Excerpt
Table of content
1. Introduction
2. Literature Review
2.1. Literature on real activities manipulation
2.2. Literature on the information content of Conference Calls
3. Hypotheses Development
3.1. Discussions and inquiries about R&D cuts during Conference Calls
3.2. The information content of Conference Calls with regard to R&D cuts
4. Sample Selection
5. Definition of Variables
5.1. Variables for assessing the probabilities of inquiries and discussions
5.1.1. Definition of inquiries and discussion about R&D cuts
5.1.2. Definition of control variables
5.2. Variables for measuring the information content of R&D cuts
5.2.1. Definition of proxies for information content
5.2.2. Definition of control variables
6. Test Design
6.1. Regression models analyzing inquiries and discussions about R&D cuts
6.2. Regression models analyzing the information content of R&D cuts
7. Descriptive Statistics
7.1. Summary Statistics
7.2. Correlation Matrix
8. Results
8.1. The effect of real activities manipulation on inquiries about R&D cuts
8.2. The effect of real activities manipulation on discussions about R&D cuts
8.3. The information content of inquiries
8.4. The information content of discussing R&D cuts
9. Summary and Conclusion
Appendix
References:
List of Tables
Table 1: Sample Selection Procedure
Table 2: Summary Statistics.
Table 3: Pearson Correlation Matrix.
Table 4: Logistic regression model with INQUIRY as dependent variable
Table 5: Logistic regression model with DISCUSSED as dependent variable
Table 6: The information content of INQUIRY..
Table 7: The information content of DISCUSSED.
Table A1: List of words used in the search procedure
Table A2: Number of different firms.
Table A3: Number of Firm Years by Year
Table A4: Percentage of Firm Years by Year
Table A5: Descriptive Statistics DISCUSSED = 0
Table A6: Descriptive Statistics INQUIRY = 0
Table A7: Descriptive statistics of the logarithm of analysts following.
1. Introduction
This study is an attempt to combine two important streams of accounting research: The problem of earnings management (hereafter EM) and the role of conference calls (hereafter CCs) as disclosure medium. In doing so, I focus on real activities manipulation (hereafter RM) through cutting R&D expenses. I contribute to the existing literature by answering two questions: Firstly, whether the risk of managers engaging in RM via R&D spending affects the probability of analysts, or management addressing those cuts during CCs. Secondly, if the analyst community benefits from such discussions, by obtaining useful information, not accessible via other information channels. To answer these questions, I examine the 4th quarter earnings conference call transcripts of 300 firm years with an increased RM risk. I use content analysis to measure analysts’ and managers’ reactions, and create dummy variables that contain the information found.
In the first part of my empirical analysis, I examine the reactions to R&D cuts, as a function of the risk of RM. To determine this risk level, I use the context of a situation in which managers have strong incentives to manipulate earnings: meeting the mean analyst forecast. I use logistic regression models to determine the effect of RM risk on the probability that analysts and managers will address the R&D cuts. My results show weak evidence that indicates a positive relation between the RM risk level and analysts’ likelihood of inquiring about R&D cuts. This result is consistent with the notion that analysts serve as monitors to the firm they cover. Managers’ reactions on the other hand, are not affected by the risk of RM. For them, the decision to address R&D spending seems to depend on whether CCs are generally their primary medium to disclose such information. This is a strong indication that CCs work as substitutes for other mediums, when it comes to disclosing information about R&D expenses.
This substitute theory is examined in more detail in the second part of my analysis. I use a linear regression model to examine whether discussing R&D cuts during CCs has any incremental information value. My proxies for information content are forecast dispersion and forecast error. I find no evidence that such discussions benefit analysts’ forecast ability, which supports the substitute theory. This result is rather surprising, given the well-established idea in prior literature, that CCs are generally used to disclose new information. The remaining part of my paper is structured as follows: Section 2 provides a literary framework on RM and CCs. Section 3 develops the hypotheses for the empirical analyses. Section 4 describes the sample, Section 5 the variables and Section 6 my test design. Section 7 displays the descriptive statistics and Section 8 the results of my tests and their interpretation. Section 9 gives a brief summary and an outlook into future research. Finally, the Appendix provides tables with additional information.
2. Literature Review
2.1. Literature on real activities manipulation
Earnings management occurs when “managers use judgement in financial reporting and structuring transactions to alter financial reports to either mislead some stakeholders about the underlying economic performance of the company, or to influence contractual outcomes that depend on reported accounting numbers.” (Healey & Wahlen 1999, p.6) This definition - and accounting research generally – distinguishes between two types of EM: Accrual-based management (financial reporting) and real activities manipulation (transactions). While managers might use a combination of both types, the two are usually considered to be substitutes, with the level of RM being chosen first (Cohen & Zarowin 2010, Zang 2012). This choice depends on the specific costs of EM a firm faces. Accrual-based management is considered to increase the risk of litigations, and scrutiny by auditors and analysts, whereas RM might represent a sub-optimal corporate decision that leads to inferior subsequent performance (Cohen et al. 2010, Gunny 2005, Zang 2012, Roychowdhury 2006).
Although the research on accrual-based management is more extensive, some results suggest that managers actually prefer RM. In their survey of financial executives, Graham, Harvey & Rajgopal (2005) found that managers are not only willing to sacrifice future cash flows for accounting numbers in certain situations, but they also prefer RM over accrual-based management. Consistent with this outcome Cohen, Dey & Lys (2008) provide evidence of a significant decline in the use of accrual-based management and a significant increase of RM after the passing of the Sarbanes-Oxley Act of 2002. Other possible reasons for these preferences – besides avoiding the risk of litigation and scrutiny mentioned above - are lower private costs for managers from RM, such as the influence on their reputation (Roychowdhury 2006). Furthermore, RM might be harder to detect, since it can be disguised as a strategic decision. Lastly, managers are limited in the use of accrual-based management by the type of business operation and the reversal of prior accrual-based management (Barton & Simko 2002).
Prior literature suggests at least three different motives behind EM: First, personal gains for the management (Baber, Fairfield & Haggard 1991, Matsunga & Park 2001, Schipper 1989). Those gains are mostly the compensation or bonuses that managers receive for reaching certain earnings goals. With the exception of Burgsthaler & Dichev (1997), who argue that using EM to move from a small loss to a small gain maximizes the manager’s individual utility - an idea based on the prospect theory. Second, EM can be used as a signal for future growth (Lev 2003, Subramayan 1996). Gunny (2010) shows that firms engaging in RM to meet earnings benchmarks have a superior subsequent performance to firms that refrain from doing the same and therefore slightly miss the benchmark. However, some research suggests that firms using EM actually perform worse subsequently (Degeorge, Patel & Zeckhauser 1999, Gunny 2005, Bohjraj, Hribar, Picconi & McInnis 2009). Lastly, managing earnings might result in benefits that have a positive impact on future firm performance. This benefit can consist of a premium the market pays for meeting certain thresholds (Bartov, Givoly & Hayn 2002). Furthermore, meeting thresholds might increase the credibility and help to achieve more favorable contract terms with stakeholders (Burgsthaler et al. 1997, Lev 2003). Finally, the firm can benefit from lower borrowing costs by appearing less risky due to smooth earnings (Truemann & Titman 1988) and meeting the requirements of debt covenant contracts (Lev 2003).
Previous research documents different types of RM. One of the most frequently used types – and subject of this paper - is the reduction of R&D cost in order to decrease reported expenses (Baber et al. 1991, Bens, Nagar & Wong 2002, Bushee 1998, Dechow & Sloan 1991, Gunny 2010). They are an attractive target for RM, since they usually don’t provide benefits in the current period and, depending on the GAAP applied, firms might not be allowed to capitalize them.[1] Other types include SG&A expenses (Gunny 2010) and the reduction of discretionary expenses as a whole (Cohen et al. 2008, Roychowdhury 2006, Zang 2012). Moreover, timing the sales of fixed assets (Bartov 1993, Gunny 2010) and the increase of cash flows from operations (Cohen et al. 2010, Burgsthaler et al. 1997). Finally, temporary increases of sales through price discounts (Roychowdhury 2006, Cohen et al. 2010) and decreasing the cost of goods sold via overproduction (Cohen et al. 2010, Gunny 2010, Zang 2012).
RM is hard to detect, even more so than accrual-based management, because in theory one would have to know the intentions of the management behind certain decisions. Therefore, researchers often conduct their analyses in the context of situations with high incentives to engage in RM (Dechow & Skinner 2000). One of these situations occurs when managers can use RM to just meet/beat certain thresholds. The most commonly used thresholds are avoiding losses (Hayn 1995, Burgsthaler et al. 1997, Roychowdhury 2006, Gunny 2010), meeting last year’s earnings (Baber et al. 1991, Bushee 1998, Jacob & Jorgensen 2007) and the mean analyst forecast (Bartov et al. 2002, Matsunaga & Park 2001), with a hierarchy in this order (Degeorge et al. 1999). Another situation involves the reduction of R&D spending by chief executive officers in their final years in office (Dechow & Sloan 1991). Bens et al. (2002) document that managers cut R&D expenses for share repurchases in order to avoid earnings-per-share dilution by employee stock options. Cohen et al. (2010) examine RM in the context of seasoned equity offerings. Moreover, some researchers use these settings to apply models that predict the “normal” spending levels of RM targets and compare them with the actuals, to measure RM activity (Cohen et al. 2010, Gunny 2010, Roychowdhury 2006). Additional research focuses on ways to influence the use of RM. Besides the findings of Chen et al. (2007) mentioned above, Bushee (1998) shows that institutional investors mitigate the problem of myopic R&D cuts and Cheng (2004) suggests that compensation committees can curb the opportunistic reduction of R&D investments for personal gains.
2.2. Literature on the information content of Conference Calls
CCs are one of the most important disclosure mediums to date. In their annual survey of Investor Relations Officers in 2014, the National Investor Relations Institute found that 97% of the firms interviewed hold quarterly CCs, an increase of 21% over the last two decades. While there are CCs for special occasions (e.g. to announce mergers, acquisitions, etc.), this paper focuses on CCs held in conjunction with quarterly earnings announcements. CCs are usually composed of two parts: First, the presentation held by the management, followed by a question and answer session, which allows for discussion between management and participants. Those participants are mostly analysts. Another important distinction has to be made between CCs that are only accessible for anyone specifically invited (hereafter closed CCs) and CCs that are accessible to anyone interested in listening, although not everyone is allowed to participate by asking questions (hereafter open CCs).
Among the first to examine CCs is Tasker (1998). She documents a negative relation between the information content of a firm’s financial statement and its probability of hosting CCs. She concludes that CCs help bridging the information gap by conveying information that is hard to convey using other mediums. Frankel, Johnson & Skinner (1999) are the first to suggest that CCs have incremental information value and are not just a substitute for the accompanying press releases, as measured by increased stock price volatility and trading volume (mostly large trades) during the period the CC is hosted. The proposed reason is the less formal situation of CCs, in which manager might be more inclined to give forward looking information, due to lower legal liability. Similarly, Bowen, Davis & Matsumoto (2002) confirm those results, using the change in forecast error and dispersion of analysts’ forecasts as proxy for information content. Since their sample is entirely composed of closed CCs, they infer that closed CCs represent a form of selective disclosure. Hence, they widen the information gap between analysts allowed to attend (and their clients, usually institutional investors), and other, usually small investors. This approach was later used by Bassemir, Novotny-Farkas & Pachta (2012) and applied on a sample of German companies with similar results, thereby being the first to focus on a setting outside of the U.S. Bushee, Matsumoto & Miller (2002) provide evidence that open CCs are associated with an elevated stock price volatility and number of small trades during the period the CC is held, compared to closed CCs. This indicates that open CCs allow small investors a more equal access to information. Brown, Hillegeist & Lo (2004) on the other hand, argue that any asymmetry caused by regularly held, closed CCs is only temporary, and they do in fact mitigate information asymmetry among equity investors over the long-term. Kimbrough (2005) suggests that CCs mitigate the under-reaction of investors to earnings surprises, using post earnings announcements drifts as a proxy. This implies that CCs help investors’ understand the future implications of current earnings surprises, and accelerate their reactions to them.
Partly due to the findings above, the U.S. Securities and Exchange Commission decided to pass Regulation Fair Disclosure (hereafter Reg FD), which prohibits firms from selectively disclosing material information to analysts and institutional investors.[2] This essentially banned the use of closed CCs. Causing a lot of controversy, several papers examined the impact of this regulation. While some argued that it will reduce information quality (e.g. Janakiraman, Radhakrishnan & Szwejkowski 2002), most results pointed to a success. Sunder (2002) provides evidence that Reg FD mitigated the problem of asymmetric information (measured by the bid-ask spread) caused by closed CCs. Irani (2004) argues that both, forecast accuracy and consensus of analysts’ earnings forecasts improved after the passing of Reg FD, implying that the information quality has increased. Finally, Bushee, Matsumoto & Miller (2003) show an increase in price volatility and the number of small trades conducted during the CCs for firms that previously held closed CCs, consistent with a more equal access to information.
While Reg FD has significantly reduced the problem of information asymmetry, another concern is managers using the flexibility they have left in open CCs to discriminate among analysts. Chen & Matsumoto (2006) document that analysts, issuing more positive stock recommendations, generally have more accurate forecasts in the subsequent quarter. They conclude that managers reward those analysts by providing them with private information. Based on these findings, Mayew (2008) shows that analysts with strong buy recommendations are twice as likely being allowed to ask questions in CCs, as analysts with strong sell recommendations are. Even within the group permitted to ask questions discrimination exists, with the order of the questions determined by the analysts’ view of the firm. Mayew, Sharp & Venkatachalam (2010) suggest that analysts allowed to ask questions during CCs are more likely to adjust their forecasts to good news than they are to bad news. Moreover, asking questions enables analysts to obtain private information, which results in an increased forecast accuracy and timeliness of their forecasts. Both sides profit from this exchange of benefits: Analysts gain an advantage over their peers, while management can pressure analysts into issuing favorable forecasts and recommendations, making it harder to compete for those analysts that refuse to do so.
The most recent research has started to examine the actual content of CCs in more detail, as well as the reasons why they are more than just substitutes for other mediums. Hollander, Pronk & Roelofsen (2010) analyze management’s disclosure choices during CCs and find that managers regularly decide to withhold information from the public. In 6 out of 10 cases the management left questions posed by analysts unanswered. Matsumoto, Pronk & Roelofsen (2011) compare the presentation part and the question and answer session, and demonstrate that the question and answer session contains significantly more information. They use abnormal stock returns during each segment as proxy. Doran, Peterson & Price (2010) and Price, Doran, Peterson & Bliss (2012) use computer-based content analysis to examine the incremental information effect of linguistic tone in CCs. Their findings point to a significant positive relation between textual tone and abnormal stock return, as well as trading volume. Moreover the subsequent post earnings announcement drift is affected in the same way. Similarly, Brochet, Naranjo & Yu (2013) examine the complexity of linguistic tone. Using the same proxies, their results indicate a reduction in information content with an increase in the complexity of tone. Highlighting the importance of the verbal communication aspect of CCs, Mayew & Venkatachalam (2012) use vocal cues to assess the emotional state of the management during CCs. They find that when put under analysts’ scrutiny, the emotion exhibited by the management are positively related to future performance. Finally, Chen, Demers & Lev (2013) document a significant relation between the time of day a CC is held and the tone of the CC. The communication between management and analysts turns significantly negative as the day passes by. This negative tone results in more negative abnormal stock returns, during, and immediately after the CC.
3. Hypotheses Development
3.1. Discussions and inquiries about R&D cuts during Conference Calls
In the first part of my empirical analysis, I examine the reactions of managers and analysts in CCs to the reduction of R&D expenses, during the year concluded by the call. More specifically, I assess how the elevated risk of RM through cutting R&D spending influences the decision to discuss said cuts. I use two different variables to measure the reactions of managers and analysts. INQUIRY focuses exclusively on the group that is the main subject of this paper, analysts. The second variable, DISCUSSED, on the other hand, measures if there is any reaction to the R&D reductions, either by analysts or managers. As previous literature suggests, meeting certain benchmarks is a common incentive for managers to engage in RM, possibly via the reduction of the firm’s R&D spending. Consequently, firms that just meet certain benchmarks can be considered high risk firms with respect to RM. In order to detect firms with such an elevated risk, I use the mean analyst forecast as a benchmark to detect potential RM. The prospect of missing this benchmark creates a situation in which managers are tempted to use RM to avoid this failure. I use earnings surprise as a proxy for meeting and beating the mean analyst forecast. Consequently, firms with low earnings surprises – those that just meet the benchmark – are the ones I consider “high risk” firms. Hence, by measuring the effect of just meeting the benchmark on analysts’ and managers’ probability to discuss R&D cuts, I can assess the influence of RM on this probability.
Two things are essential to determine the effect RM has on INQUIRY: Firstly, if analysts are able to identify firms that cut R&D spending due to RM. Gunny (2005) provides evidence that analysts do recognize the myopic reduction of R&D expenses and its future implications. Assuming these findings hold, the second crucial issue is, whether analysts benefit from questioning R&D cuts during CCs to uncover potential RM. If analysts are more likely to inquire in cases of a high RM risk, this behavior is consistent with the monitoring role analysts hold according to Yu (2008). A decreased probability indicates that analysts avoid the topic, if the risk of RM is high. Such a result would support the suggestion of Graham et al. (2005), who argue that analysts have an interest in making the firm they cover look good. Moreover, it might be possible that analysts with a more critical view of the firm are not allowed to ask questions during the call, and those that are, fear they might lose this privilege if they put management under scrutiny (Mayew 2008, Mayew et al. 2010). Finally, it is possible that analysts recognize RM and consider meeting the benchmark as a sufficient reason for cutting R&D spending, which makes any further inquiries in these cases unnecessary. An insignificant effect of the risk of RM could be the result of a failure of my test design to properly capture RM. Moreover the assumption that analysts are able to detect RM might not hold. Lastly, RM simply might not be an issue that analysts consider important enough to take into consideration when asking questions. Lastly, CCs might not be the primary medium through which they pose these questions. Given these conflicting possibilities, I do not make any prediction about the influence of RM risk on analysts’ decision to address R&D cuts:
Hypothesis 1: The risk of RM does not affect analysts’ decision to question R&D cuts during CCs.
The effect of earnings surprise on DIFFERENCE is a combination of the effects on management and analysts. Thus, I likewise need to take the firm’s management into consideration. Obviously, the managers know whether they engaged in RM during the preceding period. Generally, it seems reasonable to assume that they have no interest in discussing R&D cuts, if the reason is meeting an earnings benchmark. This discussion would clearly undermine this achievement. However, in such cases managers might initiate a discussion to shift analysts’ focus from RM to other potential reasons for reducing R&D. Nonetheless, I consider this scenario unlikely, since it presumes that analysts are incapable of recognizing this strategy. Finally, it is possible that RM has no influence on management’s decision to address R&D spending reductions, since it is not an important topic for managers. The results for the effect of earnings surprise on DISCUSSED just show an overall effect, without isolating the potentially different effects on managers and analysts. However, due to reasons discussed later, it is possible to draw conclusions for both groups. Given the different potential motives, I do not make any predictions on whether managers are influenced by RM risk. Therefore, I state my hypothesis in a non-directional form:
Hypothesis 2: The risk of RM does not affect the probability of R&D cuts being discussed during CCs.
3.2. The information content of Conference Calls with regard to R&D cuts
In the second part of my empirical analysis I examine, if discussing R&D cuts (or inquiries about them by analysts) in CCs has any incremental information value for the analysts participating. As mentioned above, there has already been a great deal of research that examines the information content of CCs. From the impact they have on the information environment as a whole, to the individual segments and the tone. My proxies for measuring information content are forecast error and forecast dispersion. This choice is based on the work of Barron, Kim, Lim & Stevens (1998) (hereafter BKLS), which links the properties of the information environment with the properties of analysts’ forecasts. In their research they present a model that suggests analysts will use new, more precise information to derive at more accurate forecasts. Their model distinguishes between public information, shared by all analysts, and private information that is idiosyncratic and uncorrelated. It seems reasonable that an increase in the precision of public information will decrease uncertainty about the future, enabling analysts to make more precise forecasts, which in turn reduces the mean forecast error. However, in the BKLS model this is only the case if public information is more precise than private information, since the increase will result in analysts relying more on public and less private information. Consequently the increased precision in public information might be offset by losing the advantage from diversification using idiosyncratic, private information. Due to the reasons brought forward in the work of Bowen et al. (2002) and Bassemir et al. (2012), I will assume that public information is more precise and thus will decrease the forecast error.[3] The effect of public information on dispersion is less complex in the BKLS framework: An increase in public information will decrease the forecast dispersion, as long as private information is held constant. The effect on forecast error is identical for private information: An increase in the amount of information will lead to more precise forecasts. Since private information is uncorrelated, increasing it will result in a higher information asymmetry among analysts and will cause dispersion of their forecasts to increase.
Assuming that discussing (or posing questions about) R&D cuts in CCs does in fact provide additional information to analysts, it can be assumed this information is public, given the context in which it is revealed. Hence, I will interpret a decrease in forecast error as an increase in public information. Despite this public setting, it might be possible for analysts to derive additional private information from the CCs in general and with respect to R&D expenses. Prior literature on this subject suggests that earnings announcements and CCs might enable analysts to generate private information from publicly available information (Barron, Byard & Kim 2002, Mayew et al. 2010). Since a possible increase in private information will increase forecast dispersion, the overall effect of incremental information on forecast dispersion remains unclear.
As will be shown later, there is only a small amount of inquiries in my sample (16 out of 223), which suggests that there is nothing to be gained for analysts from questions on this subject. This either implies that talking about the reduction of R&D costs has no incremental information value, or that in firm years in which R&D cuts are not discussed, it is not necessary to do so because this information is communicated through a different channel (e.g. accompanying press release). This suggests that CCs are substitutes for other mediums, when it comes to conveying information about R&D expenses. My hypotheses are as follows:[4]
H3: Forecast dispersion is not affected by analysts’ inquiries about R&D cuts during CCs.
H4: Forecast error is not affected by analysts’ inquiries about R&D cuts during CCs.
H5: Forecast dispersion is not affected by discussions about R&D cuts during CCs.
H6: Forecast error is not affected by discussions about R&D cuts during CCs.
4. Sample Selection
My sample selection process is described in Table 1. I start with a sample of 553 firm years that fulfill the following criteria: book value of assets above 100 Mio U.S. Dollar, at least 3 analysts following the firm, difference between actual earnings and mean analyst forecast ≥ 0, a decrease in R&D expenses and R&D intensity (R&D expenses divided by sales) of at least 3%. Those criteria assure that for all firm years there is a risk that the firm has engaged in RM by reducing R&D spending. Moreover, all firms are listed in the U.S. This is important because under U.S. GAAP, R&D costs are generally charged as expenses when incurred.[5] Thus, it is not possible to manage earnings by changing the R&D capitalization rate. I sort those firm years by date and select the 300 latest. The main reason for preferring more recent firm years is the decrease in availability of CC transcripts, the older the firm years get. I replace 7 firm years of 2014 with the next ones on the list, since I need data for the subsequent year to calculate forecast dispersion and error. I use the LexisNexis database and the webpage seekingalpha.com to obtain the year-end (usually identical with the 4th quarter) CC transcripts of those firm years. I replace 45 firm years due to missing CC transcripts and 11 because some information in the transcript does not match the sample criteria. The CC transcripts of those 300 firm years are the ones I check for inquiries and discussions of R&D cuts. For all models tested in this paper, I have to delete 71 firm years due to missing data for changes in (intensity of) R&D expenses and 6 firm years which I consider outliers. Those removed firm years are not being replaced. This composes my initial sample of 223 firm years for all regression models. For the regression models with DISCUSSION and INQUIRIES as dependent variables, I have to delete another 34 firm years due to missing data for several control variables (mainly components of operating expense), resulting in a sample of 189 firm years. Form my test of forecast dispersion I have to eliminate 48 firms due missing data, necessary to calculate dispersion itself, leaving 175 firm years. The sample for testing forecast error consists of 193 firm years, because 30 had to be deleted due to missing data for forecast error.[6]
Abbildung in dieser Leseprobe nicht enthalten
5. Definition of Variables
5.1. Variables for assessing the probabilities of inquiries and discussions
5.1.1. Definition of inquiries and discussion about R&D cuts
In order to determine if the reduction of R&D expenses is discussed during CCs I define the variable DISCUSSED. It is a dummy variable that equals 1 if R&D cuts are discussed in any way during the CC and 0 otherwise. It is neither important in which segment the discussion occurs, nor by whom it was initiated. Moreover, it should be noted that I do not asses the credibility or plausibility of the discussions and the reasons stated in any way, but simply check for their existence. However, it is not enough that the cuts are merely mentioned by management or analysts, without any further explanation or reason provided. This decision is based on the fact that the management typically provides a summary of the financial results - often including R&D expenses - in their presentation without any strategic intention. Typical reasons stated in the CCs examined include increased efficiencies, business restructuring and the postponement of R&D expenses to the next year. To detect relevant parts of CCs that might contain such discussions I use the search function in Microsoft Office Word to analyze the CC transcripts. I search for words from a list with terms I consider to be relevant to this specific subject. I developed this list after reading the first 10 transcripts from top to bottom and then added synonyms for the terms found in them.[7] For terms other than variations of “R&D”, I examine if their context leads to the conclusion that R&D costs are being discussed. When I have exhausted this list without finding any discussion about R&D cuts, the variable equals 0. As a precaution to this approach, I also read every 20th transcript entirely.
The second dependent variable that analyzes the CC content with respect to R&D cuts is INQUIRY. It equals 1 if at least one analyst addresses those cuts during the CC. The search procedure and list of terms applied are the same as for DISCUSSED. Contrary to the variable DISCUSSED however, I consider it sufficient if an analyst merely states that R&D spending has been reduced over the prior year. There are two reasons for this decision. Firstly, it seems unlikely that analysts waste the possibility to ask a question by simply stating something rather obvious. This assumption is supported by the findings that analysts personally profit from asking questions during CCs (Mayew et al. 2010). Therefore, I assume that they make such statements with the intention of eliciting a response from the management. Secondly, the two times analysts limited their participation to such a statement, it nonetheless initiated a discussion about R&D cuts with the management.[8] I do not distinguish between analysts taking the initiative with their question and questions in response to a statement made by the firm’s management. Neither do I assess the credibility or plausibility of the statements made. Finally, it should be noted that while unlikely, the possibility exists that my list failed to include crucial terms that caused me to miss discussions or inquiries.
5.1.2. Definition of control variables
ES is the earnings surprise for year t, calculated as the difference between actual earnings and the mean analyst forecast, deflated by assets at the beginning of year t.[9] It functions as the main measurement of RM in my regression models. Since I expect manager to use RM in order to just meet or beat analysts forecast, a lower earnings surprise indicates a higher risk of RM. Thus, a negative relation between ES and DISCUSSION/INQUIRY would imply that managers/analysts tend to address R&D cuts if there is a high probability of RM through R&D expenses. A positive relation on the other hand, implies that managers/analysts avoid this topic, given a high probability that the reason behind R&D cuts is RM. For the reasons mentioned in my hypothesis, I do not make any predictions with respect to significance or direction of the effect ES has.
The other dependent variables’ main purpose is to control for additional factors influencing analysts’ and managers’ decision making. However, there has not been much research, at least to my knowledge, about what influences analysts’ and managers’ decision to address certain topics, and avoid others, during CCs. Hence, there is no established approach with common control variables for this regression model. Therefore, I will explain my control variables for this model in greater detail. RDGUIDANCE is an indicator variable that equals 1 if the management provides any guidance for the R&D expenses to be expected in the subsequent quarter or year. The search procedure used to find relevant content and the list of terms applied, are identical to those for DISCUSSED. I do not make any distinction between a positive, negative or constant outlook. The reason I include this variable is to assess the information environment of the firms with respect to information about R&D expenses. A significant, positive relation between RDGUIDANCE and DISCUSSED/INQUIRY suggests that there is generally more information about R&D expenses - beyond the R&D cuts - conveyed during those CCs, in which analysts or managers address the spending reductions. This is a strong hint that those firms use CCs as their primary communication channel to talk about R&D, whereas other firms might use different ways to communicate the same news. Consequently, the reason R&D reductions are being discussed is the fact there are only few other sources for that information available for those firms, compared to firms that have no discussions or inquiries with regard to this topic in their CCs. Therefore, guidance on R&D expenses provided by management indicates a weak information environment, which is compensated by discussions on R&D cuts and questions by analysts.
RDCHANGE is the change in R&D intensity. It is calculated as the difference between the R&D expenses for year t and year t-1, divided by the sales of their respective years. This variable accounts for the magnitude of the R&D cuts. It seems reasonable that more severe reductions are more likely to draw the attention of managers and analysts. Consequently, I expect a more severe decrease in R&D spending to increase the probability of a discussion/inquiry on this subject.
DIFFERENCE is the difference between the change in R&D spending and the change in operating expenses, scaled by sales. Operating expenses are calculated as the sum of R&D expenses, COGS (cost of goods sold), SG&A (selling, general and administrative) expenses and depreciation. The changes are calculated by subtracting the R&D expenses and the operating expenses of the previous year t-1 from the respective expenses of the CC year t. This difference is divided by the value of sales in year t:
Difference
The variable compares the spending on R&D costs with the overall operating expenses of the firm. Due to the sample selection, the change in R&D expenses (ΔR&D) is always negative. A decrease in the operating expenses for year t leads to an increase in value of DIFFERENCE. If the change in operating expenses is large enough, DIFFERENCE will actually be positive. Hence, a higher value signals an overall cost reduction, which implies that the proportion and significance of R&D reductions, relative to the change in other cost components, decrease. The variable controls for the possibility that an overall cost reduction might make it less likely that, out of all reductions, R&D is singled out during the CC. Since analysts might as well point out a reduction of SG&A, or COGS. Generally, it is not possible to infer that a higher or lower value of DIFFERENCE can be associated with an increased probability of RM using R&D cuts. SG&A expenses and cost of goods sold are common targets for RM as well, and managers might engage in more than one type of RM at a time (Gunny 2010). Therefore an operating cost reduction could be a strategic cost savings initiative (as frequently argued by managers during CCs) or a combination of several RM types. However, the significance of the correlation between DIFFERENCE and ES, displayed in the correlation matrix in section 7.2., will provide an answer to this question. A positive correlation would imply that managers usually limit themselves to one type of RM. A negative correlation on other hand, would suggest that managers prefer to combine several methods. In both cases, DIFFERENCE could also be interpreted as a second proxy for RM.
[...]
[1] For more information on the GAAP applied by the firms in this sample see Section 4.
[2] Selective Disclosure and Insider Trading, Securities and Exchange Commission, 17 CFR pts 240, 243, and 249
[3] One of the important reasons suggested is the quality and quantity of public information available. For more information see Bowen et al. 2002, Appendix.
[4] It should be noted that the interpretation of the predicted results for forecast dispersion remain somewhat ambiguous. Due to the effect of private information on forecast dispersion it is possible that both, public and private information are in fact increasing and just offsetting their influences on the dispersion of analysts’ forecasts.
[5] See Statement of Financial Accounting Standards No. 2, Paragraph 12 and FASB Accounting Standards Codification 730.
[6] The Tables A2, A3 and A4 provide more detailed information about the composition of the sample. For more information, see Appendix.
[7] Table A1 in the Appendix provides a detailed list with all terms used for the search procedure.
[8] Consequently, for all firm years with inquiries, the variable DISCUSSED equals 1.
[9] Hereafter, I will label the year concluded by the 4th quarter/year-end CC examined as “year t”. The year preceding year t will be labeled as “year t-1”, the year subsequent to the CC as “year t+1”.