Do Auditors Respond to Clients’ Climate Change-related External Risks? Evidence from Audit Fees

Abstract

In this study, we investigate whether auditors consider their clients’ climate change-related external risks when making audit pricing decisions. Using county-level proxies based on the number of declared natural disasters and the level of societal climate change awareness, we discover that clients with greater exposure to climate change risks pay significantly higher audit fees. After performing several additional tests, we conclude that auditors consider climate change risks and their potential consequences as a systematic business risk that is factored into the audit fees. For instance, we demonstrate that clients’ climate risk exposure has become more strongly associated with audit pricing in recent years, as climate change has gained greater importance in public debate. Moreover, we discover that auditors place a greater emphasis on clients’ climate risks when they themselves are located in regions with higher climate change awareness, indicating that auditors’ climate change perception also matters. Given the growing interest in climate change-related risks in practice and research, as well as the significance expected to be placed on these risks in the future, our findings are timely and should appeal to a wide range of readers, including investors, regulators, and scholars.

Keywords:

• Audit fees
• Audit risk
• Climate change
• Climate risk
• External factors

JEL codes:

• M42
• Q54

Acknowledgements

We thank Beatriz García Osma (editor) and two anonymous referees for their valuable comments and guidance. We also thank Max Göttsche, David Hay, Seppo Ikäheimo, Lasse Niemi, Hannu Ojala, Nicole V. S. Ratzinger-Sakel, Frank Schiemann, and seminar participants at the 26th International Symposium on Audit Research (ISAR 2022), 83rd VHB Annual Conference, Aalto University, University of Göttingen, University of Hamburg, University of Innsbruck, and University of Ingolstadt for valuable comments and suggestions. Part of this research was conducted while Hartlieb was visiting Deakin University. All errors remain our own.

Data Availability

Data are available from the public sources cited in the text.

Disclosure Statement

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

Supplemental Data and Research Materials

Supplemental data for this article can be accessed online at https://doi.org/10.1080/09638180.2022.2141811.

Online Appendix A. Variable Definitions.

Online Appendix B. Distribution of Average Climate Risk and Climate Change Awareness by County.

Online Appendix C. Interaction between Physical Climate Risk and Climate Change Awareness.

Online Appendix D. Robustness Tests.

Notes

1 We can only consider a smaller sample for the analyses with the annual climate change awareness measure, which is only available for the period 2014–2018 (13,729 firm-year observations).

2 See Auditing Standards (AS) Nos. 1101, 2101, 2110, 2201, 2301.

3 According to Asante-Appiah & Lambert (in-press), external auditors may even assist clients in improving their ESG performance.

4 External risks are those factors that are beyond a firm’s control but can have a significant impact on its success. Firms, for example, may face costly litigation claims in regions with a high level of climate change awareness, despite actual compliance with existing environmental regulations. For a literature review on external factors and auditing, we refer to Eierle et al. (2022).

5 Using a three- or seven-year window does not alter our primary conclusions (see Online Appendix D). Following Dal Maso et al. (in-press), we remove declarations that have the same date, title, and place code to avoid double counting.

6 Most recent data for 2020 is available at https://climatecommunication.yale.edu/visualizations-data/ycom-us/. The survey results are based on responses obtained from over 25,000 participants.

7 As a robustness test (untabulated), we consider the question ‘Do you think that global warming will harm future generations?’ instead of the question about climate change harming people in the US, and the results remain largely unchanged.

8 Online Appendix A presents the definitions of all variables used in the study. We use client-year as our unit of analysis; we remove the subscripts for expositional ease.

9 We thank Chris Pantzalis for providing us with the data for the political alignment index.

10 As a prominent example, former president Donald Trump rejected California’s disaster request for wildfires in October 2020.

11 As a robustness test, we include county-fixed effects and our results remain unchanged (Online Appendix D).

12 In some cases, for example, we were unable to determine certain clients’ headquarter county based on ZIP codes, resulting in missing county variables. Furthermore, some county controls, such as the social capital measure, are unavailable for clients located in Hawaii, the Virgin Islands, and Puerto Rico.

13 We present a figure in Online Appendix B that visualizes the geographical distribution at the county-level and shows that our 558 sample counties are evenly distributed across the US. However, our sample does not cover all counties. This is due to the geographical distribution of the Compustat universe rather than specific data restrictions for some of our sample variables (e.g., the climate change-related variables). The figure in Online Appendix B also provides some insights into the variation of physical climate risk and climate change awareness at the county level, demonstrating that counties with a higher climate change awareness and greater exposure to physical climate risk are concentrated in US coastal areas.

14 Continuous variables are winsorized at 1% tails. Furthermore, some data for the county-level variables (e.g., $\mathrm{Social}\mathrm{\_Capital}$ or $\mathrm{Religion}$) are only available for specific years. Following previous research (e.g., Callen and Fang, 2018; Leventis et al., 2018), we use linear interpolation or extrapolation to obtain the missing values. Using time-invariant measures such as the average value does not change our main conclusions.

15 The robustness of these findings is also proven by a variety of sensitivity tests (see Online Appendix D).

16 According to Jha and Chen (2015), $\mathrm{Big}4$ becomes insignificant when it is also controlled for $\mathrm{County}\mathrm{\_Presence}$, which is the natural logarithm of the sum of audit fees collected from an audit firm in a county. Thus, the effect of $\mathrm{Big}4$ is absorbed by $\mathrm{County}\mathrm{\_Presence}$. It is noteworthy that $\mathrm{Big}4$ becomes positively significant if we exclude $\mathrm{County}\mathrm{\_Presence}$ as an explanatory variable.

17 The agriculture industry is based on the Fama/French 48-industry classification ‘Agriculture’, the coal mining industry based on ‘Coal’, and construction industry is based on ‘Construction Materials’ and ‘Construction’. Because agriculture and coal mining comprise only one group, no industry-fixed effects are considered here.

18 This is discussed by the European Commission: https://eur-lex.europa.eu/resource.html?uri=cellar:c851d397-9584-11ec-b4e4-01aa75ed71a1.0001.02/DOC_2&format=PDF.

19 Because of the already low number of observations for this industry tests, we use the time-invariant climate change awareness measure which can be applied to the full sample period and not only the period 2014–2018.

20 Please refer to: https://industrytoday.com/5-industries-benefiting-from-climate-change/.

21 This demonstrates the importance of employing a more comprehensive physical climate risk measure that does not focus solely on drought conditions (Truong et al., 2020a). It is noteworthy that our ‘heat’ climatological variable, which is based on the FEMA database, almost entirely contains declared fires as droughts are usually not declared because they have no material impact. We also consider an alternative database for natural disasters from the United States Department of Agriculture (USDA), which includes various types of natural disasters but focuses on droughts (correlation coefficient between the FEMA and USDA climate risk measure: approximately 20%; full test results are not presented here). Since declaration hurdles (and corresponding aids) are lower, this database contains a greater number of disasters than FEMA. However, these data are only available since 2012. Using a one-tailed test, we discover that this measure has a weak positive impact on audit fees. Online Appendix D presents more information on this test.

22 When we combine all individual climate change awareness items in a single regression, VIFs exceed the critical level of ten, indicating that multicollinearity is a concern and the results must be interpreted with caution.

23 Data for this test is publicly available at https://sites.google.com/site/scottdyreng/Home/data-and-code/EX21-Dataset. Since the data is only available up to 2014, our sample period for this additional test ends in 2014.

24 However, a greater geographical diversion can also be contended to raise the client’s overall risk of being affected by any natural disaster, particularly if production plants are located in regions with a higher physical climate risk, whereas the headquarters is located in a low climate risk region. However, we are not able to identify the region where potential subsidiaries are located, which is a shortcoming of our analysis.

25 Because data on subsidiaries is only available until 2014, we consider a time-invariant climate change awareness measure that can be applied throughout the sample period.

26 Detailed results of this test are presented in Online Appendix C.