Are Financial Analysts Inattentive?
Evidence from a Natural Experiment in context of FAS 123-‐R
Program: MSc. Business Economics, Finance track Subject: Master Thesis First Draft
Name: Tessa Wanders
Student number: 6144667 Supervisor: Dr. T. Ladika Date: July 6th 2015
Statement of Originality
This document is written by Tessa Wanders who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
We provide evidence that financial analysts using EBITDA and Net Income multiples are inattentive and incorrectly adjust their earnings forecasts although no new information comes to light. We document this effect by exploiting a unique event in 2005, FAS 123-‐R. Using the random variation in the timing of FAS 123-‐R -‐ firms with fiscal year ending June or later had to comply in 2005, while all other firms could postpone compliance until 2006 – we construct a treatment and a control group. Our two stage least squares estimates suggest that analysts unjustifiably adjust their earnings estimates downwards by $0,06 for the (early complier) treatment group, although nothing actually changes. Results are especially significant for commodity-‐like industry analysts, and insignificant for the tech industry analysts. We also show in an event study that not only financial analysts react inefficiently to FAS 123-‐R, but the market as a whole does so too.
* I would like to thank my supervisor dr. Tomislav Ladika for providing me with a large part of the data, extensive feedback and useful comments. I’m also grateful for moral support from Roland Wanders and Thomas Plantenga. All errors are my own.
Table of Contents
1. Introduction ... 5
2. Background FAS 123-‐R ... 7
3. Literature Review ... 9
4. Hypotheses and Methodology ... 16
5. Data and Descriptive Statistics ... 20
6. Results ... 24
6.1 First Stage Results ... 24
6.2 Second Stage Results ... 26
6.3 Event Study Results ... 28
7. Robustness Checks ... 29 8. Conclusion ... 32 References ... 35
1. Introduction
Do financial analysts depend on detailed company research when constructing earnings forecasts, or are they inattentive and merely focus on easily observable earnings headlines? This paper’s contribution is to provide an answer to this question. A vast majority of literature confirms that analysts do not incorporate all available information in an efficient way and that it matters how information is displayed. A number of studies have shown that financial analysts can be lazy, inattentive or both. Dellavigna and Pollet (2009) show that analysts are more distracted on trading days before the weekend. Similarly, Hirshleifer, Lim and Teoh (2009) find evidence that shows analysts are also distracted when concurrent earnings are announced. Herrmann and Thomas (2005) distinguish lazy analysts by observing forecast rounding. Lazy analysts tend to forecast earnings in nickel intervals more frequently than hard-‐working analysts. Analysts play a significant role in the process of impounding information into US stock prices (Potrioski and Roulstone, 2004). They have a large group of clients and other investors that follow their forecasts and recommendations (Davies and Canes, 1978). It is therefore important for investors, companies, analysts, managers, shareholders, academics and policymakers, to understand how analysts use information and that their forecasts sometimes might be flawed.
The setting in which the research question will be tested is a 2005 change in US accounting rules; statement FAS 123-‐R. With this new standard The Financial Accounting Standards Board (FASB) required firms to expense their employee stock options. A key element of FAS 123-‐R that we will exploit to answer the research question is the fact that the compliance date across firms differs almost randomly (Ladika and Sautner, 2014). Firms with fiscal year ending in June or later had to comply with FAS 123-‐R in 2005 (treatment) the rest comply in 2006 (control). According to Damodaran (2005) FAS 123-‐R would have a big impact on the profit and loss (P&L) headlines of the treated firms. Using a Two-‐Stage Least Squares model for identification and data from the Compustat and IBES databases, we will test Damodaran’s statement in the first stage of the research. Then in the second stage, we will test whether this exogenous change in P&L headlines results in significantly lower analyst forecasts for the treatment group. It is important to note that for both treatment and
control groups, the same amount of information is available. The only difference is that the treatment group has to recognize the ESO’s as an expense, whereas the control group only has to disclose the ESO’s in a footnote. If we find significant results in our second stage regression we may conclude that analysts only use ESO information when it is easy to observe. But when the expense is hidden in a footnote they don’t worry about it.
This paper contributes to two streams of empirical literature. The first stream links accounting or tax changes to real firm outcomes or investor attitude, more specifically on the impact of FAS 123-‐R on real firms outcomes and investor behavior. The other stream engages in explaining how analysts construct forecasts. The unique contribution of my paper is that it combines both research streams and analyzes the way analysts react to the introduction of the specific US Accounting Rule FAS 123-‐R.
Our results confirm our hypothesis that financial analysts incorrectly adjust their earnings forecasts for the treatment group as a reaction to FAS 123-‐R although no new information came to light. In the first stage results we show that FAS 123-‐R significantly impacts the treatment group’s earnings. The second stage results show that as a result financial analysts incorrectly adjust their forecast estimates. We find these results to be different for the tech firm subsample. Tech firm results are stronger in the first stage but insignificant in the second stage. For the majority of the remaining tests, the tech firm subsample shows different results. We believe this is the case because as ESO’s play a large role in the tech industry, tech industry analysts already efficiently incorporated ESO information before FAS 123-‐R. Using consensus estimate standard deviation as a proxy for confusion among analysts, we show that confusion increased around firms that had to comply with the accounting rule. The forecast error, used as a proxy for analyst accuracy, shows that after FAS 123-‐R analysts become more accurate. Our event study results show that also the security markets as a whole respond inefficiently to FAS 123-‐R. The average buy and hold abnormal returns are lower for the treatment group than for the control group. We also show that results are stronger for widely followed firms than for less followed firms. In addition we show that commodity-‐like industries (e.g. Chemicals, Food, Smoke) appear to be particularly affected by FAS 123-‐R and inefficient markets.
The rest of this paper is structured as follows. The second chapter provides a brief summary of FAS 123-‐R’s history. The following chapter elaborates on the two aforementioned streams of literature that blend together in this research. Chapter 4 and 5 explain our methodology and data, respectively. Chapter 6 presents the results for the first and second stage and our event study. Chapter 7 presents the results from our robustness checks and several extensions. Finally the last chapter concludes our results and hypotheses.
2. Background FAS 123-‐R
FAS 123-‐R is the outcome of a heated debate, on how firms should treat their stock-‐based compensation when reporting their income, which had been going on for multiple decades. This section presents a summary of the legal history and a brief description of this debate. In 1972 the U.S. Accounting Principles Board issued Opinion 25 (APB 25). APB 25 ruled that the intrinsic value based method should be used to account for stock option based remuneration. This entailed that the expense concerned with option-‐based payments is the difference, if any, between the firm’s stock price on the date the option is granted minus the option strike price. As firms usually granted their employees at-‐the-‐money stock options, this difference was null, and thus so was the accounting expense.
For the first time, in 1984, the Financial Accounting Standards Board (FASB) put an Exposure Draft on the agenda that mandated firms to use the fair-‐value approach to recognize share-‐ based payments on their income statement, rather than the previously mentioned intrinsic-‐ value based method. The proposal was released in June 1993, and was met with strong resistance. The FASB received over 1,700 letters of comment, almost all of them in opposition of the Exposure Draft.
The central arguments against the fair-‐value based approach could be divided into two categories, public policy issues and reliability of measuring the fair value of employee stock options (ESOs) (Rees and Stott, 1998). Firstly, opponents reasoned that fair value recognition of ESOs as an expense would harm stock prices. This would result in companies having to cut stock plans (Khalaf 1993, Beese 1994, Harlan 1994). It would 1) stifle new businesses and high-‐tech companies, 2) companies would go out of business so jobs would
be lost and 3) it would harm overall US competitiveness with foreign markets that do not follow the fair value method. The other central argument concerned accuracy of measurement methods (Rodgers 1994, Beese 1994). Coopers and Lybrand (1993) found evidence that the expense calculated using the FASB guideline option pricing model would vary widely depending on the assumptions made. Even the Accounting Standards Executive Committee, which was initially in favor of the FASB Exposure Draft, disagreed on the reliability of the measurement method (AcSEC, 1994).
The lobby urging abandonment of the proposal grew so strong, even President Clinton warned that it would possibly undermine the competitiveness of the U.S. tech industry (Harlan 1994). In May 1994, the U.S. Senate passed a resolution that urged withdrawal and Congress gave the FASB an ultimatum. Allow the alternative of disclosure (rather than expensing) or the FASB will seize to exist as an accounting rule making body (Brown and Lee, 2003). In October 1995, the FASB chose the former, and accepted a pro forma disclosure in a footnote of earnings that reveal what earnings would be if the fair value of the ESOs would be recognized. This proposal, FAS 123, recommended fair-‐value accounting, but not surprisingly the lion’s share of firms stuck to APB 25 (Ladika and Sautner, 2014).
The discussion on ESO expensing resurged in the early 2000s, after the Enron scandal and other major corporate failures. Murphy (2003) and Murphy and Hall (2003), argued that the absence of ESO expensing fueled the surge in stock option-‐based pay. Greenspan (2002) concluded that this surge incentivized managers to “artificially inflate reported earnings in order to keep stock prices high and rising”. From July through December of 2002, 150 firms started expensing ESOs voluntarily using the fair value approach (Aboody, Barth and Kasznik, 2004a). Recognizing or disclosing ESOs has a different effect on corporate investment. Bernard and Schipper (1994), Libby, Nelson and Hunton (2006) explain that managers and auditors take more care measuring recognized items, as opposed to disclosed items. It follows from Bushman and Smith (2001), Bens and Monahan (2004), McNichols and Stubben (2008), Biddle et al. (2009) Balakrishnan et al. (2014), that this increased financial reporting quality is associated with enhanced investment efficiency. Hirschleifer and Teoh (2003) find evidence that inattentive investors focus on recognized items and ignore important footnotes. As a consequence firms can issue overpriced stocks to fund additional investment.
The FASB released a new proposal in March 2004, again proposing to expense stock options using fair-‐value method. The final standard was adopted in December 2004, called FAS 123-‐ R. Public firms were mandated to apply the fair value method to 1) all ESO awards granted after June 15, 2005 and 2) unvested ESO awards granted after 1994, in their first financial statement (either quarterly or annual) released after June 15, 2005. However, on April 14 2005, in response to worries from accountants about altering fiscal standards in the middle of the fiscal year, the Securities and Exchange Commission (SEC) decided to delay the effective date. Now firms were allowed to implement FAS 123-‐R with the start of their first fiscal year, rather than the first reporting period, after June 2005. To clarify, this meant that firms with their fiscal year ending on June 30, 2005 had to comply starting in July 2005. Whereas firms with a fiscal year ending on May 31, 2005 were exempt from FAS 123-‐R until June 2006. (Ladika and Sautner, 2014). Appendix A-‐3 clarifies which firms comply when.
3. Literature Review
This study adds to several streams of empirical literature, the most important two will be discussed in this section. The first stream links accounting or tax changes to real firm outcomes or investor attitude, and specifically we will focus on the effect of accounting rule FAS 123-‐R. The other stream we will discuss, tries to understand the process of financial analyst forecasting. This section will present the most relevant publications and theories in these different fields, and explain how the current paper aims to contribute.
Firstly, this study relates to papers that study the effects of accounting changes and corporate governance measures on real firm outcomes or investor attitude. The term “perception” reappears often in this field of research. Accounting rules seldom change an actual economic cost, yet they do change the perception of certain costs. The way managers react to altered accounting regulation often seems irrational to economists. For example, when the FASB introduced a new current accounting charge for anticipated post-‐retirement benefits (PRB) (SFAS 106), managers predicted that share prices would come down with the reported income (Hall and Murphy, 2003). So companies started making considerable cutbacks in their medical benefits for pensioners (Mittelstaedt, Nichols and Regier, 1995). Firms also attempted to correct the markets perception of the magnitude of the PRB
obligation by choosing early adoption of SFAS 106 (Amir and Livnat, 1996). However, stock prices did not fall because markets are fairly efficient and the economic costs were already incorporated in the stock price (Amir 1993, Espahbodi, Strock and Tehranian, 1991). These results are inconsistent with our expectations for the current research, because we do not expect markets to be fully efficient. As opposed to SFAS 106, with FAS 123-‐R not all firms comply at the same time. We expect that exactly due to this difference, the incorporation of information into the price will be more complex, and therefore the market might be inefficient.
Although certain accounting rules have no effect on the company cash flows, companies still respond because of the changed perceived costs. There are three theories that attempt to explain why managers undertake actions to affect income but that have no cash flow effects. Firstly, Graham, Hanlon and Rajgopal (Bartov E. 2007) (2005) interview several CFO’s. The interviewees say that they do believe stock markets are efficient, on average, however when reporting their firm’s income, they’d rather not take the chance that the market inefficiently prices it. Secondly, Sloan (1996) and Xie (2001) question market efficiency with respect to the pricing of earnings components. Hirschleifer and Teoh (2003) state that partially attentive investors pay more attention to recognized than to disclosed charges to income, therefore managers have an incentive to avoid recognizing costs. Finally Graham et al. (2005) state that managers manage reported income to signal their capability to the executive labor market or, according to Bowen, Ducharme and Shores (1995), to other stakeholders such as creditors, suppliers and employees. A study by Carter and Lynch finds evidence that suggests firms trade off financial reporting benefits against reputational costs in determining the timing of repricings to get beneficial accounting treatment. Imhoff and Thomas (1988) scrutinized capital structure changes to study the effect of SFAS No.13 (capital lease disclosures went from footnotes to balance sheet) on lessees. The study shows that, following the adoption of the standard, firms systematically substituted capital leases with operating leases and non-‐lease sources of financing. Furthermore, lessees appeared to reduce book leverage by increasing equity and reducing conventional debt. Graham, Hanlon and Shevlin (2011) provide evidence that whether being able to designate earnings as permanently reinvested under APB-‐23 (accounting for income taxes -‐ special areas) affects real corporate decisions about operation location and profit reinvestment versus
repatriation. Bens and Monahan (2005) show that banks in the U.S. avoid consolidation under FASB Interpretation No. 46 by restructuring asset-‐backed commercial paper conduits. The term perception resurfaces in studies on employee stock options, where managers act as if ESOs are free because of their zero accounting cost (before FAS 123-‐R). Hall and Murphy (2003) argue that, because of the favorable accounting treatment of ESOs and the absence of a cash outlay at the time of the grant, firms act as if the ESO perceived cost is lower than the true economic costs. Using the perceived rather than the true cost, firms tend to grant more options than they would otherwise. Oyer and Schaefer (2006) confirm that the median firm is willing to incur costs of up to $0,50 to $1 to issue options and save $1 in compensation expense. Another reason for the decadent use of ESOs is the risk-‐ averseness of managers. They value ESOs beneath their economic value, because they cannot perfectly hedge the risks imposed by ESOs (Lambert et al. 1991, Hall and Murphy 2002). To compensate for this risk, firms give out more ESOs, resulting in an increased executive pay. Numerous studies found connections between the magnitude of share-‐based compensation and real firm outcomes. Cheng and Warfield (2005), Erickson et al. (2006), Bergstresser and Philippon (2006), all established a link between ESO compensation and earnings management. Dechow and Sloan (1991) show the more stocks or options in their firm an executive owns, the less likely they are to reduce discretionary expenditures prior to their departure. Bens, Nagar and Wong (2002) find that firms experiencing significant employee stock option exercises shift resources away from real investments toward the repurchase of their own stock. Weak evidence shows that the performance of these firms tends to decline in subsequent years, possibly implying a real cost in terms of foregone investment opportunities. These studies are relevant because FAS 123-‐R, the accounting standard that will be studied in this paper, addresses the perception and accounting treatment of ESOs.
More specifically this research contributes to literature on the real firm outcomes that followed the introduction of accounting statement FAS 123-‐R. Brown and Lee (2007) investigate the determinants and consequences of FAS 123-‐R on the option based components of the compensation for the top five executives of firms. They find that the decrease in the proportion of total compensation paid in ESOs is significantly increasing in two circumstances. Firstly, in the firm’s tendency to use ESO’s favorable accounting
treatment to report higher earnings in the period before mandated expensing. Secondly, in the amount of ESO expense to be recognized upon adoption of FAS 123-‐R. The paper also shows that firms are likely to replace the ESO compensation with restricted stock, rather than other forms of compensation. Choudhary (2008) also finds evidence for this substitution. Because this substitution is less than dollar for dollar, the conclusion states that FAS 123-‐R resulted in reduced abnormal compensation for the top five executives. Hayes, Lemmon and Qiu (2012) also provide evidence that is consistent with Brown and Lee. They only look at the CEO’s compensation and find that the option-‐based pay component, pre-‐ and post-‐ mandatory expensing, decreased by 17 percentage points. Carter, Lynch and Tuna (2007) also provide evidence that the introduction of FAS 123-‐R decreased option-‐ based pay. Skantz (2012) documents that these changes in CEO compensation composition are beneficial for shareholders, because FAS 123-‐R has contributed to a reduction of inefficient CEO compensation.
The most important related research is Ladika and Sautner (2014). They find evidence that some firms accelerated their option vesting and reduced investment. They conclude that executives with more short-‐term incentives engage in myopic behavior by reducing investment. Ladika and Sautner’s paper is useful as this paper’s identification strategy is similar to theirs. They also use the different FAS 123-‐R compliance dates as an exogenous instrument to identify causality. Golden and Kohlbeck (2014) find that FAS 123-‐R increased stock repurchases overall, and that this effect is stronger with increased levels of management stock options. The last research that fits into this stream of literature but also touches upon the final stream is performed by Barth, Gow and Taylor (2012). They examine how key market participants responded to FAS 123-‐R. They find that some companies exclude stock-‐based compensation expense from non-‐GAAP earnings, despite the regulation, and that some analysts exclude it from street earnings. Barth, Gow and Taylor reason that the former is explained by opportunism and the latter by predictive ability. The findings suggest that the decades-‐old controversy around the expensing of ESOs is possibly explained by cross-‐sectional variation in the relevance of the expense for equity valuation, as well as to varying incentives of market participants.
The other stream of literature that the current paper relates to is the literature that helps to enhance our understanding on analysts’ use of public information. Two papers in the early
1990’s, Schipper (1991) and Brown (1993), already call for more research into what accounting inputs analysts actually use in their decision process. Since then a vast amount on research is done, using all sorts of research methodologies. From simply asking analysts how they process information (Block, 1999), recording analysts thinking out loud (Bouwman et al., 1995), to examining errors in forecasts as a consequence to a tax reform act (Plumlee, 2003) and laboratory experiments to study how analysts use information (Maines et al. 1997). Brown (1993) and Schipper (1991) also indicate that behavioral research can play a more prominent role when trying to grasp the way analysts use accounting and other information to make stock recommendations.
Financial statement data are an important source of information for analysts (Barker and Imam 2008, Barron et al. 2002, Schipper 1991). Although Liu, Nissim and Thomas (2007) state that operating cash flows are better at explaining valuations than accounting earnings, Barker and Imam (2008) find that the earnings stated by the company is often the most important item used by analysts. Hirshleifer and Teoh (2003) state that analysts are boundedly rational and that they do not have the cognitive capabilities nor the time to incorporate all available information in their stock forecasts. This phenomenon, called information overload, can actually cause people to make worse decisions than they would have done without the information (Casey 1980, Simnet 1996). With the huge amount of information available, it is difficult to separate the important valuable information from the side-‐issues. This information overload, we believe, could be an explanation for the fact that in the current paper analysts do not make a distinction between the treatment and control group. Not only do we believe that analysts are overloaded with information, we also believe that they sometimes are a bit inattentive, maybe even lazy. A nice example that confirms inattentiveness is Dellavigna and Pollet (2009. They show, using earnings surprises, that Friday earnings announcements have a 15% lower immediate response compared to the other days of the week. Inattentive analysts are also found in a study by Herrmann and Thomas (2005). They show that analyst forecasts of earnings per share occur in nickel intervals more often than the actual earnings per share do. These analysts that round their forecasts tend to be less informed, exert less effort and have fewer resources. Hirshleifer, Lim and Teoh (2009) show that concurrent earnings announcements distract analysts.
These studies conclude that analysts can be inattentive, lazy and time constrained, and are consistent with what we expect to find.
The timing and the style of announcements and disclosures matter for the efficiency/accuracy analysts incorporate information. Information that is difficult to extract from public data is less completely incorporated in the market prices (Grossman and Stiglitz, 1980) and Bloomfield (2002). This is consistent with the notion that hard-‐to-‐process disclosure is costlier to process and delays the incorporation of the information into stock prices or forecasts. Andersson and Hellman (2007) show that analysts who receive pro forma and GAAP information make significantly higher EPS forecasts than analysts who merely received GAAP information. Fredrickson and Miller (2004) however say that analysts are less likely to be distracted by pro forma disclosures than non-‐professionals. In 1996, Lang and Lundholm published a paper that examines the relations between disclosure practices of firms, the number of analysts following each firm and properties of the analysts’ earnings forecasts. The findings suggest that firms with more informative disclosure policies have a larger analyst following, more precise analyst earnings forecasts, less dispersion among individual analyst forecasts and less volatility in forecast revisions. The way Lang and Lundholm measure forecast accuracy could be helpful for our research. Hope (2003) also finds that firm-‐level disclosures and the enforcement of accounting standards is associated with higher analyst forecast accuracy. Lehavy, Feng, Li and Merkley (2011) find, not surprisingly, that less readable 10-‐Ks are associated with lower analyst forecast accuracy and greater dispersion among analyst forecasts. Another study that could contribute to mine is Plumlee (2003). She finds evidence that indicates that a higher information complexity reduces analysts’ use of the information. She interprets that this is due either to analyst limited processing capacity or time constraints. Similar to our research, Plumlee uses a federal regulatory change as an event to examine and makes inferences on analyst behavior. Demirakos et al. (2004) and Bradshaw (2002) find that analysts refer to simple P/E multiples to support their stock recommendations rather than extensive present value techniques. They do not conclude that this is because of a time constraint. If our research indeed finds that after FAS 123-‐R, the treatment firms are punished with adjusted analyst forecasts, while the control group does not experience such an adjustment, this would be consistent with Demirakos’ and Bradshaw’s findings.
According to the (semi-‐strong) efficient market hypothesis (EMH) (Fama, 1970) capital markets should incorporate all publicly available information into stock prices and forecasts in a quick and efficient manner. It should not matter how the information is displayed, whether it’s disclosed in a footnote or recognized in the income statement. However, prior research shows investors or analysts find recognized items more pertinent than disclosed items. Davis-‐Friday et al. (1999 find that the market treats disclosed PRB liabilities as less reliable than recognized PRB liabilities and pension liabilities. Barth, Clinch and Shibano (2003) find evidence that “recognition of a highly unreliable accounting amount, rather than simply disclosing it, can result in greater price informativeness. Likewise, recognition of a highly reliable amount can result in lower price informativeness”. They also find that the coefficients in a regression of price on accounting numbers is affected by recognition and disclosure. Choudhary (2008) finds that mandatory recognition of ESOs reflects increased dividend and interest input accuracy. Hence financial statements reflect differences in behavior between recognition and disclosure reporting regimes. Choudhary (2011) investigates reliability differences across recognition and disclosure regimes. The evidence shows that opportunism increases with recognition as compared with disclosure, but that accuracy does not decline for the recognizers. Bratten, Choudhary and Schipper (2013) find that disclosed items are not handled differently from recognized items when the “disclosures are salient, not based on management estimates, and amenable to simple techniques for imputing as-‐if recognized amounts”. Finally, Balsam, Bartov and Yin (2006) show that the market valuation of the ESO expense does not differ whether the amount is disclosed or recognized. They claim that firms need not worry about the first order effect of mandated recognition of ESO’s on their share prices. This outcome is inconsistent with what we expect to find, as it shows markets are efficient and that prices already incorporate all available information. It also assumes that firms voluntarily expensing have exactly the same characteristics as the rest of the US firms.
These papers are relevant to ours, because most of them show that investors and analysts sometimes treat disclosed information differently from recognized information, and thus that markets might be inefficient. In this paper we aim to prove the same, using a treatment group that has to recognize their ESO expenses and the control group that merely discloses them in a footnote. According to the EMH there should not be a difference in analyst
forecasts in response to FAS 123-‐R, yet we expect analysts to lower their earnings forecasts for the treatment group.
4. Hypotheses and Methodology
The aim of the current paper is to show that when constructing forecasts and recommendations, analysts look at headlines of the Profit & Loss Statement, rather than researching the company in close detail. Due to globalization, technology and high frequency trading a.o., the amount of information to be processed has risen substantially, but the timespan for analysis has not. In other words, suspected is that in light of this time constraint, analysts prefer using the easy observable P&L headlines over extensive research. Hence my main hypothesis:
H1: When constructing earnings forecasts, financial analysts depend on P&L headlines
rather than in-‐depth company analysis.
To find proof for H1, an event is needed that plausibly led to an exogenous change in the P&L headlines, but did not actually change anything in the company. Thus the analyst doing an in-‐depth company analysis would not find anything to have changed and not adjust his forecasts, but the analyst who solely studies the P&L will observe a company change and adjust his estimates. To further explain the importance of an exogenous event, consider the following linear regression model that does not use such an event.
1 ∆𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,! = 𝜃 ∗ ∆𝐸𝐵𝐼𝑇𝐷𝐴!,!+ 𝛽 ∗ 𝑥!,!!!+ 𝜇!+ 𝜀!,! 2 ∆𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,! = 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,!− 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,!!! 3 ∆𝐸𝐵𝐼𝑇𝐷𝐴!,! = 𝐸𝐵𝐼𝑇𝐷𝐴!,! − 𝐸𝐵𝐼𝑇𝐷𝐴!,!!!
In this model the dependent variable is a change in the financial analysts’ forecast for future time period t+1 for firm f at current time period t. ∆𝐸𝐵𝐼𝑇𝐷𝐴!,! is the change in EBITDA
between time period t and t-‐1. 𝑥!,!!! is a vector of variables to control for observable firm
characteristics and 𝜇! is a variable that controls for time fixed effects. We assume EBITDA to
run the regression, find a significant and positive 𝜃 and conclude that a change in P&L headlines indeed significantly affects analysts’ forecasts.
However, inferring a pure causal relationship between ∆𝐸𝐵𝐼𝑇𝐷𝐴!,! and ∆𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,! is
unlikely. A change in the company’s EBITDA is probably correlated with variables in the error term that are difficult/impossible to observe, which in turn are also correlated to the change in forecasts by the analyst. For example, imagine company A whose factory burns down. This won’t only directly impact A’s EBITDA because they produce less, it also impacts other company variables (e.g. reputation, safety/insurance expenses go up, CEO of A gets fired, legal expenses etc.). The analyst adjusts his forecasts for A, but not because of the changed EBITDA, but because of the burnt down factory and its consequences for all kinds of variables. The factory burning down is an endogenous event. It simultaneously changes a lot of (unobservable) variables in the regression model and probably leads us to biased estimates of 𝜃. Using this model we could not conclude whether the analyst changed his forecasts because he thoroughly analyzed of the consequences of the burnt down factory or only observed the EBITDA change.
The setting that will be used to obtain causal coefficients, is the 2005 introduction of the accounting standard FAS 123-‐R as an exogenous event, similar to Ladika and Sautner (2014). The Financial Accounting Standards Board mandated all firms to treat their (granted) employee stock options as an expense in the first quarter of their first full fiscal year as of June 2005. Not all firms use the calendar year as fiscal year, so there is variation in the compliance date for FAS 123-‐R. Some firms have to comply in 2005, while for others the accounting rule takes effect in 2006. Before proceeding, the following conditions will be tested:
Condition 1. The variation in firms’ fiscal year ends is sufficiently large to derive a treatment
and a control group.
Condition 2. The treatment and control group have no significant differences in observable
firm characteristics.1
1 The first condition is important because it shows the control group is large enough to avoid small sample size
We will use a Two-‐Stage Least Squares (2SLS) Model, similar to Ladika and Sautner (2014). In version A of the 2SLS, FAS 123-‐R effective is a variable that indicates whether the law is in effect or not. It is set equal to 0 in all years and months before FAS 123-‐R takes effect, and it is equal to 1 for all years and months once the law has taken effect. The date that the variable changes from 0 to 1 differs from firm to firm, as it depends on their fiscal year. Appendix A-‐3 provides an illustration of the compliance of rule FAS 123-‐R. In version B, FAS 123-‐R takes effect is a variable that indicates whether the law took effect that year or not. It is set equal to 0 in all years other than the year that FAS 123-‐R took effect, and it is equal to 1 if FAS 123-‐R took effect in that year. After this year the value again takes on 0, as the law can only take effect once. The year that the variable takes on a 1 differs between firms, because it also depends on the firm's fiscal year end.
First Stage: A.: 𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠!,! = 𝜋!∗ 𝐹𝐴𝑆123𝑅 𝐸𝑓𝑓𝑒𝑐𝑡𝑖𝑣𝑒!,! + 𝜋!∗ 𝑥!,!!!+ 𝜇!+ 𝑢!,! B.: ∆𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠!,! = 𝜋!∗ 𝐹𝐴𝑆123𝑅 𝑡𝑎𝑘𝑒𝑠 𝑒𝑓𝑓𝑒𝑐𝑡!,!+ 𝜋!∗ 𝑥!,!!!+ 𝜇!+ 𝑢!,! ∆𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠!,! = 𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠!,!− 𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠!,!!! Second Stage: A.: 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,! = 𝛾!∗ 𝐸𝑎𝑟𝑛𝚤𝑛𝑔𝑠!,!+ 𝛾!∗ 𝑥!,!!!+ 𝜇!+ 𝑣!,! B.: ∆𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,! = 𝛾!∗ ∆𝐸𝑎𝑟𝑛𝚤𝑛𝑔𝑠!,!+ 𝛾! ∗ 𝑥!,!!!+ 𝜇!+ 𝑣!,! ∆𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,! = 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,!− 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,!!!,!!!
selection bias occurs. It is expected that both conditions will not be rejected because Ladika and Sautner also test similar hypotheses and find no evidence against them. If both conditions are satisfied the variation in fiscal year ends will be used as a valid and relevant instrument in the identification strategy. Ladika and Sautner (2014) also use this identification strategy to identify the effect of mandated ESO expensing on accelerated vesting of options. Van Binsbergen, Graham and Yang (2010) use the variation in firms’ fiscal year endings to establish a causal effect of the tax reform TRA86 on firms’ marginal cost of debt. This strategy is also similar to Daske et al. (2008). They identify the effect of IFRS on liquidity by using the fact that IFRS applied to firms on different dates depending on their fiscal year end. Michels (2015) uses this identification strategy to estimate the effect of disclosure versus recognition.
Where ∆𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡!,! is a measure of change in analyst forecast for period t+1 for firm f at
time t. 𝜇! is a control variable for time fixed effects. ∆𝐸𝑎𝑟𝑛𝚤𝑛𝑔𝑠!,! is the estimated value of
Change in Earnings (either Net Income or EBITDA) from the 1st stage regression. This regression model is similar to the Ladika and Sautner (2014) model. 𝑥!,!!! is a vector of firm
characteristics at time t-‐1, they are from the previous year to assure that they are not affected by FAS 123-‐R.
The vector of firm level controls include firm size, measured by Log(Assets). As larger firms tend to provide their managers with a higher fraction of equity-‐based pay (Gabaix and Landier, 2006), the variable is expected to carry a negative coefficient. The control variable Log(Assets)² will also be included because this relationship is thought to be inverted U-‐ shaped, hence will carry a positive coefficient. Book to Market Ratio will also be controlled for, because analysts will likely punish growth firms (low P/B) more severely than value firms (high P/B) when they miss earnings growth targets (Skinner and Sloan, 2002). As firms with higher leverage ratios are more likely to face constraints from earnings-‐based covenants (Billet, King and Mauer, 2007), analysts may also treat their missed earnings-‐targets differently, so we control for Debt/Assets. Using the variables Stock Return and Volatility, we control for the firms performance and risk respectively. Firms with lower stock returns will likely have lower employee stock option expenses. So the higher the stock return, the larger the effect of FAS 123-‐R on the firms EBITDA or Net Income, thus a negative coefficient is expected. Finally, we control for the Option Grant Value, as the impact of the new accounting rule is increasing with this value we expect a negative coefficient.
Sub-‐H1: FAS 123-‐R significantly impacts the earnings of companies that belong to the
treatment group after controlling for firm characteristics 𝑥!,!!! and time fixed effects 𝜇!.
Sub hypothesis 1 is not rejected if 𝜋! in the first stage regression is significant. 𝜋! is
expected to be negative and significant because the treatment group has to expense their employee stock options. Ceteris paribus, when expenses are higher (operating) earnings will be lower. Because of condition 2 and Sub-‐H1 it may be concluded that the decreased earnings in the treatment group are because of FAS 123-‐R, and nothing else has actually changed in the company. The coefficient of interest is 𝛾!. A significant value for this
for the treatment group, although nothing has truly changed in the firm. We expect the value to be positive, because we expect earnings forecasts to be positively related to reported earnings. Correspondingly:
𝐻!: 𝛾! = 0.
𝐻!: 𝛾! > 0.
With the introduction of FAS 123-‐R in 2004 no new information on the cost of option compensation came to light, it was just easier to observe. The unrevised statement from 1995 (FAS-‐123) recommended firms to adopt fair value accounting. It also mandated the disclosure, rather than expensing, of option compensation. From then on, firms had to disclose their cost of option compensation in a footnote. With the 2005 revision of the FAS-‐ 123 statement, the contents of this footnote moved to the profit and loss statement. This increased visibility. With a significant positive 𝛾!, it seems that when information on option
expenses is easy to observe, analysts use it in their forecasting. However, when it was hidden in a footnote they didn’t bother.
5. Data and Descriptive Statistics
The panel data used for this research is obtained from four databases; Compustat, Eugene Fama and Ken French industry classifications database, I/B/E/S and CRSP. Compustat is a database that contains financial and statistical market information on active and inactive companies from all around the world. It covers 99% of the world’s total market capitalization and its first observations date back to 1962. We will only use data on U.S. based firms as FAS 123-‐R is a U.S. accounting rule. The two scholars Fama and French constructed a database that assigns each NYSE, AMEX, and NASDAQ stock to an industry portfolio at the end of June of year t based on its four-‐digit SIC code at that time. There are 48 industries, and we’ll be using them in our tests to see if FAS 123-‐R had different effects on several specific industries. We also classify several industries as tech and especially scrutinize this group of firms. The I/B/E/S database provides consensus and detail forecasts from security analysts for all kinds of metrics, though we will only be using their earnings per share forecasts as those are issued for the vast majority of firms in our sample. If we