Did Fast Loan Growth Before 2008 Predict Which US Banks Failed? 7,914 FDIC-Insured Banks, Loan Growth 2004-2007, Failures 2008-2013, an Exact Attributable-Fraction Identity and Charge-Offs Among Survivors

Rapid loan growth is a standard warning sign for bank supervisors. I measure how well it worked in the last US banking crisis using public FDIC data for 7,914 banks that filed reports in December 2004 and December 2007. Loan growth is the annualised log change in gross loans and leases between the two dates, and the outcome is failure between January 2008 and December 2013 (434 failures in the sample). The failure rate was 14.47% in the top fifth of loan growth (above 15.5% a year), against 1.71% to 5.18% in the other four fifths. The relation is U-shaped at the bottom: banks whose loans shrank failed more often (3.22%) than those growing 2% to 6% a year (1.71%). The risk ratio of the top to the bottom fifth is 4.49 (95% interval 3.34 to 6.03). I prove an exact identity for the attributable fraction, which shows that 52.8% of failures, about 229 banks, were in excess of the rate among the slower-growing 60% of banks. In a logit that controls for equity, real estate concentration, size and the loan-to-asset ratio, each extra point of annual loan growth raised the failure probability by 0.103 points on average, and adding loan growth lifted the in-sample AUC from 0.787 to 0.805. Among banks that survived to file in 2010 and 2011, the median net charge-off rate rose from 0.28% in the slowest-growing fifth to 0.89% in the fastest. Because failed banks drop out, this survivor comparison understates the damage.