Analyst Revenue Forecasts and Firm Revenue Misstatements

Abstract

Earnings and revenue are ranked as the two most important performance measures reported to outsiders. However, prior literature primarily focuses on management’s incentives and willingness to manipulate earnings to meet earnings benchmarks. Increasingly, analysts are releasing forecasts of financial items in addition to aggregate earnings, with revenue forecasts being the most common. We posit that management may be under similar pressure to meet revenue forecasts given their impact on firm value. As such, we examine whether analyst revenue forecasts aggravate revenue misstatements. We find that revenue misstatements are positively associated with revenue forecasts and the association is more pronounced when beating revenue forecasts is more important. We also show that several characteristics of revenue forecasts aggravate management pressure and thus the likelihood of revenue misstatements. Further analyses suggest that firms use accruals management rather than real earnings management to inflate reported revenue. Our findings may be useful to academics, investors, and regulators in examining the relationship between analyst revenue forecasts and firms’ financial reporting behavior.

Keywords:

• Analyst
• Revenue forecast
• Revenue restatement
• Revenue misstatement

JEL codes:

• G10
• G24
• M41

Acknowledgements

We thank Beatriz García Osma (editor) and two anonymous reviewers for their valuable comments and guidance. We thank Fujiing Shiue, Rong-Ruey Duh, Jeng-Ren Chiou, Hua-Wei Huang, Jeng-Fang Chen, and Yenn-Ru Chen for their helpful comments. We are also grateful for the suggestions received from Mohan Venkatachalam, Agnes C. S Cheng, David A. Wood, and Ting-Kai Chou. We appreciate the comments from Yi-Hung Lin and Yu-Lin Huang as well as the suggestions from the reviewers and the participants of 2016 AFAANZ conference and 2016 AAA Annual Meeting. Errors and omissions are our responsibility.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data Availability

Data are available from the public sources cited in the text.

Supplemental Data

Supplemental data for this article can be accessed https://doi.org/10.1080/09638180.2021.1983447

Notes

1 We note that over ninety percent of firms with earnings forecasts also have revenue forecasts since 2003, which suggests that the number of revenue forecasts better captures the importance of meeting revenue targets than the existence of revenue forecasts. We explore alternative specifications later and find consistent results.

2 For example, in the sample used in this study, the average number of revenue forecasts and revisions is four times the average of cash flow forecasts and revisions.

3 For example, analysts are more likely to issue cash flow forecasts in addition to earnings forecasts when earnings are uninformative (DeFond & Hung, 2003). This suggests that cash flow forecasts may serve as substitutes for earnings forecasts, whereas revenue forecasts complement earnings forecasts (Rees & Sivaramakrishnan, 2007; Ertimur et al., 2011). In addition, Lerman (2020, Table 1 Panel B) examines financial message boards and shows that earnings and revenue are the no.1 and no.3 most frequent types of accounting information that attract investor attention, while cash flows are the no.11 most frequent type of accounting information.

4 For how analyst forecasts influence investor earnings fixation, see also Bushee (1998), Libby and Tan (1999), Tan et al. (2002), Libby et al. (2006), Han and Tan (2007), Hirst et al. (2007), Houston et al. (2010), and Elliott et al. (2011).

5 Our study is related to Chou et al. (2021) who find a negative association between the existence of analysts’ quarterly revenue forecasts and discretionary revenue; however, there are several key differences between our studies. First, rather than quarterly disaggregated forecasts, we examine annual disaggregated forecasts, which are arguably more common and widely examined in the literature (e.g., Givoly et al., 2009; Keung, 2010; Ertimur et al., 2011; McInnis & Collins, 2011; Lee, 2012; Cheng et al., 2020). Second, we examine the number of revenue forecasts and revisions, while Chou et al. (2021) examine the existence of quarterly revenue forecasts. As indicated in their study, initiations of quarterly revenue forecasts are scarce in recent years (e.g., 116 initiations in 2008 and 150 initiations in 2016). Similarly, an untabulated analysis indicates that more than ninety percent of observations with earnings forecasts also have revenue forecasts since 2003. Therefore, it is believed that the number of revenue forecasts can better capture the importance of revenue forecasts than the existence of revenue forecasts. Third, our paper examines revenue restatements, which are a direct and strong measure of revenue misstatements (Aobdia, 2019) and do not suffer from the potential problem of incorrect inferences when using residuals as dependent variables (Chen et al., 2018), while Chou et al. (2021) examine discretionary revenue. Lastly, we analyze several characteristics of revenue forecasts, examine the conditions that may strengthen or weaken the association between revenue forecasts and revenue misstatements, and explore the relation between revenue forecasts and the type of earnings management techniques used.

6 Ertimur et al. (2003) show that markets react more positively to revenue surprises than to cost savings, and Rees and Sivaramakrishnan (2007) find that the market premium (penalty) to meeting (missing) earnings forecasts is accentuated when revenue forecasts are met (missed) at the same time. They also find that the act of meeting revenue forecasts is associated with an equity premium that is separate and distinct from the premium associated with meeting earnings forecasts.

7 Matsunaga and Park (2001) provide evidence that managers have incentives to meet benchmarks and that the strength of that incentives may be influenced by whether the firm has missed a benchmark during the year. The study shows that CEO bonuses are negatively affected when the firm’s quarterly earnings fall short of the consensus analyst forecast or the earnings for the same quarter of the prior year (Matsunaga & Park, 2001). In addition, Mergenthaler et al. (2012) find that CEOs and CFOs are penalized for failing to meet consensus analyst forecasts.

8 Similarly, Apple Computer, Inc. experienced an 11 percent decrease in stock price after it announced earnings of $0.52 per share, which was higher than the consensus earnings estimate of $0.37 per share, and the negative market reaction was said to be caused by Apple’s failure to beat forecasted revenue (Rees & Sivaramakrishnan, 2007). Additionally, in 2016, the Wall Street Journal reported that the Burberry Group CEO was under pressure to turn around the retailer’s performance after posting disappointing second quarter sales (Chaudhuri, 2016).

9 In addition, when compared to earnings or profit margins, revenue proves more difficult to manage (Marks, 2008; Cheng et al., 2020).

10 It is possible that firms engage in revenue management techniques that allow them to beat revenue forecasts without having to restate financial reports. If this is the case, this would prevent this study from finding a significant association between revenue forecasts and revenue restatements. However, these techniques (e.g., real earnings management) could be costly and take longer time to implement, and the effect on reported figures is hard to control (McInnis & Collins, 2011; Zang, 2012). Therefore, they may not serve as perfect substitutes of revenue management practices that are inconsistent with GAAP.

11 Eight cases of revenue restatements due to clerical errors are excluded because they are less likely to be intentional. Our results remain consistent when these cases are included.

12 This study focuses on one-year-ahead analyst annual forecasts on a firm’s financial report in year t.

13 Gentry and Shen (2013) show that CEO turnover is strongly associated with firm performance when the firm is covered by many analysts, and Gleason and Lee (2003) also show that the post revision drift is more pronounced for firms with lower analyst coverage.

14 Keung (2010) finds that earnings forecast revisions that include sales forecast revisions have a greater impact on security prices. Hobbs et al. (2012) also find that frequently revising analysts yield more profits than infrequently revising analysts.

15 Industry fixed effects are based on the two-digit Standard Industrial Classification.

16 Observations with analyst revenue forecasts but without earnings forecasts are rare and hence excluded.

17 All continues variables are winsorized at the 1st and 99th percentiles.

18 Two exceptions are the high correlations between LOG_REV and LOG_EPS and between LNTA and SHARES. The variance inflation factors (VIFs) are examined to evaluate whether the results are affected by multicollinearity. The VIFs on LOG_REV and LOG_EPS are 13.04 and 14.45, marginally higher than ten. To evaluate whether multicollinearity drives our results, we exclude LOG_EPS from the regression, and the sign and significance remain consistent. The VIFs on LNTA and SHARES are 7.03 and 4.04, and are lower than 3 on other variables. These indicate that multicollinearity is not a concern (Kennedy, 2008).

19 The percent change in the odds of a revenue restatement for one standard deviation change in LOG_REV is calculated as (100 x (e^{(β x 1 standard deviation)} – 1)), which equals 50.20 (100 x (e^{(0.401 x 1.097)} – 1)).

20 Gaver and Utke (2019) indicate that ‘entropy balancing is an equal percent bias reducing matching method, which ensures that covariate imbalance improves after matching’ and that entropy balancing increases testing power because no observations are discarded and there are no random matches (King et al., 2011). In contrast, while propensity score matching (PSM) is often used in the literature, PSM does not ensure treatment firms and control firms are identical, often select random matches and fail to solve the functional form misspecification (covariate imbalance), and reduces testing power due to discarding observations (King et al., 2011; King & Nielsen, 2019; DeFond et al., 2017; Shipman et al., 2017; McMullin & Schonberger, 2020; Gaver & Utke, 2019). The problem of low-power tests is particularly relevant for our study because our proxy for revenue management is revenue restatements, which is a dichotomous variable and relatively rare. In their response, Gaver and Utke (2020) conclude that entropy balancing is preferred ‘because it accounts for observable differences across clients without random matching or discarding data (Hainmueller, 2012; McMullin & Schonberger, 2020)’ and that PSM ‘should be used with caution or not at all.’

21 Industry fixed effects are excluded when controlling for firm fixed effects.

22 Following Chen et al. (2021), we estimate a linear probability regression model when firm fixed effects are included, in order to avoid the incidental parameter problem, which biases estimates when including a large number of fixed effects in a nonlinear model (Lancaster, 2000).

23 While 26.12% may sound small, this may be explained by several potential reasons. First, some firms may focus more on beating earnings targets than on beating revenue targets. Second, some revenue misstatements may haven’t been discovered. Third and importantly, Audit Analytics only provides cumulated change in net income. As a restatement may involve more than one year and as a firm could restate financial statements more than once, the net effect of revenue restatements on the reported revenue in a year is unclear. In addition, a restatement could involve more than one account, and the restatement of one account may offset the restatement of another account. Lastly, it is likely that misstating net income by one dollar requires misstating revenue by more than one dollar.

24 We thank the Editor for suggesting this important analysis.

25 The full tables of Tables 5, 6, 7, and 8 are available in Online Appendixes D, E, F, and G, respectively.

26 See Panel A of Online Appendix D for the pair-wise correlations between different specifications. The correlation between the logarithm of the number of analysts issuing revenue forecasts and the logarithm of the number of analysts issuing earnings forecasts are extremely high (0.97) when revisions, first issuance, and last issuance are not considered.

27 The VIFs on LOG_REV and LOG_EPS are below ten under the first and second specifications.

28 We use the top forty percentile as the cutoff to increase the possibility that a treatment observation can find a matched control observation.

29 We use a relatively low caliper to ensure that the control variables are generally identical for treatment firms and control firms, and implement matching with replacement to find the most similar control firm observation.

30 Results are consistent when using year t+2 or t+3 as the cutoff point.

31 That is, reported revenue is higher than the consensus revenue forecasts, and the difference scaled by the consensus revenue forecasts is between 0 and 0.03. Results are consistent when using the median value of revenue forecasts.

32 Results are consistent when using different thresholds, such as 0.01, 0.02, 0.04, and 0.05.

33 Discretionary revenue is the calculated as the residuals from the revenue model which regresses the change in accounts receivable on the change in revenue in the first three quarters and the change in revenue in the fourth quarter. Please refer to the model (1) in Stubben (2010).

34 We thank the anonymous reviewer for pointing out this alternative explanation.

35 We thank the anonymous reviewer for suggesting this scenario.

36 The results remain consistent when using lagged PS for sample split.

37 The results are consistent when using lagged industry average LOG_REV for sample split.

38 Measures of accruals and real earnings management are estimated by year-industry groupings, with a requirement of at least fifteen non-missing observations for each grouping. Industries are classified based on two-digit Standard Industry Classification.