Accounting Bias and Debt Contracting

Abstract

This study investigates the effects of accounting bias on the efficiency of a debt-financed investment, which is subject to an interim liquidation/continuation decision based on the accounting report. We decompose the overall effect of accounting bias into a mean effect and a variance effect and transform binary state and report spaces into continuous ones. We derive two main results. (1) The highest investment efficiency is induced by a downward bias for gloomy investment prospects and an upward bias for rosy investment prospects. (2) The optimal covenant tightness is U-shaped and the optimal interest rate is hump-shaped in the degree of bias. In particular, neutral (unbiased) accounting neither minimizes the variance of the report nor maximizes investment efficiency.

Keywords:

• Neutrality
• Bias
• Investment efficiency
• Covenant tightness
• Interest rate

Acknowledgments

We wish to acknowledge the helpful comments of Robert Göx (the editor) and two anonymous referees whose suggestions led to a much better manuscript. We have also benefitted from helpful comments of participants at the 2022 AAA Midwest Region Meeting and the 2023 AAA FARS Midyear Meeting.

Disclosure statement

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

Supplemental Data

Supplemental data for this article can be accessed online at http://dx.doi.org/10.1080/09638180.2023.2248620.

Data Availability

All data are from publicly available sources identified in the study.

Notes

1 The first approach is adopted by Gigler and Hemmer (2001), Venugopalan (2001), Chen et al. (2007), Suijs (2008), Gigler et al. (2009), Göx and Wagenhofer (2010), Drymiotes and Hemmer (2013), Gao (2013), Li (2013), Nan and Wen (2014), Bertomeu and Cheynel (2015), Armstrong et al. (2016), Callen et al. (2016), Jiang (2016), Bertomeu et al. (2017), Caskey and Laux (2017), Jiang and Yang (2017), Lu et al. (2018), and Dordzhieva et al. (2022), among others. The second approach is adopted by Guay and Verrecchia (2006), Burkhardt and Strausz (2009), Gow (2009), Göx and Wagenhofer (2009), Caskey and Hughes (2012), and Beyer (2013), among others. In a different strand, Kwon et al. (2001), Fan and Zhang (2012), and Jiang and Yang (2021) use a binary partition of a state space to represent the degree of bias.

2 See Burkhardt and Strausz (2009), Caskey and Hughes (2012), Beyer (2013), Gao (2013), Li (2013), Callen et al. (2016), Jiang (2016), Bertomeu et al. (2017), and Jiang and Yang (2021), among others.

3 We can generalize this assumption as $Pr({\alpha }_{\theta }\ge t)\times v+Pr({\alpha }_{\theta }<t)\times 0>m$, where the success/failure threshold is $t\le \mathbb{E}\left[{\alpha }_{\theta }\right]$. Our main result (Proposition 3) holds qualitatively in this general case. Results are available on the online appendix. If $t>\mathbb{E}\left[{\alpha }_{\theta }\right]$, aggregating the individual projects into a megaproject makes it more difficult to succeed than running the individual projects separately; that is, it is value-destroying to form a megaproject. Hence, we assume that if $t>\mathbb{E}\left[{\alpha }_{\theta }\right]$, the firm would not form a megaproject in the first place.

4 In the example of Apple Watch introduced in the preceding section, the report $y_i$ for subproject i (say, Bluetooth 5.3) will be 1 if the technical feasibility is exceeded, and 0 otherwise. The accounting criteria for the technical feasibility will affect how the state of the subproject is represented, conservatively or aggressively. The aggregate report is an aggregation of the individual reports for the four subprojects involved in Apple Watch 8.

5 Under IFRS and U.S. GAAP, software development costs can be capitalized if certain criteria are met, most importantly the establishment of technical feasibility. Recognition of such costs as an asset following the industry norm is neutral accounting (represented by k = 0 in our model). Recognition at a later date causes a downward bias of current earnings (represented by k>0) whereas recognition at an earlier date causes an upward bias (represented by k<0).

6 When we drop the requirement of having at least 3 loan facilities for each firm, our sample consists of 13,389 observations from 2211 firms. The regression results are qualitatively the same.

7 As a robustness check, we include $Cscor{e}^3$ (Cscore cubed) in the regression and the coefficient on it is not significant, indicating that the main finding in Table 2 indeed indicates a U-shaped relation.

8 As a robustness check, we include $Cscor{e}^3$ (Cscore cubed) in the regression and the coefficient on it is not significant, indicating that the main finding in Table 3 indeed indicates a hump-shaped relation.