Do price-earnings multiples for firms with patterns of increasing earnings vary with the quality of the earnings pattern?

ABSTRACT

Prior research finds that firms with patterns of consecutive earnings increases are rewarded by market participants with higher Price-Earnings (P/E) multiples, and that some firms manage earnings to sustain the earnings stream. We propose several fast and inexpensive heuristics to classify firms as suspect or non-suspect of maintaining the stream through earnings management and examine whether market rewards differ for suspect and non-suspect firms. We find higher P/E multiples for firms with a pattern of increasing earnings supported by the same pattern of increasing cash flows (non-suspect). This is an important result as it shows that cash flows can be used to assess the quality of earnings in scenarios prone to earnings management, where earnings quality is expected to be low. We also find lower price-earnings multiples for firms suspected of using accrual-based earnings management, sales manipulation, and overproduction to achieve the earnings pattern.

KEYWORDS:

• earnings pattern
• cash flow pattern
• benchmark beating
• earnings management
• real activities manipulation
• P/E ratio
• fundamental analysis
• financial reporting

JEL CLASSIFICATION:

• G10
• G30
• M41

Acknowledgments

We are grateful for the helpful comments and suggestions received from Michael Brennan, Beatriz García Osma, Begoña Giner, Ole-Kristian Hope, Garen Markarian, Peter Pope, William Rees, Gianfranco Siciliano, Martin Walker and workshop participants at Universidad Carlos III de Madrid, the European Doctoral Colloquium in Accounting, the European Accounting Association Annual Congress, the Spanish (ASEPUC) Doctoral Colloquium in Accounting, the International Accounting Research Symposium, and the Workshop on Empirical Research in Financial Accounting. We acknowledge financial assistance from the Spanish Ministry of Innovation and Science (ECO2013-48328, ECO2016-77579 and PID2019-111143GB), from Comunidad de Madrid (Programa Excelencia para el Profesorado Universitario, convenio con Universidad Carlos III de Madrid, V Plan Regional de Investigación Científica e Innovación Tecnológica, EPUC3M12) and from FEDER (UNC315-EE-3636).

Disclosure statement

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

Notes

1. Following the results in Dechow et al. (2003) and Brown and Caylor (2005), prior research has focused mainly on the meeting or beating of earnings forecasts, and there is scarce evidence on the incidence and valuation implications of streams of consecutive increases in annual earnings.

2. While the subset of firms with a stream of at least 5 years of consecutive increases in earnings is small, we believe studying this small subset of firms is important. These are the firms that can be suspected of engaging in earnings expectations games, of the type described in Jensen (2005) that can be especially detrimental to the whole of the economic and financial system. Therefore, understanding their reporting and how capital markets respond to their reported numbers is an important issue. Prior research on benchmark beating firms also studies small samples. For example, Barth et al. (1999) focus on a subset of 419 firm-year observations, from a total sample of 21,173. Roychowdhury (2006), who studies real earnings management, focuses on a subset of 503 firm-year observations with small profits, from a total sample of 21,758 observations.

3. For a discussion of the earnings smoothing phenomenon see, Dechow et al. (2010). There is an ongoing debate in the literature on whether earnings smoothing is rewarded by capital market participants.

4. Bird et al. (2019) argue that the benefits linked to earnings management to beat targets are expected to be larger after SOX, ‘consistent with the market rationally becoming less skeptical of firms that just meet benchmarks’ (p. 1).

5. Early evidence documenting benchmark beating behaviour, its determinants and consequences, includes Burgstahler and Dichev (1997), Degeorge et al. (1999), and Peasnell et al. (2000).

6. For a review of the earnings management and benchmark beating literature, see, Walker (2013).

7. Chu et al. (2019) examine the related issue of earnings manipulation to consistently beat earnings forecasts.

8. An alternative view of cash flows to the one stated in this hypothesis is that they can also be managed, as described by Roychowdhury (2006), among others. However, managing cash flows is more difficult and costlier than managing accruals (Roychowdhury, 2006) and also, the management of cash flows does not seem to be driven by the desire to maintain a cash flow pattern, but by the desire to meet financial reporting goals (Roychowdhury, 2006) based on earnings benchmarks. Therefore, we believe that while this might be a concern, we can still use cash flow patterns as a signal of a high-quality earnings pattern. We can see it the other way around: an earnings pattern without a similar cash flow pattern would be a signal of a low-quality earnings pattern.

9. There is an important controversy on how to deal with scale problems with this type of models (see, for example, Ohlson, 2015). Following Barth et al. (1999), Barth and Clinch (2009), and Aledo Martínez et al. (2020) we use per share values.

10. We also replicate all our tests using stock prices at the end of the fiscal year, as in Barth et al. (1999).

11. DBeat5 is equivalent to the variable D_up in Barth et al. (1999).

12. We acknowledge that the number of firms with a stream of at least 5 years of consecutive earnings increases is small compared to the universe of Compustat firms that we use. Therefore, finding a significant coefficient on EPS*DBeat5 can be driven by firms with the earnings stream being different in other aspects from firms without the earnings stream. Even if that is the case, our main coefficient of interest is EPS*DBeat5*DCFO5, where the focus is only on the subset of firms with a stream of consecutive earnings increases, and whether, among those, firms whose stream is backed up by a stream of consecutive increases in cash flows receive an additional reward. Prior research on streams of earnings increases also use the universe of Compustat firms as starting point.

13. Given that our main aim is to assess methods to identify firms that are suspected of managing earnings to maintain the earnings stream, we do not use the existence of SEC enforcement actions, class action lawsuits or restatements, as these are ex-post events that lead to negative stock market reactions and that, at the same time, put an end to the earnings stream. That is, we focus on firms for which it is unclear whether the earnings stream is genuine.

14. Prior research (see for example, Zang, 2012; Di Meo et al., 2017; Mellado-Cid et al., 2019) argues that abnormal CFO is a difficult to interpret proxy, as it can also be affected (in the opposite direction) by decreased discretionary expenses to beat benchmarks. This, in our view, would bias the measure towards not identifying cases of earnings management, and, therefore, it could only lead to not being able to find the hypothesised effect, not the other way around. Consistent with our view, this proxy is still used in recent studies on real earnings management (i.e. Abad et al., 2018; Cheng et al., 2016; Doukakis, 2014; Garcia Osma et al., 2020; Haga et al., 2018; Kim et al., 2017; Kim & Sohn, 2013; Pappas et al., 2019; Shi et al., 2018; Sohn, 2016).

15. This is a common assumption in the literature. See, for example, Zhao et al. (2012) who argue that the overproduction proxy also captures overly friendly price discounts and other trade credit terms. Francis et al. (2016) also explain and provide examples of how this variable can capture ‘deep’ discounts.

16. As in Barth et al. (1999) the number of observations with streams of at least 5 years of consecutive earnings increases is small. In their case, they have 419 observations (a 2% or their sample). In our case, the 977 observations represent a 3% of our total sample.

17. The frequency of the earnings pattern is based on the sample used for our main tests (Tables 4 and 5).

18. While we have a relatively large sample, the number of suspect or non-suspect firm-year observations, which are key for our analysis, is small, ranging from 143 to 401 observations across the different specifications. Therefore, given this small sample, and following Gow et al. (2010), it does not seem advisable to use double-clustering as in Petersen (2009). Instead, we only cluster standard errors at the firm level, and use White (1980)-type heteroskedasticity consistent standard errors.

19. We replicate our tests using income streams of at least six years, obtaining identical inferences. As we use shorter streams, results are less consistent with our hypotheses.

20. In additional tests we include, as additional control variables, cash flow growth and cash flow volatility. Our results and inferences do not change when we include those two control variables.

21. One possible interpretation for the fact that EPS*DBEAT5 is insignificant in the full model when we use prices three months after the fiscal year end, and significant when we use prices at the fiscal year end, is that investors reward benchmark beating firms at the fiscal year end, but adjust and analyse more critically the full set of financial statements, including cash flow patterns, once the full financial statements are released. This would be in line with the results in DeFond and Park (2001) and Baber et al. (2006).

22. This result holds if, instead of total discretionary accruals, we use working capital discretionary accruals. It also holds if we use the modified Jones model proposed by Dechow et al. (1995), and also if we augment the Jones (1991) model with current, one-year-lagged, and one-year lead cash flows, as in McNichols (2002) (the so called Dechow-Dichev augmented Jones model). Godsell et al. (2017) provide evidence that the augmented Jones model identifies well cases of earnings management. We also use performance matched discretionary accruals in the spirit of Kothari et al. (2005), including ROA as a control variable in the first step of the estimation of the discretionary accruals. In this case, we do not find lower price-to-earnings ratios for suspect firms. Given that, in our research setting, earnings management is driven by poor performance, including a control for ROA makes it more difficult to identify cases of earnings management, leading to an under-identification of suspect firms.

23. When we use $\mathrm{\Delta }PRICE$ as the dependent variable,$\mathrm{\Delta }PRICE$, $\mathrm{\Delta }EPS$ and $\mathrm{\Delta }BVS$ are deflated by lagged $PRICE$.

24. We do not include firms suspected of managing discretionary expenditures in our overall measure because, given the results in our main tests, the proxy might be noisy. If we also consider firms suspected of opportunistically reducing discretionary expenditures in our overall identification of suspect firms, we end up classifying as suspect more than 60% of the firms with an increasing earnings trend. This is probably not accurate, and it does not help in identifying the firms that are actually suspected of managing earnings to maintain the stream.

Additional information

Funding

This work was supported by the Comunidad de Madrid [EPUC3M12]; FEDER [UNC315-EE-3636]; Spanish Ministry of Science and Innovation [ECO2013-48328, ECO2016-77579, PID2019-111143GB].