Does Disaster Risk Relate to Banks’ Loan Loss Provisions?

Abstract

We examine the relation between disaster risk and banks’ loan loss provisions (LLP). We propose a disaster risk measure based on the natural disasters declared as major disasters by the Federal Emergency Management Agency over a 15-year span. We theoretically support and empirically validate our measure using three different approaches, including the UN Sendai Framework for disaster risk reduction, which relates disaster risk to natural hazard exposure, vulnerability and capacity, and hazard characteristics. Using more than 445,000 bank-quarter observations, we document that banks located in U.S. counties with higher disaster risk recognize larger LLP after controlling for other bank-level factors related to LLP. We employ several techniques to ensure the robustness of our findings, including difference-in-differences estimation and matched samples. In additional analysis, we explore the characteristics that better enable banks to recognize disaster risk in their LLP, and investigate the consequences of managing disaster risk through LLP. Our results are important, especially because of the increasing concern about disaster risk and because they inform the growing debate on the economic consequences of disaster risk and the ability of the banking system to proactively manage the resulting credit risk through LLP.

Keywords:

• Disaster Risk
• Loan Loss Provisions
• Future Charge-offs
• Banks

JEL codes:

• M40
• M41
• G20
• G21

Acknowledgements

We thank Beatriz Garcia Osma (Editor), two anonymous reviewers, Claudia Imperatore (discussant), Thomas Bassetti, Jochen Bigus, Pawel Bilinski, Brian Bushee, Minyue Dong, Peter Fiechter, Elaine Henry, Xiaoran (Jason) Jia, Anya Kleymenova, Arthur Kraft, Christian Leuz, Tim Martens, Araceli Mora, Luc Paugam, Inês Pinto (discussant), Changjiang Wang, Joanna Wu, Lei Zhang, and workshop participants at the 2020 AAA Joint Midyear Meeting of the AIS, SET and International Sections, 2020 Swiss Accounting Research Alpine Camp, 16th European Financial Reporting Network, University of Bamberg, Nanyang Technological University, Yeshiva University, CASS Business School, University of Bologna, and Free University of Bozen.

Disclosure statement

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

Data Availability Statement

The data are available from the sources identified in the text.

Supplemental Data and Research Materials

Supplemental data for this article can be accessed on the Taylor & Francis website, https://doi.org/10.1080/09638180.2022.2120513.

Appendix OA1. Validation of disaster risk measure using the German Watch Framework and the National Risk Index from the FEMA

Appendix OA2. Definition of variables used in validating disaster risk

Appendix OA3. Sample heterogeneity: Coarsened Exact Matching and Entropy Matching Analysis

Table OA1. Sample distribution by state and year

Table OA2. Disaster type distribution

Table OA3. Validation of disaster risk measure through the German Watch framework

Table OA4. Validation of disaster risk measure through the National Risk Index

Table OA5. Pearson’s correlation coefficients

Table OA6. Relation between disaster risk and loan loss provisions: Matched sample selection

Notes

1 The UN Sendai Framework for disaster risk reduction was adopted at the Third UN World Conference in Sendai, Japan, on March 18, 2015. The Sendai Framework replaced the Hyogo Framework for Action 2005–2015. Among others, the main objectives of the Sendai Framework are the improved understanding of disaster risk in all its dimensions, the strengthening of disaster risk governance, and the mobilization of risk-sensitive investment to avoid the creation of new risk.

2 We provide a more detailed discussion of the mechanism through which banks incorporate disaster risk in their LLP in Section 3.

3 https://www7.fdic.gov/sdi/main.asp?formname=customddownload.

4 In sensitivity analysis, we show that managers explicitly incorporate disaster risk into LLP and do not reflect the effects of a recent disaster in their estimations.

5 In 2020 the FASB replaced the incurred loss model with the expected loss model. Although we develop our hypothesis using arguments based on the incurred loss model, we note that a similar line of reasoning is applicable for the expected loss model.

6 https://www.fema.gov.

7 https://www7.fdic.gov/sdi/main.asp?formname=customddownload.

8 https://www7.fdic.gov/sod/dynaDownload.asp?barItem=6.

9 Our main analysis is robust to the inclusion of interstate banks in the sample.

10 The full dataset containing disaster risk and disaster risk determinants at the county level is available upon request from the authors.

11 We assess the reliability of the causal interpretation of our results by testing for the parallel trends assumption. We first assume a linear pre-treatment trend and then test this assumption using a local approach. Untabulated results show that treated and untreated firms exhibit the same pre-treatment linear trend and that no time-to-time difference arises between treated and untreated units during the pre-treatment period.

12 Full results are available upon request from the authors.

13 https://apps.bea.gov/regional/downloadzip.cfm.

14 The data in the Duke CFO Magazine Global Business Outlook survey, which is available at https://www.cfosurvey.org, starts in 2004. Therefore, our sample size is reduced to 308,125 bank-quarter observations for this test.

Additional information

Funding

Kiridaran Kanagaretnam thanks Social Sciences and Humanities Research Council of Canada (SSHRC) for its financial support [grant number 430-2020-00824].