Non-GAAP Earnings Disclosure and Analysts’ GAAP Earnings
Forecasts
Name: Yi Li
Student number:11594152 Thesis supervisor: Dennis Jullens Date:24th June, 2018
Word count:12715
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam
Statement of Originality
This document is written by student Yi Li 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
Non-GAAP disclosure by management has been a disputable topic in recent years, some papers argue that it gives a more complete picture of firms since managers voluntarily give out more private information to the outsider markets (private information hypothesis of Non-GAAP disclosure), while others suggest that Non-Non-GAAP disclosure has been used by managers as a method of earnings management because it gives managers discretion to change the measurement bases (earnings management hypothesis of Non-GAAP disclosure). Analysts’ GAAP forecast is broadly available recently from big forecast data providers, such as I/B/E/S. This newly available forecast metric is a straightforward result of analysts’ evaluation of financial performance and future earning ability of firms. This paper examines whether the availability of Non-GAAP disclosure by managers impacts the accuracy of analysts’ GAAP forecast. From the insignificant regression results, my study suggests that Non-GAAP availability does not show a significant association with the accuracy of analysts’ GAAP earnings forecast. One of the possible explanations for this is earnings management hypothesis of Non-GAAP disclosure. This hypothesis means that analysts do not trust the Non-GAAP reporting voluntarily provided by the managers, or analysts do not find Non-GAAP disclosure informative or useful to their GAAP forecast. Other possible reasons are presented in the conclusion.
Contents
1 Introduction ... 5
2 Literature Review and Hypothesis Development ... 8
2.1 Fundamental Concepts ... 8
2.1.1 GAAP Earnings ... 8
2.1.2 Non-GAAP Earnings ... 9
2.1.3 Street Earnings ... 11
2.1.4 Earnings Surprises ... 12
2.1.5 Meet-or-Beat Analysts Forecasts ... 12
2.1.6 Measurement Error in Traditional Calculation of Earnings Surprises ... 13
2.2 Analysts GAAP earnings forecasts ... 14
2.2.1 Why GAAP earnings forecast is of relevance to the market? ... 15
2.2.2 Challenges of forecasting on GAAP-based earnings ... 16
2.3 Hypothesis Development ... 18
3 Research Design and Data Sources ... 21
3.1 Data Source and Measure of Non-GAAP disclosure ... 21
3.2 Dependent Variable ... 22
3.3 Control Variables ... 23
3.4 Regression Model ... 24
3.5 Sample Period and Sample Selection ... 25
4 Empirical Results ... 27
4.1 Descriptive statistics and correlation matrix ... 27
4.2 Regression Results ... 30
5 Conclusion ... 34
1 Introduction
This paper examines whether Non-GAAP disclosure by management impacts analysts’ GAAP earnings forecast. One of the primary roles of analysts in the market is to help mitigate the information asymmetry between firms and investors (Mark and Edward, 2018), GAAP earnings forecast by analysts is a direct result of their processing of all the information disclosed. However, compared to street earnings forecasts, GAAP earnings are way more difficult to make precise forecasts due to the broader forecasting base which includes transitory items.
At the same time, Non-GAAP disclosures by management are regarded as extra private information voluntarily provided by firms which could help mitigate the information asymmetry between firms and investors (Huang and Skantz, 2016). However, Non-GAAP earnings have a tendency to be more ambiguous for normal investors to read and interpret than GAAP figures, because Non-GAAP figures are unregulated and differ from firms to firms. In addition, due to the discretion for managers to define the calculation base of Non-GAAP earnings, the quality of Non-GAAP earnings are still questionable for researchers. Previously, some scholars (Lougee and Marquardt, 2004; Bhattacharya et al. 2003) have shown that Non-GAAP earnings provide more private information to outsiders. Therewith, the information asymmetry between firms and outsiders is mitigated (private information hypothesis); However, it is also suggested by some other scholars (Lougee and Marquardt, 2004; Black and Christensen, 2009; Doyle et al, 2013) that Non-GAAP earnings are frequently used by managers as a method of earnings management, thus lowering the quality of reporting and giving more “noise” to outsiders. Therefore, I want to research the question whether Non-GAAP information from managers can help analysts, who are the more sophisticated information users and also intermediary between firms and investors, do a better job. To be more specific, my research question is written as follows:
RQ: Do Non-GAAP disclosures by managers increase the accuracy of analysts’ GAAP forecasts?
The primary objective of financial reporting is to provide useful information to the users of financial reports (Pelger 2016). However, in practice this is hard to achieve due to multiple reasons. For example, the information asymmetry between managers and investors, the complex accounting standards and the difficulty for normal investors to process the information. Therefore, in the imperfect market, the role of analysts as the intermediary between firms and investors is to address the gap between the two and to build a better functioning market by giving precise analysts forecasts. Analysts look at all the market factors and financial results including Non-GAAP information, if available, and then use these data to try to predict the
future earnings of the company and advise their clients whether to buy or sell stocks of a certain company.
Explicit GAAP earnings forecasts by analysts have been tracked by I/B/E/S since 2003. “In North America this figure is referred to as GAAP Earnings Per Share and is calculated according to Generally Accepted Accounting Principles (GAAP), which is reported in SEC filings” (Thomson Reuters 2009). GAAP earnings forecasts have been available for nearly 90% of I/B/E/S firms since 2009. This addition to the I/B/E/S forecasting dataset is valuable to investors because this figure itself and the following calculation of GAAP-based earnings surprises significantly improve the comparability and consistency, two tenets of earnings quality.
However, not surprisingly, it is more difficult for analysts to make GAAP earnings forecasts than to make Non-GAAP earnings forecasts, because of the inclusion of transitory items in GAAP basis (Bradshaw et al 2017). The first problem following making GAAP forecasts is that analysts have to distinguish transitory items from recurring items, and how to distinguish these two items is determined by the nature of related transactions while the information on these transactions is not always accessible. In addition, even if analysts are able to distinguish the transitory items from all, it is more complicated to make forecasts on these items than recurring items due to the higher inconsistency.
According to previous researches, Non-GAAP disclosure by management is probably of great help for analysts to formulate GAAP earnings forecasts and can increase the accuracy of analysts GAAP earnings forecasts, because Non-GAAP disclosures by managers provide more complete information from the internal side of the firms (we call it the private information hypothesis). More precisely, the private information hypothesis assumes that managers possess more private information that is hard to get access to by outsiders. In this context of asymmetric information, Non-GAAP disclosure, as a supplement to the mandatory GAAP reporting, conveys a bigger part of private information to analysts.
On the other hand, Non-GAAP earnings information provided by management might be useless or even noisy for analysts to forecast future GAAP earnings because managers have incentives to hide bad news and exaggerate good news in their Non-GAAP numbers (we call it earnings management hypothesis). In other words, earnings management hypothesis argues that managers will provide unreliable Non-GAAP numbers to outsiders because they have conflicts of interests with investors and managers are motivated by management opportunism. Hence, analysts do not revise their expectations based on these voluntary and less regulated information disclosed by managers because they assume these figures are noisy or they are simply useless due to the low reliability of the Non-GAAP numbers.
To test hypotheses of my study, I use forecasts dispersion (!"#,%) and forecasts error (!&#,%) as two proxies of accuracy of analysts’ GAAP forecast, and I regress analysts’ GAAP forecasts dispersion on the control variables and independent variable ('(#,%) which indicates whether managers disclose or not Non-GAAP information in that year. As a result, both of the regressions do not show significant relations between analysts’ forecast measures and Non-GAAP disclosure by managers. These two insignificant results confirm the earnings management hypothesis of Non-GAAP disclosure, which means managers use their discretion and provide unreliable Non-GAAP information, and analysts do not trust this and therefore do not change their forecasts, regardless whether there are Non-GAAP disclosures available. Other explanations are also possible for the insignificant results, for example, the limitation of the data source and the endogenous problem deriving from the complex data transforming.
My study contributes in several aspects. First, most prior literature makes use of ‘street’ earnings as the replacement of managers Non-GAAP earnings when examining the informativeness of Non-GAAP earnings compared to GAAP earnings. I point out the differences between these two measurement bases and the problems of using ‘street’ earnings as the replacement of managers’ Non-GAAP earnings. Secondly, though Non-GAAP earnings forecast is already a popular topic in extant literature, not a lot researchers are looking into GAAP earnings forecast, while GAAP forecast figures are already commonly available from forecast data providers. Most researches use Non-GAAP forecasts when talking about analysts’ forecasts, while analysts’ forecasts can mean Non-GAAP based or GAAP based forecasts. There is only Bently et al (2017) specifying GAAP forecast. I illustrate the importance of GAAP earnings forecasts to the market and this is a quite new topic which deserves further researches. At last, it’s relevant for standard-setters to know more impacts of Non-GAAP disclosure, which has been a disputable topic recently.
This thesis proceeds as follows. In section 2, relevant literature will be reviewed and hypotheses for my research question are developed. Section 3 describes the data and research design. Section 4 presents the main empirical results. Section 5 concludes the article.
2 Literature Review and Hypothesis Development
The thesis mainly relates to two streams of previous literature. It first relates to the literature on analysts GAAP earnings forecast. Secondly it relates to the literature on Non-GAAP earnings disclosure. Both of these two topics are quite new and controversial in accounting field. In addition, both of these two are built on series fundamental concepts. Therefore, before going to the hypotheses development, I will first introduce these fundamental concepts and the related two accounting topics.
Fundamental Concepts
2.1.1 GAAP Earnings2.1.2 Non-GAAP Earnings 2.1.3 Street Earnings
2.1.4 Meet-or-Beat Analysts Forecasts 2.1.5 Earnings surprises
2.1.6 Measurement Error in Traditional Calculation of Earnings Surprises Figure 1.
2.1 Fundamental Concepts
2.1.1 GAAP Earnings
GAAP, or Generally Accepted Accounting Principles, is considered the “gold standard” of financial reporting. In its standards that constitute GAAP, the Financial Accounting Standards Board (FASB) requires rule-based GAAP performance measures such as net income, earnings per share, and operating cash flows.
GAAP earning is a corporate profit performance measure that follows US GAAP rules and includes all earnings items from income statement, regardless of how those profits (or losses) were generated. Because of the rule-based or standardized nature, GAAP earning allow investors to directly compare current earnings of a firm to (1) its own historical earnings, which is useful in measuring growth in firm performance, and (2) the contemporaneous earnings of other firms, which is useful in assessing a firm’s relative performance. This significantly enhances consistency and comparability of financial reporting.
However, the rule-based or standardized nature also means GAAP earning includes transitory asset write-downs and one-time charges such as restructurings, divestitures, and acquisitions. These items are of course important for investors to consider, but the inclusion of these nonrecurring items into the base of GAAP earning makes GAAP forecasting more difficult. In addition, a big criticism of GAAP earning is that the relevance of this metric is controversial to outsiders, since it contains all nonrecurring items which make the current earning numbers not predictive and informative about future earning ability.
U.S. law requires businesses that release financial statements to the public and companies that are publicly traded on stock markets to follow GAAP guidelines. While FASB strives to alleviate incidents of inaccurate and uninformative reporting, GAAP earning is by no means comprehensive, there is no way to modify it to become one metric suitable for all conditions and for all different users. Therefore, to compensate for the drawbacks of standardized GAAP earning, Non-GAAP earning appears as an alternative reporting way and is becoming more and more popular both practically and academically.
2.1.2 Non-GAAP Earnings
Non-GAAP earning is an alternative measure of performance to GAAP earning, and as explained above, Non-GAAP reporting by managers has reasonably become a popular topic nowadays for both researcher and practitioners. From a practical perspective, it has been used very broadly. Among S&P 500 firms, the proportion of firms which disclose Non-GAAP information has substantially increased, from 53% in 2009 to 71% in 2014, indicating that firms’ use of non-GAAP earnings is now commonplace (Black et al. 2017).
These Non-GAAP measures provided with discretion by firms (common ones include adjusted EBITDA, operating earnings, and free cash flow) are often based upon information contained in GAAP financial statements. The companies believe that presenting both GAAP and Non-GAAP data creates a more complete picture of its past performance and would be more predictive to future earnings results.
Discretion in Non-GAAP reporting allows managers and analysts to exclude earnings components required under GAAP (non-recurring items and other items). Researchers have expressed significant interests in examining these exclusions and the managerial motives in deciding exclusion bases. Although a great amount studies find evidence consistent with Non-GAAP reporting being primarily motivated by providing a more informative earnings metric (e.g., Gu and Chen 2004; Curtis, McVay, and Whipple 2014), most conclude that Non-GAAP reporting is at least partially motivated by management opportunism, such as benchmark beating (e.g., Doyle, Lundholm, and Soliman 2003; Black and Christensen 2009; Frankel et al. 2011; Brown, Christensen, Elliott, and Mergenthaler 2012a; Barth, Gow, and Taylor 2012; Doyle et al. 2013, Christensen, Drake, and Thornock, 2014).
To highlight the key of academic discussions around Non-GAAP measure, most of them stem from one single question: do managers disclose Non-GAAP to provide more private information to capital markets or to opportunistically show a better performance to fool investors? Regarding this question, we find two general hypotheses in the academic literature:
(1) Private Information Hypothesis;
For the first private information hypothesis, the market is not perfect, private information is defined as the information only known to the managers of the firms and unknown to investors, while public information is the information already available in the market to investors. Ball (2013) recalls that there are ‘many competing information sources available to investors, and many are timelier than periodic financial reporting’. Private information hypothesis argues that Non-GAAP disclosure by management is one of the additional information sources to periodic GAAP reporting. Managers provide timelier, more complete and more predictive information to outsiders by providing Non-GAAP earnings as an addition, rather than only provide mandatory GAAP information.
Consistent with (1) Private information hypothesis, Doyle et al. (2003) find that the increasingly popular Non-GAAP measure of earnings excludes certain expenses that the companies deem non-recurring, non-cash, or otherwise unimportant for understanding the future value of the firm. Curtis et al. (2014) find that Non-GAAP earnings are more informative than GAAP earnings and managers motivated to inform stakeholders about sustainable earning ability will disclose Non-GAAP earnings information excluding the gain. In addition, some studies find evidence that investors have a preference for Non-GAAP earnings (Heflin and Hsu, 2008; Doyle et al., 2013) and Non-GAAP reporting is primarily motivated by providing a more informative earnings metric (e.g., Gu and Chen 2004; Curtis, McVay, and Whipple 2014).
For the second hypothesis, managers always have some incentives to opportunistically provide earnings to outsiders according to agency theory (Healy and Wahlen, 1999; Fields, Lys, and Vincent, 2001; Watts and Zimmerman, 1986), and Non-GAAP earnings is obviously a perfect tool because they are allowed to have larger discretion in this case. Though managers frequently argue that they exclude transitory items from GAAP income in an effort to better reflect core operating performances of the firms, regulators and critics of pro forma reporting contend that these exclusions can overstate operating results.
Consistent with this hypothesis, extant literature provides evidence that some managers exclude some items other than transitory items for opportunistic reasons (Black and Christensen 2009; Frankel, McVay, and Soliman 2011; Barth, Gow, and Taylor 2012). To be more specific, Barth et al. (2012) find that opportunism is the primary explanation for exclusion of the expense from pro forma earnings. In addition, the Securities and Exchange Commission (SEC) issued a warning that under certain circumstances, pro forma reporting “can mislead investors if it obscures GAAP results” (SEC 2001) and the SEC’s chief accountant of the Enforcement Division, Howard Scheck, emphasizes that Non-GAAP metric is a “fraud risk factor” (Leone 2010). Furthermore, most researches also conclude that Non-GAAP reporting is at least partially motivated by managerial opportunism, such as benchmark beating (e.g., Doyle, Lundholm, and Soliman 2003; Black and Christensen 2009; Frankel et al. 2011; Brown,
Christensen, Elliott, and Mergenthaler 2012a; Barth, Gow, and Taylor 2012; Doyle et al. 2013, Christensen, Drake, and Thornock, 2014).
Though different people have different opinions regarding Non-GAAP measures, it’s undeniable that Non-GAAP earnings have become more and more popular for mangers to disclose in addition to GAAP earnings.
2.1.3 Street Earnings
‘Street Earnings’ are calculated by analysts, and analysts have discretion to value a firm based on the measurement base which according to their judgment best depicts the performance and financial position of the firm. ‘Street’ base decided by analysts is normally different from GAAP base and the Non-GAAP base decided by managers. Analysts forecast data providers such as I/B/E/S publish ‘Street Earnings’ regularly. And for comparison’s sake, I/B/E/S does not only provide analysts’ ex ante earnings forecasts but also ex post earnings realizations calculated on the same base as the forecasts.
One of the most commonly used ‘street earnings’ metric is “earnings per share” (EPS), Thomson Reuters 2009 defines “street EPS’ as the metric that “the contributing analyst considers to be that with which to value a security. This figure may include or exclude certain items depending on the contributing analyst’s specific model”. Similarly, the base of ‘Street EPS’ metric is also in analysts’ discretion, it can be on a GAAP or Non-GAAP basis, depending on what the majority of analyst think is the best way to value a firm.
Therefore, the base of street earnings can vary across firms and over time, raising concerns about the comparability of the earnings metrics. Lambert (2004, p. 212) affirms in his paper that “if the definitions of earnings analysts are forecasting is not the GAAP number, then researchers have to be careful about their interpretation of forecast surprises and forecast ‘errors’”. This statement highlights the challenges underlying the use of ‘street earnings’ in traditional researches on forecast surprises and errors.
Challenges regarding low comparability of street earnings can be categorized into two aspects. First, the definition of street earning has shifted in recent years (Bentley et al. 2017), making it difficult for researchers to compare the same earnings metric across firms and over time.
A second concern is that analyses of GAAP-based earnings surprises are problematic without explicit GAAP forecasts. To investors, earnings surprises are regarded as an important metric to consider when making investment decision; To researchers who do research on Non-GAAP topic, they are interested in measuring Non-GAAP-based earnings surprises to examine (1) the informativeness of GAAP earnings relative to Non-GAAP earnings and (2) whether firms
report Non-GAAP earnings when their GAAP earnings miss analysts’ forecasts. However, prior studies did not have a GAAP-based expectation when I/B/E/S EPS was on a Non-GAAP base, subjecting GAAP earnings surprises to significant measurement error. In the following parts we will introduce more on the concepts: earnings surprises and measurement errors. 2.1.4 Earnings Surprises
An earnings surprise, or unexpected earnings, in accounting, is the difference between the reported earnings and the expected earnings of an entity.
Earnings Surprises=the actual earnings-the expected earnings forecast by analysts Measures of a firm's expected earnings include analysts' forecasts of the firm's profit and mathematical models of expected earnings based on the earnings of previous accounting periods. Analysts’ forecasts are used in many different streams of accounting research, and one of the most common uses of these forecasts is to calculate a firm’s earnings surprise. Earnings surprises, serving as a new information of earning ability in the past period, are received by investors at the earnings announcement.
Previous literature has found that earnings surprises have significant impacts on stock markets (e.g. Rai and Tartaroglu 2015; Keung et al 2010)—positively to positive earnings surprises and negatively to negative earnings surprises—although a proportion of earnings surprises result in stock markets reacting not in the expected direction, which may be a reaction to other relevant information released with the earnings announcement or inaccurate measurement of the earnings surprise.
Besides impacts on stock market, large negative earnings surprises may also have legal and reputational costs to managers. Firstly, managers can be held liable personally when shareholders sue the firm for failing to disclose negative earnings news timely. Secondly, investment managers may choose not to hold, and analysts may choose not to follow, the stocks of the firms in which managers have reputations for withholding bad news. This may to some extent forces managers to voluntarily disclose information related to negative earnings surprises: quarterly earnings announcements containing large negative earnings surprises are preempted by voluntary disclosures more frequently than are other earnings announcements. 2.1.5 Meet-or-Beat Analysts Forecasts
As mentioned above, earnings surprises have such big influences on reporting firms, so managers have great incentives to meet analysts’ earnings forecasts. This has been examined by a large stream of research. Bartov, Givoly, and Hayn (2002) find that firms can receive a stock price premium for meeting or beating analysts’ forecasts. Similarly, Brown and Caylor
(2005) point out that meeting analysts’ forecasts is an important benchmark for managers themselves, and managers would take actions to meet the benchmark for personal benefits.
The literature categories different actions managers take to meet or beat analysts’ forecasts into two big scopes, which are earnings management and expectations management (e.g., Bartov et al 2002; Matsumoto 2002; Richardson, Teoh, and Wysocki 2004). As one way of earnings management, prior researches present that when GAAP earnings fall short, managers use their discretion for Non-GAAP earnings to appear to meet analysts’ forecasts. For instance, when GAAP earnings miss analysts’ forecasts and managers exhaust other forms of earnings and expectations management, managers can modify the base of Non-GAAP earning to make it meet or exceed analysts’ forecasts (Black et al., 2017). Consistent with this, Bhattacharya et al. (2003) find that while 80% of firms (Meet-or-Beat firms) report Non-GAAP earnings that exceed analysts’ street forecasts, only 39% of them report GAAP earnings that also exceed the same forecast.
The possibility for managers doing earnings management lies in the measurement error underneath earnings surprises calculation. More detailed introduction to measurement error underlying in traditional calculation of earnings surprises is presented as follows.
2.1.6 Measurement Error in Traditional Calculation of Earnings Surprises
GAAP earnings forecasts were not available historically, so traditionally earnings surprises were computed as the difference between actual GAAP earnings and analysts’ street forecast figures, while Non-GAAP surprises equal street earnings less street forecast figures (i.e., an apples-to-apples comparison). Therefore, GAAP earnings surprises calculation actually contains measurement errors underneath, as the measurement bases used in calculation do not always match. More detailed explanations are presented as follow. We can see clearly the problems from the following listed three equations:
&)*%+,,% = Street Earnings – Street Consensus Forecast (1) &)?@@ABCD#*E = GAAP Earnings – Street Consensus Forecast (2) &)?@@ABIJ,KL = GAAP Earnings – GAAP Consensus Forecast (3)
Equation (1) is how to calculate street (Non-GAAP) earnings surprises, and equation (3) is the “clean” way to calculate GAAP earnings surprises, which demands GAAP consensus forecasts. However, as has explained earlier, GAAP consensus forecasts were not available before, so equation (2) was used as the replacement to (3), which deployed two misaligned measurement bases and made results noisy.
This disproportionate measurement error under GAAP earnings surprises has caused considerable negative impacts on previous researches, and Non-GAAP earnings literature extensively recognizes the problems associated with the traditional unavailability of GAAP forecasts. First, traditional way to calculate GAAP earnings surprises is subject to measurement error due to misaligned forecasts and realizations, leading to misclassification of meet-or-beat firms by researchers.
Second, numerous previous studies examine the informativeness of Non-GAAP earnings by disentangling whether investors actually respond more to Non-GAAP earnings or GAAP earnings. In discussing the method deployed in extant literature, Bradshaw (2003, p. 330-333) presents the flaws resulting from using the Non-GAAP earnings expectations to calculate GAAP surprises. For example, conclusions made from “the conventional analysis of GAAP and Non-GAAP ERCs” need to be taken with caution, because underlying measurement errors “may well induce” the validity of prior results. While the Non-GAAP earnings surprises are free from this form of measurement error, as they compare both street bases of expectations and realizations (i.e., an apples-to-apples comparison), GAAP earnings surprises are plausible in nature. This flaw gives an alternative explanation for the prior results that investors respond less to GAAP earnings relative to Non-GAAP earnings. Therefore, future researches on this topic should figure out certain solutions to address this measurement flaw. Towards this, Bradshaw (2003) argues researchers should have a unique expectation corresponding to each earnings measure.
In addition, same argument was also brought up by Cohen et al. (2007), they use simulations to explore the magnitude of possible measurement error in prior researches, and question the validity of prior research, concluding that “results from conventional analyses of GAAP and [non-GAAP] ERCs...are significantly contaminated by measurement errors in earnings surprises”. Similarly, Beyer, Cohen, Lys, and Walther (2010, p. 323) highlight the same underlying measurement error and conclude that, “the debate over whether investors are better informed by Non-GAAP figures or pro forma earnings numbers is still not settled.”
However, this form of measurement error related with GAAP surprises calculation can be avoided by deploying GAAP earnings actuals and forecasts instead of GAAP actuals and street forecasts, which refers to the method of equation (3), the “clean” way to calculate GAAP earnings surprises. This method makes analysts GAAP earnings forecasts become necessary, in the following chapter, we will discuss more details about the analysts GAAP forecasts. 2.2 Analysts GAAP earnings forecasts
Since 2003, I/B/E/S first began tracking GAAP earnings forecasts for some companies. The availability of GAAP forecast has outpaced that of all the other metrics of forecasts, till
now GAAP earnings forecasts are available for nearly all firms that have street earnings forecasts (more than 90%). Besides, the number of analysts contributing bottom-line GAAP earnings forecasts in addition to the common street earnings has grown considerably, with around 6 analysts following per company recently.
2.2.1 Why GAAP earnings forecast is of relevance to the market?
From theoretical perspective, GAAP-based earnings forecasts can effectively improve the relevance, credibility, reliability, comparability and also understandability of different measures under current financial reporting system. GAAP forecast, as a direct result of analysts’ interpretation of all complex information in the market, is helpful to build the trust between the market and firms.
First of all, the earnings metric in GAAP basis or accounting standard basis is the mainstream to measure how successful a company is, and earnings in standardized base give us a benchmark to compare the profitability between companies and from different time periods. Though recently we hear a lot voice promoting Non-GAAP based earnings, it is still premature to conclude that investors recognize higher value of street earnings relative to GAAP earnings (Abarbanell and Lehavy, 2007).
Secondly, Non-GAAP earnings give managers a big room to decide and change the earnings calculation base, and different firms which are similar economically can have very dispersed results of profitability. Therefore, credibility of Non-GAAP based earnings is questionable. Although some studies (e.g., Gu and Chen, 2004; Curtis et al., 2014) argue that Non-GAAP reporting by managers is primarily motivated by incentives of helping investors make better decisions, in the way of providing them extra information. Also, some studies (Heflin and Hsu, 2008; Doyle et al., 2013) find that investors do have a preference for Non-GAAP earnings. Most researchers conclude that Non-Non-GAAP reporting is at least partially motivated by management opportunism, such as benchmark beating (e.g., Doyle et al., 2003; Black and Christensen, 2009; Frankel et al., 2011; Brown et al., 2012a; Barth et al., 2012; Doyle et al., 2013; Christensen et al., 2014). In this case, GAAP basis is relatively more regulated and leaves less room for management opportunism compared to Non-GAAP.
In addition, some new researches also confirm that GAAP forecasts increase the credibility of earnings surprises (Bradshaw et al. 2018). GAAP forecasts as an addition to I/B/E/S allow researchers to calculate clean GAAP earnings surprises (equation 3) and mitigate the challenges associated with using street surprises instead (equation 2). By deploying GAAP forecast number to calculate GAAP surprises, measurement error underneath the traditional way can be addressed, which has already been discussed thoroughly in previous section. Under the new method, meet-or-beat firm classification would be more valid to help the market
participants make investment decisions, and clean GAAP earnings surprises give a more solid foundation for those researches investigating how the market interprets GAAP-based earnings. Furthermore, GAAP forecasts provide disaggregated information which increases the credibility of earnings reporting (Hirst, Koonce, and Venkataraman 2007; Merkley, Bamber, and Christensen 2013). For GAAP base, researchers are required to forecast some earnings components that have been traditionally excluded from street forecasts (e.g., transitory items, stock compensation). Bradshaw et al (2017) find that investors view street earnings to be more credible when GAAP forecasts of the same firms are also available at the same time, because GAAP forecasts provide disaggregated information (Hirst et al., 2007; Merkley et al., 2013). The difference between GAAP and Non-GAAP forecasts actually gives investors one extra information, which is analysts’ forecast of street earnings exclusions. Exclusions forecasts can at least be useful to enhance the market in two ways. First, they set ex ante Non-GAAP exclusions for managers and analysts, which constrains opportunism of both managers and analysts to change the base of Non-GAAP earnings at the end of fiscal period. Second, disaggregated earnings provided by exclusions forecast depict a clearer scope of a firm’s core earnings by highlighting items which to analysts are less persistent (i.e., the exclusions).
At last, to investors it is also easier to understand the GAAP earnings numbers because of the higher comparability and verifiability. While for Non-GAAP basis, both managers and analysts following have discretion to exclude nonrecurring items, thus increasing the processing costs of earnings reporting to the market.
Though the relevance of GAAP forecast to the market is high, there are limited guidance on how to make GAAP-based earnings forecast and limited researches on how to improve the quality of GAAP forecasts.
2.2.2 Challenges of forecasting on GAAP-based earnings
Analysts GAAP earnings forecasts are of high value to the market and deserves more attention from standard-setters and researchers while a big criticism of GAAP forecast is that errors in GAAP earnings forecast appear much more frequently than in street forecast. The larger difficulty of providing high quality GAAP-based earnings forecast than street forecast lies mainly in how to make precise forecast on exclusions, which are the differences between GAAP-based earnings and Non-GAAP-based (or street) earnings.
The first problem regarding exclusions forecast is the difficulty in distinguishing these exclusion items. Unfortunately, there is limited empirical evidence about the different components between Non-GAAP and GAAP earnings (Lambert 2004). Common assumption in extant Non-GAAP studies is that for Non-GAAP earnings forecast, analysts only forecast firm’s recurring earnings (e.g. Bhattacharya et al. 2003; Christensen 2007; Frankel et al. 2011).
However, recently Brown et al. (2015) and Heltzer et al. (2014) survey analysts and find that some analysts exclude some items relating with recurring transactions from their Non-GAAP earnings forecasts. This finding is also consistent with that analysts might exclude not purely transitory items when making Non-GAAP earnings forecast to influence the earnings surprises, thus currying favor with managers or making their Non-GAAP forecasts appear more accurate ex post (e.g. Gu and Chen. 2004; Bank et al. 2009; Barth et al. 2012; Brown et al. 2015).
According to all those extant literature, exclusions between Non-GAAP and GAAP earnings when analysts make forecast can be generally partitioned into two categories: (1) Non-recurring items (e.g. litigation charge) represent one-time costs to the company that analysts exclude when forecasting the Non-GAAP earnings; (2) Other items (e.g. amortization expense and investment gains/losses) represent the remaining Non-GAAP adjustments after considering non-recurring items, and these other items generally relate to recurring transactions.
As shown in Figure 2, the first line is the categorization of all items in GAAP earnings. In the second line, Exclusions 1 presents the ideal situation, which excludes all non-recurring items from Non-GAAP Earnings Base 1, leaving complete recurring items in GAAP earnings composing the base of Non-GAAP earnings. However, in real world there always exist two variations of the exclusions bases, which we can see from the third and forth lines. Exclusions 2 contains only partially of non-recurring items, leaving Non-GAAP Earnings Base 2 containing not only recurring but also some non-recurring items. While Exclusions 3 covers
Recurring items Non-recurring items
GAAP Earnings Base
Recurring items
Non-recurring items
Non-GAAP Earnings Base 1. Exclusions 1.
Recurring items
Non-GAAP Earnings Base 2. Exclusions 2.
Figure 2. GAAP and Non-GAAP Earnings Bases
Non-GAAP Earnings Base 3. Exclusions 3.
not only non-recurring but also a part of recurring items, in this case Non-GAAP Earnings Bases 3 include incomplete recurring items.
Therefore, the first big challenge of analysts GAAP earnings forecasting is how to distinguish non-recurring items precisely from recurring items, and standardize a same exclusions base for all analysts following the same company. Because if every analyst following the same company can have a clearer and thus more similar classification of recurring and transitory items, the forecast results made by different analysts on both recurring and non-recurring parts would be closer, leading to the aggregated GAAP forecasts also closer.
The other big challenge of GAAP forecasts is that the forecasts on transitory exclusions from GAAP earnings to Non-GAAP earnings are inherently more difficult for analysts (Whipple, 2015). If more reliable information about those transitory transactions is disclosed from the inside of companies, the accuracy of exclusions forecast by analysts should be higher theoretically, thus the accuracy of GAAP earnings forecast would be improved.
Besides the two main challenges faced by analysts GAAP forecasts, it’s surprising that some researches (Whipple 2015) find even for recurring part, analysts are also not very accurate in their forecasts on these items which relate to transactions that happen every quarter. One explanation Whipple brought for analysts’ inability to accurately forecast these items is that certain other items are volatile in nature even though they recur every quarter (e.g. investment gain/loss exclusions).
Consequently, it is much harder for analysts to accurately forecast GAAP-based earnings than Non-GAAP earnings. As the value of forecasting GAAP earnings is high, it worth further researches on how to improve the accuracy of analysts GAAP earnings forecasts.
2.3 Hypothesis Development
Relied on the literature discussed above, we want to develop the hypotheses to test. Two specific hypotheses (see Figure 3) are generally developed to test our research question.
H1 Non-GAAP earning disclosures decrease the dispersion of analysts GAAP earnings forecasts.
H2 Non-GAAP earning disclosures decrease the errors of analysts GAAP earnings forecasts.
Figure 3. Hypotheses
Non-GAAP earnings disclosures by management can be viewed as a supplement of relevant information to the mandatory GAAP financial reporting, with more details disclosed about the transitory transactions occurred in this financial period. One likely effect of increased
information from management Non-GAAP earning disclosures is that analysts following can have a clearer picture of the company’s performance in this period and thus make a more precise prediction on the future profitability of this company. In normal circumstances, one would expect greater transparency increased by insider managers to be associated with fewer absolute forecast errors and smaller dispersion in the forecasts. Therefore, holding other factors constant, Non-GAAP disclosures from managers should be positively related to lower levels of analysts’ absolute forecast errors and forecast dispersion.
We acknowledge the contradictory view that Non-GAAP reporting is used more as a method of earnings management, and the information disclosed by managers is in lack of reliability and possibly cause analysts to make wrong judgments. Following the earnings management hypothesis in one stream of Non-GAAP reporting literature, managers use their discretion to change the exclusions bases and selectively disclose incomplete information or release unreliable information about these transactions. This leads to greater difficulty for analysts following to first distinguish nonrecurring items from recurring items and second to make precise forecasts on GAAP earnings. For example, some managers exclude investment loss from Non-GAAP earnings, which actually relates to recurring transactions in their companies. These exclusions from Non-GAAP reporting may cause different analysts to have different conclusions on the nature of the investments the companies have. Some would lower their forecasts on future investment loss because they assume this part of loss won’t occur any more because managers classify them to nonrecurring items, but some may still keep the loss for future periods based on other sources of information or their own experience and knowledge. Therefore, the forecast dispersion is greater and the chance of forecast error is also greater.
Despite the possibility of the above opposing view, we believe the more likely effect of Non-GAAP disclosure from management is decreasing the forecast dispersion and forecast errors, because apparently Non-GAAP disclosure in addition to mandatory financial reporting provides more complete information to outsiders. Under the private information hypothesis in extant Non-GAAP literature, managers voluntarily release Non-GAAP earnings as a supplement to GAAP earnings, with disclosure on related transactions. At the same time, the mandatory GAAP earnings will still be presented in the same way and still under the standardized scheme. Hence, there is a more complete picture about the company depicted by increased information available to outsiders.
As the object of my research is analysts, who are more sophisticated information users and mostly follow the same company for long periods. Bently et al (2017) find the analysts GAAP earnings bases mainly originate with managers’ and that analysts echo these Non-GAAP metrics when assessing performance. Analysts, however, do not always copy managers’ Non-GAAP items, but instead monitor numbers of Non-GAAP earnings and exclude lower
quality Non-GAAP items. This finding is consistent with our argument above that analysts are capable to make good judgments on the information disclosed by managers. Therefore, the quality of analysts GAAP forecasts would be improved by more completed information provided by managers, even though a part of the information is questionable.
Analysts can make GAAP forecasts when they only have access to normal GAAP financial reporting and other public information, but it is highly possible that different analysts have different classifications on recurring and nonrecurring transactions, thus the exclusions base they forecast on are different. If Non-GAAP disclosure by management in addition to GAAP reporting is available, analysts would have much more direct information on the questionable transactions when deciding the recurring or nonrecurring nature of them. This would help different analysts to confirm their judgments when classifying nonrecurring or recurring items and to achieve a consensus on the exclusions base. In addition, more details on transitory transactions make the expected numbers of these items in forecast also closer to the actual numbers. Therefore, the forecast dispersion can be lowered since analysts’ forecasts on the exclusions converge, while forecasts on the recurring street items have been always viewed as the easier part which always has closer results to actuals.
In summary, we believe our original argument, positing a negative relationship between the availability of Non-GAAP disclosure and analysts forecast dispersion, will prevail empirically. By specifying the null hypothesis as the no-relationship case, we can present our first hypothesis as: Non-GAAP earning disclosures decrease the dispersion of analysts GAAP earnings forecasts.
Along similar lines, we argue that Non-GAAP disclosures decrease the GAAP forecasts errors for the following reasons. Assume that only GAAP reporting is available, the possibility for an experienced analyst to make wrong forecasts on recurring items is 5%, and the possibility for that on nonrecurring items is 30%. Among all mistakes, there must be some errors are caused by wrong classification of transactions between recurring and nonrecurring. If now Non-GAAP disclosure is available, analysts can have a direct help when they categorize recurring and nonrecurring transactions. Therewith, at least errors due to wrong classifications can be lowered, and the two possibilities of forecast errors would be both lowered by mitigating classification mistakes. This leads to the second hypothesis as: Non-GAAP earning disclosures decrease the errors of analysts GAAP earnings forecasts.
3 Research Design and Data Sources
3.1 Data Source and Measure of Non-GAAP disclosure
The database for my independent variable '(#,%, whether managers disclose Non-GAAP earnings, is just published in 2018. It is provided by Jeremiah Bentley, Ted Christensen, Kurt Gee, and Benjamin Whipple, authors of Disentangling Managers’ and Analysts’ Non-GAAP Reporting. Journal of Accounting Research, Forthcoming.
They create and make publicly available a large dataset of quarterly-disclosed Non-GAAP earnings by management. The original information they use is from all quarterly earnings announcements filed in 8-Ks. The variables provided by their dataset include a dummy variable that determines whether managers explicitly disclose a Non-GAAP EPS metric quarterly, and also a variable specifying the figures of managers’ Non-GAAP earnings from the earnings announcements. In total their dataset contains 115,370 firm-quarter observations of 6,335 firms spanning fiscal years 2003-2012.
Their data sample begins with all consolidated firms in the CRSP, Compustat, and I/B/E/S universe with fiscal years ending between 2003 and 2012. They exclude: (1) real estate investment trusts (i.e., REITs) because these companies’ Non-GAAP metrics are customary to become a standardized, industry-specific (i.e., Funds From Operations), (2) firm-quarters that report extraordinary items, and (3) firm-quarters for which they cannot identify an 8-K earnings announcement. Besides, they dismiss firm-quarters for which manager and I/B/E/S exclusions are missing.
To test our hypotheses and to see whether Non-GAAP disclosure would impact analysts GAAP forecasts accuracy, we define the dummy variable '(#,% which takes the value 1 when managers from the firm disclose Non-GAAP information and takes the value 0 when the managers do not do so. Since the direct variable provided in the dataset from Bently et al (2017) records quarterly availability of Non-GAAP earnings from managers, while we need the variable which represents yearly availability of managers Non-GAAP disclosure. Therefore, we define our independent variable as:
'(#,% = 0 If managers of firm N do not disclose once Non-GAAP earnings in year O;
'(#,% = 1 If managers of firm N disclose at least once Non-GAAP earnings in year O;
3.2 Dependent Variable
Measuring the accuracy of analysts GAAP earnings forecasts is quite clear and all variables are from public datasets. For analysts’ forecast data, we obtain from I/B/E/S. For financial data, we search required information from Compustat.
According to extant literature, I use two proxies to measure analyst forecast accuracy. First, analyst forecast dispersion on GAAP earnings (!"#,%) is used, which is defined as the logarithm of the standard deviation of analysts’ estimates for firm i in year t divided by the stock price at the beginning of the year for firm i in year t:
!"#,% = 1
' (!R,#,%− !#,%)U
V.X
/Z#,% where:
!"#,% =analysts forecast dispersion of firm i, in year t divided by the stock price at the beginning of the year t;
!R,#,% =forecast by analyst j for firm i in year t (with !#,% as the mean);
Z#,% =stock price at the beginning of the year for firm i in year t.
Similar with other researches, (e.g.Barniv 2009; Hope 2003; Hope and Kang 2005; Lang and Lundholm 1996), my study also uses analyst forecast error proxy (!&#,%) (Dhaliwal et al 2012) as an inverse for forecast accuracy. It is defined as the average of the absolute errors of all forecasts made in the year t for target earnings of firm i, scaled by the stock price at the beginning of the year t:
!&#,% = 1
' !R,#,%− &Z)#,% /Z#,% where:
!&#,%[ =analyst forecast error of firm i, in year t; !R,#,% = forecast by analyst j for firm i in year t; &Z)#,% =actual earnings per share for firm i in year t;
3.3 Control Variables
The literature on analysts’ absolute forecast errors has established several factors as important determinants of cross-sectional differences in analysts’ forecast accuracy and forecast dispersion. Five control variables were included in this study to control for factors that are known (from the prior literature) to be associated with cross-sectional differences in analysts’ absolute forecast errors and forecast dispersion.
These factors are (1) age of the forecast, (2) the number of analysts following a firm, (3) the market capitalization of the firm, (4) the volatility of earnings, and at last, (5) firm risk.
(1) The average age of the forecasts included in the analysis (\]^#,%). According to Sinha et al. (1997), forecasts issued earlier are likely to be less accurate than forecasts issued closer to the time earnings are announced. The reason for this is that the analysts have less information available and thus higher uncertainty about a company’s results in a fiscal year. Accordingly, this relationship is also documented by several other studies (e.g., Brown, Richardson, and Schwager 1987; Lys and Soo 1995; Das and Saudaragan 1998; Jacob, Lys, and Neale 1999; Duru and Reeb 2002). Therefore, I expect a positive sign for this variable in the regression on forecast dispersion and forecast error.
(2) The number of analysts following the firm during the quarter ('#,%) stands for the number of analysts releasing an earnings forecast for the certain company in a quarter. Lys and Soo (1995) point out that a greater analyst following means more intense competition among fellow analysts, and therefore greater incentive for analysts to put more effort in and enhance their forecast quality (Dhaliwal et al. 2012). Hence, the number of analysts who provide a forecast as a control variable is included as a control variable in regression models. A positive relation between the number of analysts following and forecast accuracy is expected according to previous literature. Therefore, '#,% should have negative relations with forecasts errors and also forecasts dispersion.
(3) Firm size, which is measured by the lagged market capitalization of the firm (_`#,%Ba). Firm size in the literature can be a proxy for several factors. On the one hand that size reflects information availability about a firm, as large companies are more likely to provide additional information or more private information (Jaggi and Jain 1998) to the public than smaller companies. In addition, more regulations on information quality are put into effect for bigger firms rather than for small firms. Therewith, bigger firms should have a more complete and reliable information environment or a greater transparency which can help analysts make better forecasts. Accordingly, negative relations to forecasts dispersion and also forecasts errors are expected. However, though many researches include firm size as a control variable in studying analyst forecasts, the net effect of firm size is ambiguous (Frankel et al. 2006). Probably
because firm size can also proxy for a host of other factors, such as managers’ incentives, for which predictions for the relation with forecast accuracy are unclear temporarily.
(4) The volatility of earnings, which is measured by the coefficient of standard deviation of earnings per share before extraordinary items deflated by the stock price over the past five years (b`_d#, %Ba B(%BX)). Extant literature proves that longer-term earnings volatility is associated with more optimistically biased and thus less accurate forecast numbers (Kross et al. 1990; Lim 2001). Accordingly, we also control for earnings volatility, which is expected positively related with dispersion and errors of analysts GAAP forecasts.
At last, (5) firm risk as reflected by the cumulated stock returns over the previous five years (e^O#, %Ba B(%BX) ). e^O#, %Ba B(%BX) serves as a risk measure, with higher values representing higher risk, consistent with the risk-return framework of financial economics. Based on the related prior literature (Mensah et al. 2004), the signs expected for the firm risks in both of regressions are negative.
3.4 Regression Model
To test the hypotheses of our research, we regressed the individual firm’s forecast dispersion and absolute forecast errors on the independent variable of Non-GAAP earnings disclosure ('(#,%) and the other five control variables. The regression model can be written as:
\!#,% = f + halog \(&#,%+ hU'#,%+ hjlog _`#,%Ba+ hkb`_d#, %Ba B(%BX)
+ hXe^O%, %Ba B(%BX)+ la'(#,%+ m#,%
where,
\!#,% =analysts’ forecasts attribute, either !"#,% or !&#,%;
!"#,% =analyst forecasts dispersion for firm i of year t deflated by the stock price at beginning of the fiscal year t;
!&#,% =absolute forecast error of firm i, in year t deflated by the stock price at the beginning of the year t;
!R,#,% =forecast by analyst j for firm i of year t (with !#,% as the mean);
log \(&#,% =natural log of the average (in days) of the forecast for firm i in year t;
'#,% =number of analysts whose forecasts are included in the statistics computed for firm i of year t;
log _`#,%Ba =natural log of total market capitalization for firm i at the beginning of the fiscal year t;
b`_d#, %Ba B(%BX) =coefficient of variation of the last five years’ earnings before extraordinary items ending at period t-1 deflated by absolute median;
e^O%, %Ba B(%BX) =previous five years’ cumulative stock returns;
'(#,% =whether the firm disclose Non-GAAP earnings information, equals 0(not disclose) or 1(disclose).
Table 1. Variables Definitions
For both !"#,% and !&#,%, support for the directional hypotheses (H1 and H2) would be indicated by negative coefficients for '(#,%. '(#,% measures whether managers from firm i disclose Non-GAAP earnings, with value at 1 indicating managers disclose Non-GAAP information at least once that year, value at 0 otherwise. According to our hypotheses development, Non-GAAP disclosure leads to smaller forecasts dispersion and fewer forecasts errors, therefore negative coefficients should emerge in the regressions.
3.5 Sample Period and Sample Selection
The investigated period is 2010–2012. The reasons for choosing this period as my sample period are as follows. First, firms started disclosing Non-GAAP information since 2003, and till 2009 the percentage of firms which disclose Non-GAAP information has substantially increased (Black et al. 2017). The sample period of my study should best be after 2009. In addition, my study directly use data from the dataset of Bently et al (2017) to determine the independent variable '(#,%, and the most complete observations of their dataset are from 2010 to 2012 among the whole recording period 2003-2012. Therefore, our sample period is 2010-2012.
Table 2 presents the sample selection process. The firms included in Bently et al (2017) dataset are all consolidated firms in the CRSP, Compustat, and I/B/E/S with fiscal years ending between 2003 and 2012. I initially obtain all firms from their dataset over 2010-2012 period with a gvkey identifier (6355 firms), and then exclude financial firms which do not have all observations for the three fiscal years (3593 firms left). In the next step we match the original firm identifier, gvkey, with cusip, which is the firm identifier used in I/B/E/S and Compustat.
Unfortunately, not all gvkey can be matched with an 8-digit cusip and 1318 gvkey firms which cannot be matched with cusip codes have to be dropped (2275 firms left). The next step is to get all the initial data for calculating dependent and other control variables, more firms are excluded because the key variables of these firms for 2010-2012 period in I/B/E/S and Compustat are missing. (1311 firms left).
The whole selection procedure leads to 1311 firms left. For the whole sample period 2010-2012, all left firms have complete variables needed for regression model. Hence, our sample contains 3933 firm-year observations.
Selection criteria used to obtain sample of firms
Description n
All firms recorded in the dataset from Bently et al (2017) with a unique gvkey 6355
-Drop if missing observations in 2010 to 2012 -2762
-Drop if missing cusip -1318
-Drop if missing key I/B/E/S variables -648
-Drop if missing key Compustat variables -316
Amount of firms satisfying preliminary data requirements 1311
4 Empirical Results
4.1 Descriptive statistics and correlation matrix
Summary statistics on analysts’ forecasts accuracy data, numerical control variables and also independent variable used in my study are presented in Table 3. Columns 1 to 3 of Table 3 present the summaries for the three years from 2010 to 2012 and column 4 presents the summary for the entire period.
For the analysts’ forecast measures, the means of forecast error (i.e. the inverse of the forecast accuracy as defined above) for the three years are close, with average 0.055 for the whole sample period. Similarly, analysts’ forecast dispersions for the three years are also close, with average 0.068 for the whole period. This similarity is probably because our sample firms for the three years are the same, and analysts who follow these firms didn’t change significantly.
For control variables, average forecast age (log(Age)) shows a slight but steady decline from the mean of 5.117 in the first year to 5.050 at the end. Within the whole sample period, numbers of analysts following don’t change a lot, with the same median 2 analysts following per company. Similarly, the firm size is also very stable, this is probably because I use the same sample firms for three years, and the economy is quite stable in 2010-2012. Earnings volatility shows a minor increase in this period, with the mean from 1.940 in 2010 to 1.955 in 2012. Notably, Previous return shows a big and steady increase, this makes sense because this variable covers stock return from five years before till latest, which start from economic crisis at the beginning and recover stably afterwards.
It is surprising to see that the independent variable in my study shows a small decrease in 2010-2012, because Non-GAAP disclosure has been very popular for firms recently. One of the possible explanations for this is that original dataset we use to determine whether managers disclose Non-GAAP earnings doesn’t provide observation every quarter for every company, sometimes there would be missing observations from some companies in a certain year. This is a flaw of my research, my independent variable may not be complete when transfer quarterly observations to become a yearly variable due to the missing observations.
Table 4 presents the correlation coefficients of the regression variables in my regression model. The correlation coefficient between the two analysts’ forecast measure is highly positive and significant.
Among all the control variables, significantly negative correlations are found between the firm size and both two analysts’ forecast measure. This means that the result of the univariate analysis generally supports the predicted positive relationship between firm size and analysts’ forecast accuracy, which is also confirmed by the regression results later. Moreover, firm size is also related with the number of analysts following, earnings volatility, previous return and
managers’ Non-GAAP disclosure, which is consistent with that firm size can be an ambiguous proxy for a lot of related factors (Frankel et al. 2006). In addition, the number of analysts following is found positively related with firm size, earnings volatility and manages’ Non-GAAP disclosure, which is consistent with prior expectation that more information and higher higher difficulty of forecasting would need more analysts to interpret.
Variable Observation Statistics
Year 2010 N=1311 Year 2011 N=1311 Year 2012 N=1311 Entire Period N=3933 Analysts Forecast Attributes
(dependent variables)
Analysts’ forecast error, Mean 0.051 0.064 0.055 0.057
!&#,% Median 0.008 0.007 0.007 0.007
Std.Dev. 0.204 0.876 0.561 0.612
Forecast dispersion, Mean 0.075 0.079 0.068 0.074
!"#,% Median 0.046 0.039 0.040 0.041
Std.Dev. 0.165 0.675 0.280 0.432 Control Variables
Average forecast age, Mean 5.117 5.096 5.050 5.088
log \(&#,% Median 5.308 5.276 5.252 5.278
Std.Dev. 0.806 0.791 0.841 0.813 Number of analysts following, Mean 3.330 3.258 3.301 3.296
'#,% Median 2.000 2.000 2.000 2.000
Std.Dev. 2.832 2.737 2.701 2.757
Size, Mean 7.061 7.055 7.055 7.057
log _`#,%Ba Median 6.990 6.997 6.989 6.990
Std.Dev. 1.737 1.734 1.742 1.737
Earnings volatility, Mean 1.940 1.948 1.955 1.948
b`_d#, %Ba B(%BX) Median 1.280 1.290 1.310 1.300
Std.Dev. 2.684 2.690 2.703 2.692
Previous returns, Mean 0.761 0.985 1.051 0.932
e^O%, %Ba B(%BX) Median 0.545 0.631 0.666 0.610
Std.Dev. 1.381 1.699 1.856 1.661 Non-GAAP disclosure measure
(testing variable)
Managers' Non-GAAP Mean 0.480 0.478 0.475 0.478
disclosure, '(#,% Median 0.000 0.000 0.000 0.000
Std.Dev. 0.500 0.500 0.500 0.500
FE FD logAge N logMV CV_X Ret NG
Analysts' forecast error, FE 1
Forecast dispersion, FD 0.7782* 1
0
Average forecast age, logAge -0.0024 -0.0266 1
1 0.9405
Number of analysts’ following, N -0.0215 -0.0127 0.0016 1
0.9957 1 1
Size, logMV -0.0592* -0.0747* 0.0419 0.0978* 1
0.0056 0.0001 0.2157 0
Earnings volatility, CV_X -0.018 -0.0042 -0.0033 0.0530* 0.3252* 1
0.9998 1 1 0.0242 0
Previous returns, Ret -0.0115 -0.0179 0.0024 -0.0011 0.2599* 0.5108* 1
1 0.9998 1 1 0 0
Managers' Non-GAAP disclosure,
NG 0.0065 -0.0084 0.0089 0.0515* 0.2212* 0.0378 -0.0127 1
1 1 1 0.0339 0 0.3959 1
4.2 Regression Results
In this session, I present the results of regressing analysts’ forecast measures on the control variables and the independent variable examining the availability of Non-GAAP disclosure. Table 5 shows the results of regressing analysts’ GAAP forecast dispersion(!"#,%). Column I indicates the expected signs of all control variables and independent variables, column II to IV present the results for year 2010 to 2012, and the last column presents the results for the whole period.
Among all the control variables, the most significant one is firm size, which is negatively associated with forecast dispersion in year 2010, 2012 and also for the whole period. This is consistent with our expectation, confirming that larger organizations normally are more regulated and under higher scrutiny (Jaggi and Jain 1998). Larger firms have to provide more complete private information to outsiders, and their disclosures are often with higher reliability and transparency. Hence, the higher information quality of reporting by larger firms helps analysts improve their forecast quality and decrease the forecast dispersion.
The positive association between earnings volatility and forecast dispersion is significant in 2010, but not significant in the following two years. Similarly, the variable of previous returns is also found 95% significantly related with forecast dispersion in the first year, while in the last two years the significant relation doesn’t exist anymore. Though these two relations are not consistent, both confirm to some extent that higher earnings volatility and returns are associated with lower forecast dispersion due to the higher difficulty in making forecasts on companies facing bigger volatility.
Unexpectedly, average forecast age is negatively associated with forecast dispersion in 2010. This association is significant in 2010, but becomes insignificant in the other two years. One possible explanation for this unexpected finding is that less information earlier causes analysts to have closer forecasts, while more information provided at the later stage that is closer to the actuals’ announcement date could cause analyst to make more judgments and bring greater dispersion.
For the other control variables, no more significant associations are found with analysts’ forecast dispersion on GAAP earnings.
Now we turn to the examination of the first hypothesis. The coefficients of managers’ Non-GAAP disclosure are not consistent over the sample period. In addition, I don’t find a significant relation between the availability of Non-GAAP disclosures by managers and analysts’ forecast dispersion.
Table 6 presents the results of regressing analysts’ forecast error(!&#,%). Similar with table 5, Column I lists the expected signs of all control variables and independent variables, column II to IV present the results for year 2010 to 2012, and the last column presents the results for the whole period.
Table 5. Regression results of the relation between Analysts' Forecasts Dispersion(!"#,%) and Managers' Non-GAAP
disclosure
Coefficient (t-value shown below)
Variable Expected Sign Year 2010 N=1311
Year 2011 N=1311 Year 2012 N=1311 Entire Period N=3933 Intercept 0.2550*** 0.3460** 0.2250*** 0.2730*** (7.39) (2.50) (4.01) (5.32) Control Variable Average forecast + -0.0130** -0.0260 0.0010 -0.0120 age, log *+&#,% (-2.36) (-1.09) (0.09) (-1.45)
Number of analysts - 0.0020 -0.0050 -0.0010 -0.0010 following, ,#,% (1.37) (-0.76) (-0.23) (-0.41) Size, - -0.0180*** -0.0200 -0.0230*** -0.0200*** log -.#,%/0 (-6.53) (-1.64) (-4.67) (-4.65) Earnings volatility, + 0.0080*** 0.0030 0.0020 0.0040 1._3#, %/0 /(%/5) (4.10) (0.33) (0.64) (1.45) Previous returns, - -0.0070* -0.0030 0.0001 -0.0030 789%, %/0 /(%/5) (-1.73) (-0.20) (0.09) (-0.57) Non-GAAP disclosure measure (testing variable) Managers'
Non-GAAP disclosure, - -0.0010 0.0340 -0.0100 0.0080
,+#,% (-0.08) (0.88) (-0.65) (0.54)
Adjusted R^2 0.0374 0.0230 0.0205 0.0052
F statistic 9.49 3.28 5.56 4.45
* Statistically significant at a probability of less than 0.05. ** Statistically significant at a probability of less than 0.01. *** Statistically significant at a probability of less than 0.001.
