Marital Status and Earnings Management

Abstract

In this note, we examine the effect of CEO marital status on the riskiness of financial reporting. Using multiple proxies, we find that firms headed by a single CEO display a higher degree of earnings management than those headed by a married CEO. The effect is economically significant. Our results persist in an instrumental variable regression, suggesting that our results are not driven by innate heterogeneity in preferences.

Acknowledgements

The authors thank Laurence van Lent (the Editor), three anonymous referees and seminar participants at Singapore Management University, INSEAD and Tsinghua University for useful comments. The authors also thank Nikolai Roussanov and Pavel Savor for making CEO marriage information publicly available.

Supplemental Data and Research Materials

Supplemental data for this article can be accessed on the Taylor & Francis website, doi:10.1080/09638180.2016.1266958.

Notes

1 Specifically, we use the dataset Roussanov and Savor made available at http://dx.doi.org/10.1287/mnsc.2014.1926. We thank them for sharing their data.

2 We have randomly selected 400 CEOs in our sample and manually searched for their marital information. We were able to replicate 98.5% of their coding. Furthermore, we form a stratified sample comprised of the 100 CEOs for whom we have verified the marriage date and 17 randomly selected single CEOs (to match the proportion of single CEOs in our main sample). We re-estimate our baseline regression using these 117 CEOs. Our conclusions are not affected (See the Online Supplemental Material for details).

3 Our analysis focuses on accruals earnings management because accruals management carries significant legal risk, but real earnings management does not. Therefore, we expect the effect of a CEO's marital status to be concentrated in accruals management.

4 Some married CEOs ultimately divorce (e.g. Jack Welch of GE), which may introduce measurement errors in the coding of marital status. We discuss and address the potential impact of measurement errors on our estimates in the Online Supplemental Material. Our tests suggest that such measurement errors are unlikely to affect our conclusions.

5 To mitigate the concern that the biological difference in testosterone between men and women, independent of their marital status, could drive our results (e.g. Jia et al., 2014), we perform two untabulated robustness tests: (1) we control for the CEO gender, and (2) we exclude observations with female CEOs. Our conclusions are not affected.

6 The t-statistic associated with Single is 1.51 when we use the Modified Jones (1991) Model.

7 We report the second stage regression for earnings management measure using the Modified Dechow and Dichev (2002) Model in Column 2 of Table 1 and the other two measures in Table A6 of the Online Appendix).