Understanding How the Effects of Conditional Conservatism Measurement Bias Vary with the Research Context

Abstract

We re-examine previous seminal studies on conditional conservatism (CC) that apply the asymmetric timeliness (AT) measure of Basu [(1997). The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting and Economics, 24(1), 3–37. https://doi.org/10.1016/S0165-4101(97)00014-1] and compare the outcomes with those based on the modified AT (MAT) measure of Badia et al. [(2021). Debiasing the measurement of conditional conservatism. Journal of Accounting Research, 59(4), 1221–1259. https://doi.org/10.1111/1475-679X.12366] and the spread in conditional variances (SCV) measure of Dutta and Patatoukas [(2017). Identifying conditional conservatism in financial accounting data: Theory and evidence. The Accounting Review, 92(4), 191–216. https://doi.org/10.2308/accr-51640]. Our conclusions are threefold. First, all three measures yield similar inferences in interrupted time-series settings that examine the change in CC following a policy mandate. Second, the inferences drawn from the AT measure in studies that model the determinants of CC based on cross-sectional settings are more sensitive to test specifications and research designs. Third, across the three measures, MAT shows the best empirical performance in terms of aligning with existing theories while being less affected by AT bias.

Acknowledgements

The authors thank the Editor and two anonymous referees for constructive comments and suggestions. The authors are also grateful for helpful comments of Colin Clubb, Joachim Gassen, James Ohlson, Panos Patatoukas, Andrew Stark, Konstantinos Stathopoulos, Norman Strong, as well as participants of presentations at the EAA annual conference 2018 in Bocconi University (Milan, Italy), Fudan University (Shanghai, China), Humboldt University of Berlin (Berlin, Germany), Lebanese American University (Beirut, Lebanon), and the University of Manchester (Manchester, UK).

Disclosure statement

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

Notes

1 For useful reviews on the CC literature, see Beyer et al. (2010), García Lara et al. (2014), Mora and Walker (2015), Penalva and Wagenhofer (2019), Ruch and Taylor (2015), Wang et al. (2009), and Watts (2003b, 2003a).

2 To help motivate our study, we conducted a search of all original research articles, since Basu (1997) to date, which apply the AT construct and are published in five leading accounting journals. We counted a total of 148 articles, with the time trend as well as topic and journal distribution provided in Appendix A.

3 In Appendix B, we provide a selected list of research publications that cite each of these four previous studies we re-examine, showing their importance and relevance to the literature through citation counts based on Google Scholar.

4 We discuss the issues that may arise when using the ratio instead of the spread of conditional variances in Section 4.

5 In our empirical analyses, we follow recent CC studies and calculate expected returns as the average returns of 5×5 portfolios formed by sorting firms first based on the beginning of year market value of equity and then based on beginning of year book-to-market equity ratio (Badia et al., 2021; Ball et al., 2013b; Dutta & Patatoukas, 2017).

6 In line with Patatoukas and Thomas (2011), Breuer and Windisch (2019) provide theoretical insights and empirical evidence suggesting that the earnings-returns concavity is increasing in the volatility of firms’ underlying shock processes (i.e., uncertainty). This results in an asymmetric effect in the Basu (1997) piecewise linear regression absent accounting influences.

7 Ball et al. (2013b) use the two-digit SIC code industry classification. We require each industry-year cross-section to have at least 10 observations and thus we use the Fama and French twelve industry classification to maximize the number of observations and limit data attrition in international datasets where the earnings expectation model is performed at the country-industry-year level. Nevertheless, our results remain unchanged when using the two-digit SIC industry classification.

8 Since the SCV is a non-linear parameter, we test the statistical significance of its differences across sub-samples through a non-linear combination of estimators using the delta method (Casella & Berger, 2002; Feiveson, 1999).

9 To the extent that the effect of IFRS adoption is found to be more prominent in Europe due to better legal enforcement (Christensen et al., 2013), we have also replicated André et al. (2015) who examine the change in CC following the mandatory IFRS adoption in 16 European countries. André et al. (2015) employ the C_Score measure of CC (Khan & Watts, 2009) and find a significant reduction in C_Score post-2005. Our re-examination of their inference confirms their findings and supports the alignment of AT with the MAT and SCV measures in interrupted time-series settings.

10 As mentioned earlier, for brevity and exposition purposes, we only report the coefficients on the variables of interest. We state the full regression equation in the notes of Table 3.

11 Given that Ahmed et al. (2013) controls interactively for the book-to-market ratio, we exclude the market-to-book ratio (i.e., the reciprocal of BTM) and its interactions from the MAT measure to avoid multi-collinearity.

12 We are grateful to Soeren Hvidkjaer for providing the PIN measure dataset on his website (https://sites.google.com/site/hvidkjaer/). LaFond & Watts’ (2008) footnote 10 indicates that they also acquired the PIN score from his earlier website (http://www.smith.umd.edu/faculty/hvidkjaer).

13 Ball et al. (2008) use the Global Vantage database, which has been succeeded by Compustat Global.

14 To illustrate, assume that a researcher is interested in comparing the level of CC exhibited by publicly listed firms in two different countries with different economic environments and institutional settings. Country A has Variance (X | RET<0) = 0.09 and Variance (X | RET>0) = 0.06, while Country B has Variance (X | RET<0) = 0.02 and Variance (X | RET>0) = 0.01; the SCV measure for Country A is 0.03 while that of Country B is 0.01, suggesting that Country A exhibits a higher level of CC. However, that is an unreliable conclusion given that the conditional variances of earnings in both countries are incomparable in magnitude due to high cross-sectional heterogeneity, and thus the ratio of conditional variances (RCV) should be used. When computing the RCV measure for Country A and Country B, the values are, respectively, 1.5 and 2, resulting in a contradicting conclusion to that inferred from using SCV.