The effect of natural disasters on big bath earnings management of banks: evidence from the 2005 US hurricane season

AbstractWe investigate the effect of the 2005 US hurricane strikes on big bath earnings management in the banking industry. Using a difference-in-differences approach, we find that banks affected by the hurricanes add more discretionary loan loss provisions (DLLP) after the shock relative to unaffected banks. Further tests suggest that the hurricane-induced DLLP increase is attributable to opportunistic big bath accounting rather than to a precautionary motive. We also show that the effect of the hurricanes on big bath earnings management is more pronounced in banks that were managing earnings upwards more aggressively before the hurricanes.Keywords: big bath earnings managementbankingdiscretionary loan loss provisionsnatural disastersJEL classification: G21M41 AcknowledgementsWe are grateful to Juan Manuel García Lara (the editor) and two anonymous referees for their valuable and constructive suggestions. We also appreciate comments and suggestions from seminar participants at Southwestern University of Finance and Economics.Disclosure statementNo potential conflict of interest was reported by the author(s).Data availability statementThe data that support the findings of this study are available from FEMA and the Statistics on Depository Institutions database of the FDIC. These data were derived via https://www.fema.gov/disasters and https: //www.fdic.gov/bank/statistical/.Notes1 Ng and Roychowdhury (Citation2014) point out that the increase in LLPs can increase a bank’s total regulatory capital under certain conditions. We detail these conditions in Section 4.3.1.2 In Section 3.2, we detail the definition of the treatment group (affected banks) and the control group (unaffected banks).3 Following Schüwer et al. (Citation2019), non-performing loans are measured as total assets past due 90 or more days and still accruing interest. Unreported findings indicate that the results are robust when we compute non-performing loans as the sum of total assets past due 30 days or more and still accruing interest and total assets no longer accruing interest.4 In Section 3.3, we detail the procedure of PSM.5 In Section 3.3, we detail the selection of the sample period.6 Data are acquired via https://www.fdic.gov/bank/statistical/.7 Data are acquired via https://www.fema.gov/disasters.8 According to the National Bureau of Economic Research, the 2001 recession ended in November 2001, and the 2008 financial crisis started in December 2007. Thus, we limit the sample period to Q1 2002–Q3 2007 to avoid the overlap with two recessions. In the robustness section, we shorten the sample period and the results remain similar.9 Specifically, Size equals the natural logarithm of total assets. Loan (Deposit, CAP) equals total loans (total deposits, the value of equity) divided by total assets. ROA is the return on assets.10 We only include the interaction term in our regressions because we focus on future performance improvement of those treatment banks increasing DLLP in response to the hurricanes. Untabulated findings show that the results are robust when we merely use the sample of the treatment group and examine the effect of DLLP_Pos on ΔROAt+4 and ΔROAt+8, respectively.11 We only consider upward earnings management for four quarters prior to the hurricanes, because upward earnings management before this period might be affected by the reversion of abnormal provisions from previous periods. Therefore, relative to earlier periods, recent upward earnings management is more likely to be reversed by big bath accounting during the hurricanes.12 We obtain the data on realised and unrealised gains and losses on AFS securities from the Bank Regulatory Database of Wharton Research Data Services.13 Untabulated findings indicate that the results are robust to the period of three and four quarters before and after the hurricane season.14 The Home Mortgage Disclosure Act (HMDA) data files provide information about the amount of banks’ mortgage loans extended to each county.15 Untabulated findings indicate that our results remain robust when we employ total assets, total loans, and return on assets as matching variables.