested here, largely resembles the Blau et al. (2015) regression:
∆𝑆𝐼 = 𝛽0+ 𝛽1𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽2𝐵𝑀𝑖𝑡+ 𝛽3𝐴𝑛𝑎𝑙𝑦𝑠𝑡𝑠𝑖𝑡+ 𝛽4𝑃𝑜𝑠𝑆𝑢𝑟𝑝𝑖𝑡+ 𝛽5𝑇𝑜𝑛𝑒𝑖𝑡+ 𝛽6𝑃𝑜𝑠𝑆𝑢𝑟𝑝 ∗ 𝑇𝑜𝑛𝑒𝑖𝑡+ 𝜀 (6)
Variables Size and BM are control variables and are discussed after the next regression. Important for this regression is the number of analysts present during the call (Blau et al., 2015) which is a control variable, expected sign is uncertain. PosSurp reflects whether there is a positive earnings surprise or not (Blau et al., 2015) and Tone is the tone in the conference call as computed by the Loughran and McDonald or Henry wordlist. Also an interaction variable on tone and positive surprise is included, following Blau et al. (2015). Expectation on these variables would be a positive effect on short interest change as short sellers would target positive surprises with positive tone (Blau et al., 2015). Number of analysts present at the call is added to the regression following Blau et al. (2015). I will also run a regression of short interest on tone to look into the possible ‘disciplining’ effect that short sellers could have as suggested by Massa et al. (2014) which I discussed in section 2. This will be a very basic regression just looking for the effect of level of short interest, earnings surprise and some controls:
𝑇𝑜𝑛𝑒𝑖𝑡 = 𝛽0+ 𝛽1𝐸𝑆𝑖𝑡+ 𝛽2𝑆𝐼𝑖𝑡+ 𝛽3𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽4𝐵𝑀𝑖𝑡+ 𝜀 (7)
18
The third hypothesis is about a possible mediating effect of short sellers on the effect that tone can have on stock prices, which is based on Blau et al. (2015). Here I will conduct an event study. The event date is the date on which the earnings conference call takes place. A benchmark of expected returns will have to be created in order to compare whether returns can be indicated as abnormal. I choose to use the market model, as given in Bodie et al. (2011) to determine these expected returns.
𝐸(𝑅) = 𝛼 + 𝛽1𝑅𝑚+ 𝜀 (8)
The estimation window used to calculate the expected returns is -100 to -10 trading days before the announcement, which is about 4,5 months. My expectation, based on the research by Blau et al. (2015) is that when more short sellers are present -as measured by change in level of short interest following the announcement- the less effect tone will have on the stock price and so cumulative abnormal returns. Expectation following Huang et al. (2014) is that in the short run stock prices will go up following the overly optimistic tone used in the announcement because investors are possibly misled in the short run, as Huang et al. (2014) find in their study. However the study by Blau et al. (2015) suggests that companies will have less opportunity to use tone to keep their stock prices at favourable levels for the firm when short sellers are present. They find short sales to go up after unjustified overly optimistic tone, that these sales predict negative returns and they conclude that short sales have a strong negative relationship with the companies’ future return (Blau et al., 2015). Blau et al. (2015) show that short sellers process qualitative information, such as use of tone, more efficiently and contribute to price discovery. Therefore, the expectation regarding my third hypothesis is that when short sellers are present, there will be less positive impact of a positive earnings surprise in combination with an overly optimistic tone and following their contribution to price discovery as indicated by Blau et al. (2015) I expect that the cumulative abnormal returns will be lower when more short sellers are shorting the firm’s stock.
Important for this hypothesis is that the event window will include different sizes after the event took place to indicate whether the present of short sellers influences cumulative abnormal returns after the conference call. I will use a window of [0, +1] to see what the direct effect is on cumulative abnormal returns and to keep in mind possible post earnings announcement drift as discussed earlier, I also include a [0, +20] window which is smaller than the possible effect might last according to Hirsleifer et al. (2008) but I don’t want to use a larger window because there might be overlap with a following announcement then. However, I don’t expect the
19
cumulative abnormal returns of the [0, +20] window to show significant results because the presence of short sellers and their assumed level of sophistication (as based on Blau et al. (2015)) suggests that prices would not drift. But then again, as discussed in section 2, Mendenhall (2004) says post earnings announcement drift could still be possible if short selling is too risky or costly. The regression that I run in order to test my third hypothesis is:
𝐶𝐴𝑅 = 𝛼 + 𝛽1𝐸𝑆𝑖,𝑡+ 𝛽2𝑆𝐼𝑖,𝑡 + 𝛽3𝑇𝑜𝑛𝑒𝑖,𝑡+ 𝛽4𝐵𝑀𝑖,𝑡+ 𝛽5𝑆𝑖𝑧𝑒𝑖,𝑡+ 𝛽6𝑆𝐼 ∗ 𝑇𝑜𝑛𝑒𝑖,𝑡+
𝛽7𝐹𝑟𝑖𝑑𝑎𝑦 + 𝜀 (9)
The SI variable is the change in short interest after the announcement, I can only obtain bi-monthly short interest data but I fitted the days to nearest bibi-monthly short interest reported. Expected sign for short interest is negative, as short interest grows CAR is expected to be lower based on research by Blau et al. (2015). BM variable stands for book to market ratio and the size variable refers to the size of the firm (as measured by market capitalization), these two variables are used as control variables in Blau et al. (2015), Huang et al (2014) and Massa et al. (2014) as well. These variables will be measured as natural logarithms. Following Huang et al. (2014, p.1107) expectation is that BM will be positively related to abnormal returns, Size negatively related to abnormal returns and former returns positively related to abnormal returns. For measuring cumulative abnormal returns I largely follow a regression by Huang et al. (2014) although adding the presence of short sellers –as based on Blau et al. (2015) and a possible interaction effect of short seller presence with tone. The Friday variable is a control variable added to see whether the day of the announcement influenced the returns. This is based on the investor inattention literature that investors might be paying less attention to announcements made on Fridays (DellaVigna & Pollet, 2009).
Since the finding of Blau et al. (2015) is that short sellers target firms with positive surprises and positive tone I use the suest option in Stata to compare whether the coefficient on short interest in case of a positive earnings surprise with a positive tone is different from the coefficient on short interest in case of negative earnings and positive tone or negative earnings and negative tone.
Testing for robustness of my results will include amongst other things, adding some variables as discussed in section 4.2. Also I use the Loughran and McDonald wordlist as well as the Henry wordlist to determine tone of conference call transcripts.
20
5. Descriptives
5.1 Dataset construction
As mentioned in section three, the first step gathering my data included downloading transcripts from seekingalpha.com, using a webcrawler in excel VBA written by dr. Jochem. After that, the transcripts are transformed to text files also using this program. Running the second step of the program left me with a sample of 3643 conference call transcripts from December 2012 until March 2015. Using excel VBA I read calls into excel colums and was able to let excel count the number of positive and negative word occurrences based on the two wordlists. The sum of the positive and negative words was calculated and the tone variable is determined by the formula as described in section 4.1. I also needed other information from the transcript calls, such as ticker, date, number of analysts, the quarter it concerned and CEO name. This data was all hand collected and stored in an excel file which was later merged with the computed tone for the concerning conference calls.
After completing the tone dataset, I merged the set with the variables needed in order to run the regressions. As described in section 3, these variables were gathered from I/B/E/S, CRSP, Execucomp and Zephyr. Specification of all variables, including computation and name can be found in Appendix A. Following Blau et al. (2015) observations for financial firms are dropped. Also I deleted pennystock from the sample, Huang et al.(2014) dropped stock with a price below one dollar. In this sample penny stock is dropped when price is lower than five dollar. I was left with a total of 1245 conference calls dating from 2013 until 2015. Variables such as executive compensation and whether a CEO is a longholder or age of the CEO was rather difficult to get for the entire data sample and leads to missing values. Therefore the tone regression with manager variables will have less observations than for example the regression concerning tone and short interest.
5.2 Summary statistics
Following Huang et al. (2014), I winsorize the financial variables in my data sample. Like Huang et al. (2014) I also exclude returns from winsorizing. The cut off using winsorizeJ is (0.05 99.5) for the following variables: ROA, ROAt+1, BM and Growth. Most important descriptive statistics of regression variables can be found in table 1. Looking at table 1 it shows that for the dataset used in this research the tone measured by the Henry wordlist is more positive than for tone measured using the Loughran and McDonald wordlist. Which might be because the Loughran and McDonald wordlist was especially developed for financial texts
21
(Loughran & McDonald, 2011). This difference in tone between the Henry and Loughran and McDonald wordlist is also present in Davis et al. (2014), in their sample the mean of the Henry wordlist is also above the mean of the Loughran and McDonald wordlist. Mean tone measured with the Loughran and McDonald wordlist -which can be found in table one row one, column four- is not much different from the mean tone in Davis et al. (2014), 0.295 versus 0.3397 in their sample (though they have a higher standard deviation which is 0.5238) Average age in the sample is 56, however the dataset only contains age information on a smaller subset of data. Mean earnings surprise is slightly negative, with a value of -0.005 which is below the average of Davis et al.(2014) -who have an average surprise of 0.001- and the Blau et al.(2015) sample. However Blau et al. (2015) have higher outliers. Mean earnings surprise in this dataset is very close to zero which means that the average company in the dataset just about meets the earnings forecast. The Friday variable based on DellaVigna and Pollet (2009) who find that announcements made on Fridays tend to have higher post earnings announcement drift. In this sample 8.43% of the earnings conference calls takes place on a Friday, which is slightly higher than in DellaVigna and Pollet (2009), there Friday announcements make up around 6% of the sample. The precise distribution can be found in appendix B. What is notable in this dataset is the amount of men in proportion to women, it is commonly known that there are much more male than female CEOs but in this sample women are underrepresented. The data gathering however was random and independent so this underrepresentation of women is considered to be a coincidence. Mean short interest of the data sample is around 6% and the variable is winsorized to exclude extreme outliers.
22
Table 1: Descriptive Statistics
Variable N Min Q1 Mean Median Q3 Max Stand. Dev.
ToneLM 1245 -0.412 0.167 0.2950 0.308 0.706 0.746 0.1917 ToneH 1245 0.1 0.538 0.6250 0.649 0.725 0.959 0.1447 Male 1245 0 1 0.9679 1 1 1 0.1764 Age 257 38 52 56 56 60 87 6.6451 VarPay 257 0 0 0.0267 0 0 0.6042 0.0884 Longholder 257 0 0 0.4358 0 1 1 0.4968 Friday 1245 0 0 0.0843 0 0 1 0.2780 Litrisk 1245 0 0 0.0562 0 0 1 0.2304 MBE 1245 0 0 0.6763 1 1 1 0.4681 ES 1245 -0.4918 -0.0006 -0.0005 0.0004 0.0021 0.2144 0.0224 ESr 1245 -0.4855 -0.0007 -0.0006 0.0004 0.0021 0.215 0.0225 Growth** 1245 -0.5767 0.0072 0.2053 0.0804 0.2001 4.1993 0.5658 Loss 1245 0 0 0.2594 0 1 1 0.4385 Returni 1245 -0.2639 -0.0149 0.0053 0.0039 0.0255 0.5951 0.0529 Returnvw 1245 -0.0232 -0.0015 0.0021 0.0016 0.0052 0.0218 0.0061 Returnew 1245 -0.0213 -0.0002 0.0022 0.0021 0.0037 0.0224 0.0056 Returnsp 1245 -0.0228 -0.0015 0.0019 0.0012 0.0057 0.0240 0.0062 ROA* 1245 -0.3486 -0.0006 0.0059 0.0103 0.0215 0.1801 0.0415 ROAt+1* 1182 -0.3121 -0.0009 0.0039 0.0086 0.0187 0.1939 0.0361 SEOt+1 1245 0 0 0.0112 0 0 1 0.1054 MAt+1 1245 0 0 0.0193 0 0 1 0.1376 Analysts 1245 0 4 7.2771 7 10 22 4.007 BM* 1240 -3127 0.182 0.413 0.348 0.573 3156 0.405 Size 1245 10,245 13,498 14,708 14,558 15,842 20,270 1,688 SI* 1245 0.0006 0.0175 0.06046 0.03323 0.07889 0.4219 0.0679 This table provides information on the distribution of the most important variables used in the regressions. Financial variables are winsorized following Huang et al (2014). LitRisk reflects the litigation risk in the industry based on research by Rogers and Van Buskirk (2009) and longholder is a dummy variable based on research by Malmendier and Tate (2008) and Campbell et al. (2011). An overview of the definitions of all of the variables used can be found in Appendix A. *winsorized (0.5 99.5). **winsorized(1 99).
5.3 Correlation
Next to using the dependent, independent and control variables in a regression I also look at correlation of these variables. Table 2 shows a correlation matrix on a subset of the most important dependent and independent variables used in several regressions. I will shortly discuss some of the correlations displayed in table 2.
23
The correlation between short interest and tone measure based on the Loughran and McDonald wordlist which can be found in the first observation of the last row in table 2 below, show a weak but significant negative relation between the level of short interest and the tone based on Loughran and McDonald. This indicates that the two variables move in opposite directions so when short interest goes up, the tone measure will go down. This negative relationship might relate to the disciplining hypothesis as researched by Massa et al. (2014), for they find that the presence of short sellers could discipline companies to use a less positive tone. The correlation between short interest and the Henry tone measure shows an even weaker negative relationship between tone and short interest however this correlation is not significant. The Henry wordlist is used for robustness, amongst other things but the insignificant relation between this wordlist and short interest might be because this wordlist is less suited for financial texts, which was the reason for Loughran and McDonald (2011) to create their wordlist.
The earnings surprise variable shows a positive relation with respect to both tone measures, both of them show a weak but positive significant relation. This means that when the earnings surprise goes up, the tone measure gets more positive. This is in line with expectations as mentioned in section 4.2 that a higher (positive) earnings surprise would lead to use of more optimistic tone (Davis et al., 2014). Looking at the relation between short interest and the earnings surprise -as can be found in table 2, last row and third column- these variables have a weak negative but significant correlation. This indicates that there is a lower earnings surprise when there is higher short interest. Correlation only gives an indication of the direction in which the variables moves, it doesn’t show a causal effect so I can’t say whether higher short interest leads to a lower earnings surprise or that lower earnings surprise causes higher short interest. Relating back to Blau et al. (2015) idea is that short sellers target firms with positive earnings announcements and positive tone, a positive sign was expected for correlation between short interest and earnings surprise.
The longholder variable, which is used as a measure of overconfidence based on research by Malmendier and Tate (2008) and Campbell et al. (2011), has a positive weak but significant correlation with both tone measures. This positive sign indicates that when a CEO is a longholder, the tone measure is more positive. This is in line with expectations as formulated in section 4.2. Age and tone also show a negative weak but significant relationship, which is in line with the expectation based on Davis et al. (2014) that the older one gets, the less positive the tone used in conference calls will be. The gender variable ‘Male’ shows a weak negative significant relation between the two tone measures and gender, however interpretation of this
24
coefficient is difficult because men are overrepresented in the dataset. It would however indicate that tone and gender move in opposite directions, so for women the tone is lower which is in line with the expectation based on the research by Davis et al. (2014). Variable pay, which is used as an alternative measure for overconfidence based on Schrand and Zechman (2012) has a weak positive significant relation with the tone measure based on the Loughran and McDonald wordlist but a weak negative significant relation with the tone measure based on the Henry wordlist. So in case of a positive tone on the Loughran and McDonald wordlist there is higher variable pay and in case of a positive tone on the Henry wordlist there is lower variable pay. The negative correlation is what would be expected based on Schrand and Zechman (2012) because they find that being less dependent on variable pay in combination with positive tone indicates overconfidence/overoptimism.
This table provides information on the correlation between different dependent and independent variables at a 5% significance level. The first observation in the second row for example shows a weak but significant positive correlation between the earnings surprise and the tone as measured by the Loughran and McDonald wordlist. This indicated that when the earnings surprise goes up, the tone will go up as well so if the earnings surprise is more
25
positive the tone is more positive. The SI variable is a measure of short interest which is calculated as shares sold short divided by total shares outstanding. *significant at 5% level.
6. Results
I will discuss the main results of my regressions per hypotheses. Robustness checks and other regression results will be discussed in the next section or can be found in the appendix.
6.1 Manager characteristics results
First regressions that are run concern the hypothesis that manager characteristics influence the tone that is used in conference calls. The results of this regression with age, gender and longholder as identifiers of manager characteristics are shown in table 3. The SEO dummy variable, which originally was in the regression drops out when running it because this variable turns out to be zero in all observations.
What follows from the regression results in table 3 is that none of the manager characteristic variables included here has a significant impact on the tone used in conference calls. The regression on the Loughran and McDonald tone measure, table 3 column one, shows that as age goes up the tone variable goes down. So less positive tone is used when one gets older, which was expected based on Davis et al. (2014) however the result here is not statistically significant. The gender dummy, which is labelled ‘Male’ shows a sign opposite from what was expected based on Davis et al. (2014) for the Loughran and McDonald measure excluding time fixed effect. The coefficient, shown in table 3 column one, is negative indicating less positive tone is used by men. However the coefficient is not significant and the coefficients on ‘Male’ in the other three regressions are small but positive, however also insignificant. For the Longholder variable, which should measure manager overconfidence as based on Malmendier and Tate (2008) and Campell et al. (2011) also no significant relation was found. In column two of table three, where time fixed effects are included, the coefficient indicates a very small negative effect (-0.6%) on Loughran and McDonald tone when manager overconfidence goes up. I performed an F-test with three restrictions: that the coefficients of Longholder, gender and age are equal to zero. Again this statistic appears to be insignificant in all four regressions. This means that no evidence is found that these manager characteristics have a significant effect on the tone used in conference calls. In section 7 I will discuss the results of the ‘Variable pay’ variable, which is the other measure used for overconfidence based on Schrand and Zechman (2012), in combination with the manager characteristics age and gender.
26
This table shows the regression coefficients and clustered standard errors of 𝑇𝑜𝑛𝑒𝑖𝑡= 𝛼 + 𝛽1𝑀𝐵𝐸 + 𝛽2𝐸𝑆 +
𝛽3𝐿𝑂𝑆𝑆𝑖𝑡+ 𝛽4𝑅𝑒𝑡𝑢𝑟𝑛𝑖𝑡+ 𝛽5𝐺𝑟𝑜𝑤𝑡ℎ + 𝛽6𝑅𝑂𝐴𝑖𝑡+ 𝛽7𝑅𝑂𝐴𝑖𝑡+1+ 𝛽8𝑆𝐸𝑂𝑖𝑡+1+ 𝛽9𝑀𝐴𝑖𝑡+1+ 𝐹𝑖𝑟𝑚𝑖+
27
here are age, gender and longholder. Column 1 contains the results of the regression run on the Loughran and McDonald tone measure without time fixed effects(year and quarter) and the second column contains the results of the regression run on the Loughran and McDonald measure including time fixed effects. Column three and four contain the results for the regression run on the Henry tone measure, where the third column is without time fixed effects and the fourth column including time fixed effects. The F-statistic on time fixed effects in the regression with Loughran and McDonald tone as dependent variable as shown in column 2, is statistically significant at the 5% level. All regressions exclude firm fixed effect, the idea was to run panel regressions however the average observation per firm was 1.2, so panel regressions were not the ideal measure anymore and checking for firm fixed effects did not lead to meaningful output results. None of the manager characteristics has a significant effect on the tone used in conference calls in this sample. The Henry tone measure shows more significant results than the Loughran and McDonald measure, which may be a result of the Loughran and McDonald wordlist better suiting to analyse financial documents (Loughran & McDonald, 2011). The SEO variable dropped out of the regression because it had the value of 0 for all observations. Meeting or beating the expectations by the market seems to be a significant indicator of tone used for all tone measures, significant at 5%. Clustered standard errors are given within parenthesis unless specifically stated otherwise. *significant at 10%; **significant at 5%; ***significant at 1%.
What is interesting to see in the regressions for tone and the possible determinants of it is that for MBE, all regressions have significant coefficients indicating that meeting or beating expectations leads to more positive tone used which is about 5% for the Henry tone measure and about 7% for the Loughran and McDonald measure. Looking at the effect of the earnings surprise, table 3 columns one and two show a positive effect for the Loughran and McDonald tone measure which turns out to be insignificant. The effect of the earnings surprise on the Henry tone measure, table 3 column three and four, even shows a negative but insignificant relation. This is remarkable since meeting or beating expectations has the same sign for all regressions, while this measure is based on the earnings surprise. The earnings surprise however, indicates the size of the earnings surprise while the MBE is a dummy variable. The assumption that next period merger announcements could cause a more positive tone in this quarter for strategic reasons as suggested by Davis et al. (2014) is only significant for the Henry tone measure in this sample. The control variable of LitRisk, representing litigation risk in the industry shows no significant results so I can not conclude that having higher litigation risk in the industry the firm operates in would lead to managers being more careful with their use of tone in conference calls as suggested by Rogers and van Buskirk (2009).
The other variables on firm financial performances included in the regressions following Davis et al. (2014) show that for example the effect of having a loss in this quarter leads to a slightly more positive tone used for the Loughran and McDonald tone variable of about 4.28% when controlling for time fixed effects. This is in contrast with Davis et al.’s (2014) results, they find
28
a significant negative relation between firms experiencing losses and the tone used in the conference call. The coefficients on Loss in my regressions however have no statistical significance. In line with Davis et al. (2014) results, also show a positive relation of about the same size between this period’s return on assets and tone used in conference calls although only for the Loughran and McDonald tone measure. Including firm fixed effects, table 3 column two, shows that a 1% increase on return on assets in this quarter increases tone with about 1.27%.
The R2 of the regressions run in table 3 are quite low, ranging from about 8 to 16%. This means that the variables used in these regressions are not very good at estimating the tone variable. While Davis et al. (2014) find that adding manager characteristics significantly increase the predictability of the tone used in conference calls, the regression results in my study show no support with respect to the first hypothesis that manager’s own characteristics influence the tone used in conference calls in this dataset. This could be because I use a different time period, while Davis et al. (2014) look at 2002-2009, my dataset contains conference calls from beginning 2013 to the first quarter of 2015. However, Davis et al. (2014), because of their large dataset, were able to do panel regressions and take into account firm fixed effects. This may also be an explanation for the difference in results.
6.2 Short seller presence results
The first regression on short interest change and last regressions on cumulative abnormal returns concern the idea by Blau et al.(2015) that short sellers target firms with positive earnings and a positive tone. The results of this regression of the impact of tone on change in short interest can be found in table 4, the independent tone measure here represents the tone as determined using the Loughran and McDonald wordlist. The interaction variable of tone used in the call in combination with there being a positive earnings surprise has no significant impact on the change in short interest from the day of the call until the next day. The sign of the coefficient, although insignificant, suggest a negative relation of more positive tone in combination with a positive surprise leading to lower short interest change. This is different from Blau et al.’s (2015) results who find that short selling goes up when there is a positive announcement with positive tone.
The coefficient on positive surprise in the regression without fixed effects which is shown in table 4 column one, is positive but very small (0.9%). It would suggest that when there is a positive earnings surprise, short interest goes up which is in line with expectations based on
29
Blau et al. (2015) that short sellers target firms with positive surprises. However the effect is insignificant. The regression results of table four, do not support the second hypothesis that short interest increases when there is a positive announcement and positive tone in the conference call.
This table shows the regression coefficients and clustered standard errors of ∆𝑆𝐼 = 𝛽0+ 𝛽1𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽2𝐵𝑀𝑖𝑡+
𝛽3𝐴𝑛𝑎𝑙𝑦𝑠𝑡𝑠𝑖𝑡+ 𝛽4𝑃𝑜𝑠𝑆𝑢𝑟𝑝𝑖𝑡+ 𝛽5𝑇𝑜𝑛𝑒𝑖𝑡+ 𝛽6𝑃𝑜𝑠𝑆𝑢𝑟𝑝 ∗ 𝑇𝑜𝑛𝑒𝑖𝑡+ 𝜀. Short interest is calculated as shares
sold short divided by total shares outstanding. Change in short interest represent the percentage change in the short interest after the earnings conference call. PosSurp indicates a positive earnings surprise and ToneLM*PosSurp is an interaction variable of the Loughran and McDonald tone measure and a positive earnings surprise. Idea following Blau et al. (2015) is that short sellers target firms with positive earnings surprises and positive tone. None of the coefficients in this regression show significant results, not even after adding time and firm fixed effects which is the regression in the third column. (size shows a significant negative result on change in short interest
30
after fixed effects are added however it is only significant at the 10% level and the fixed effect f-tests show no statistical significance). Firm fixed effects are excluded from other regressions since the average observation per firm is too low for the firm fixed effects to control for omitted variables. All clustered standard errors are given within parenthesis unless specifically states otherwise. *significant at 10%; **significant at 5%; ***significant at 1%.
The regression run in order to answer the third hypothesis whether the presence of short sellers has a mediating effect on cumulative abnormal returns around earnings announcements is tested in two ways, one regression and the other a comparison of means of the coefficient of short interest change as mentioned in section.
Table 5 shows the results on the cumulative abnormal return regression using the tone as measured by Loughran and McDonald. The cumulative abnormal return here is measure over a 2-day window, from the day the conference call took place until the day after that. I used two benchmarks in calculating cumulative abnormal returns, table 5 column one and two contain the value weighted market return benchmark CAR and column three and four contain the S&P500 benchmark CAR. The results show a significant positive relation between the earnings surprise and the cumulative abnormal returns. Effect of change in short interest is the variable of interest here since since -based on Blau et al. (2015)- the expectation is that short sellers target firms with positive announcements and high positive tone thereby differently processing the information that is included in this positive tone since short sellers would see this as an indicator of negative future performance (Blau et al., 2015). The expected effect on abnormal returns, based on Blau et al. (2015) would be that these returns are lower for firms who have higher short interest changes than firms with lower short interest changes as represented in the third hypothesis investigating whether short seller presence mediates the effect of CAR. The coefficients on the short interest variable show a very small negative relation between short interest change and cumulative abnormal returns for a two day window, which is the day of the call until one day after. It indicates that as short interest goes up, cumulative abnormal returns will slightly decrease. However the effect is not statistically significant. The interaction variable on short interest change in combination with tone indicates a negative effect on cumulative abnormal returns. This means that when the change in short interest goes up and the tone is more positive, the cumulative abnormal returns are lower (although hardly different from zero). The sign of the interaction variable is as expected based on the literature read, however it is not statistically significant.
