CEO Facial masculinity and accounting conservatism

Abstract

In this study, we examine the consequences of CEO facial masculinity for accounting conservatism. Facial masculinity is associated with an array of masculine behaviours including aggression, ambition, egocentricity, risk taking, and an increased desire to maintain social status. We predict that such behaviours lead to aggressive financial reporting practices that incorporate bad news into earnings in a less timely manner (i.e. lower conditional conservatism). Using a sample of S&P 1500 firms from 1993 to 2015, we find that CEOs’ facial masculinity is associated with less conservative accounting. This finding is robust to the use of several measures of conservatism. Further, we document that stronger external monitoring dampens the negative relationship between CEO facial masculinity and conservatism. Our findings complement recent work that reveals CEO facial masculinity is positively associated with fraud and AAERs, and contributes to the literature by documenting the effect of an ‘off the job’ CEO characteristic on accounting conservatism.

Keywords:

• CEO Characteristics
• accounting conservatism
• masculinity

Acknowledgement

We thank Rajiv Banker, Sudipta Basu, John Daniel Eshleman, Robert Felix, Chansog (Francis) Kim, Mark Robinson, Zhifeng Yang, Haoran Zhu and workshop participants at Stony Brook University for helpful feedback. Peng Guo acknowledges financial support from the Ten Haken fellowship.

Data availability

Data used in this study are publicly available at sources cited in the text.

Disclosure statement

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

Notes

1 There is still debate in the literature as to whether testosterone causes men to have more masculine faces (e.g. Deaner et al. 2012). However, the key finding in this literature that we rely on is that facial masculinity is associated with a set of masculine behaviours.

2 See also Lindberg et al. (2005).

3 Prior work posits that accounting conservatism cannot, in large part, be explained by earnings management (Watts 2003). Therefore, even if a significant portion of the accounting fraud and AAERs documented by Jia et al. (2014) is caused by the earnings management behaviour of masculine CEOs, it is distinct from accounting policies impacting the timeliness of incorporating good news and bad news into earnings.

4 As defined, we focus on conditional conservatism in this study. Conditional conservatism requires a higher degree of verification for recognising good news relative to bad news (Basu 1997). By contrast, unconditional conservatism refers to a consistent downward bias in net asset valuation. Unlike conditional conservatism, unconditional conservatism does not consider economic news events after the inception of assets and liabilities (Beaver and Ryan 2005). Moreover, Ettredge et al. (2015) document a significant relationship between litigation risk and conditional conservatism, but not unconditional conservatism. Therefore, we expect the link between masculinity and conditional conservatism to be more salient. Our additional tests consider unconditional conservatism to demonstrate the contrast.

5 This is similar to the audit pricing literature, which finds that CFOs with untrustworthy faces are charged higher audit fees (Hsieh, Kim, Wang, and Wang 2020).

6 ExecuComp only covers the S&P 1500. Therefore, we do not collect photographs of CEOs belonging to non-S&P 1500 firms in the year 2008. As such, if an S&P 1500 firm in 2012 was a CEO at a non-S&P 1500 firm in 2008, then we would not have collected this CEO’s photograph.

7 This measure is associated with more masculine behaviours in men, but not women (Geniole et al. 2012). Therefore, our sample consists of only male CEOs.

8 Due to the subjectivity and skill involved in selecting and measuring CEOs’ facial photos, we replicate Jia et al.’s (2014) analysis to validate our measure. More discussion is provided in Section 6.3 of this study.

9 The percent is calculated as (−0.428 × 0.221)/1.196 = 0.08, where −0.428 is the coefficient on $D\times \mathrm{RET}\times \mathrm{fWHR}$, where 0.221 is the difference between the 1st and the 3rd quartile of fWHR (obtained from Table 2), and 1.196 is the coefficient on $D\times \mathrm{RET}$.

10 Nevertheless, we acknowledge that we cannot completely rule out the potential for selection bias. That is, more aggressive firms may seek to hire more aggressive CEOs, with ‘masculine’ characteristics. Time invariability of CEO facial masculinity and limited time variability of conditional conservatism constrain our ability to use empirical remedies to address such endogeneity issues. Thus, the findings of our study should be interpreted with this caveat in mind.

11 A notable exception is Graham, Harvey, and Puri (2013) who conduct psychometric tests to a large sample of CEOs and find that CEO characteristics such as risk aversion and optimism influence corporate policy. This type of analysis is difficult for most researchers to replicate, as it requires a large number of CEOs willing to undergo psychological testing.

12 As we discuss in the research design and results sections, we take steps to alleviate endogeneity concerns but also acknowledge that data constraints limit our ability to completely rule out the possibility of self-selection.

13 Given that CEO overconfidence is measured as the propensity for CEOs to hold in-the-money stock options, it is difficult to say whether the auditors are reacting to CEOs’ overconfident personality or the fact that their personal stock holdings indicate that they are betting on their firm.

14 For evidence that men with higher testosterone levels are rated as being more masculine and more attractive, see Penton-Voak and Chen (2004).

15 There are two studies which fail to find a relationship between facial masculinity and aggression (Deaner et al. 2012; Ozener 2012). A limitation of Deaner et al. (2012) is that the authors examine a sample of professional hockey players, which may be quite different from the general population. A limitation of Ozener (2012) is that he relies on undergraduate students’ self-reported aggression, rather than observing aggressive behaviour directly.

16 Hereafter, we refer to conditional conservatism as ‘conservatism’ for brevity except in cases where we specifically refer to unconditional conservatism to draw contrast.

17 For example, Watts (2003) points out that conservatism ‘offsets managerial bias [and] constrains managements’ opportunistic payments to themselves and other parties’ which increases firm value and overall welfare.

18 If facial width is largely the result of craniofacial bone growth occurring during adolescence, a man’s FWHR should be relatively stable after puberty. Jia et al. (2014) verify this for a small sample of 19 CEOs for which they were able to obtain two photographs for each CEO at two different points in their lives.

19 In section 6 we replicate the findings of Jia et al. (2014) to validate our measure of fWHR.

20 Ball, Kothari and Nikolaev (2013b) recommend researchers to ‘simply control for firm-specific effects, at least as a robustness check, to avoid potentially spurious inferences’ (p. 757). Our main model includes controls for a host of firm-specific characteristics, in addition to industry and year fixed effects. As a robustness check, we also replace industry fixed effects with firm fixed effects and find that the results are qualitatively similar. Nevertheless, for our main tests, we rely on industry fixed effects due to the relatively limited within-firm variation in fWHR.

21 Although we follow prior work (e.g. Ettredge et al. 2012) showing the usefulness of C-score in capturing variations in accounting conservatism, Byzalov and Basu (2022) issue caution regarding the use of C-score, arguing that such regression-based scores used as a dependent variable in a second stage can cause biases and difficulties in interpretation.

22 Khan and Watts (2009) show that their C-score captures asymmetric earnings timeliness at horizons up to 3 years ahead, making that a reasonable measurement window. We follow prior work in measuring C-score in this way (e.g. García Lara et al. 2014).

23 As we do with the Basu (1997) model, we conduct a robustness test with respect to CScore in which we replace industry fixed effects with firm fixed effects. We find that the results are qualitatively similar.

24 That is, relying on a fiscal year that is not too recent and not too far in the past allows us to exploit the number of years of available data for each CEO both before and after the year that our initial list is based upon. For example, a given fiscal year will include CEOs that will soon retire and CEOs that will continue to work for additional years. If we alternately select a more recent fiscal year, then our sample would lose out on observations pertaining to CEOs that would continue to work for more years after the initial selection. If we select a fiscal year farther in the past, then our sample would be limited by data availability during those years and may not be representative of the current environment. Selecting a year in the late 2000s (i.e. 2008) balanced these trade-offs. Given that the mean tenure of CEOs in our sample is approximately 7 years, choosing 2008 ensures that most CEOs will have completed their tenure by the end of our sample period.

25 Jia et al. (2014) report the mean (median) of fWHR 2.009 (2.006) for CEO-level sample with good-quality pictures, and 2.013 (2.011) for CEO-level sample with all measurable pictures.

26 The percent is calculated as (−0.428 × 0.221)/1.196 = 0.08, where −0.428 is the coefficient on $D\times \mathrm{RET}\times \mathrm{fWHR}$, 0.221 is the difference between the 1st and the 3rd quartile of fWHR (obtained from Table 2), and 1.196 is the coefficient on $D\times \mathrm{RET}$.

27 Although our main tests do not control for $\mathrm{EXTERNAL}$ because it is not brought up until this section, the main findings are not sensitive to controlling for it.

28 To test the moderating effect of external monitoring using the Basu model, we would have to include 4-way interactions in the model. One of the appealing features of a firm-specific measure of conditional conservatism is that such high degrees of interactions are not required to examine a moderating effect. Given our sample size due to our reliance on hand-collected data on CEO photographs from S&P 1500 firms in 2008, the model is likely to lack sufficient power with the introduction of a 4-way interaction and could generate statistically insignificant results even when an association may exist. As such, we only conduct this test using the CScore model which does not require such high degrees of interactions.