Does the Texas Ratio Predict US Bank Failures? An Out-of-Sample Test on 8,466 Banks, 2007 to 2013, and the Distribution in June 2026

The Texas ratio divides a bank's noncurrent loans and repossessed real estate by its tangible equity plus loan loss reserves. Analysts treat a value above one as a danger sign. I test it out of sample: the ratio for 8,466 US banks at the end of 2007, from FDIC call reports, against the 488 of them that failed between 2008 and 2013. I prove that the buffer is exhausted exactly when the loss rate on problem assets exceeds one over the ratio, and that the area under the ROC curve equals the Mann-Whitney probability and so does not change under any strictly increasing transformation of the ratio. The 2007 ratio ranks banks well: the AUC is 0.739 (bootstrap 95% interval 0.714 to 0.765), above the noncurrent asset ratio (0.732), the tangible equity ratio (0.605) and the Tier 1 leverage ratio (0.602). Failure rates rise from 2.6% for ratios below 0.05 to 40.4% between 0.5 and 1 and 54.8% above 1. But only 17 of the 488 failures had a 2007 ratio above one; the median failed bank would have needed losses of 584% on its 2007 problem assets to exhaust its buffer, so the ratio worked by ranking early deterioration, not by measuring imminent insolvency. Adding bank size raises the cross-validated AUC to 0.758. In June 2026 the median ratio across 4,206 banks is 0.032, against 0.053 in 2007, and 7 banks have a finite ratio of at least one, against 30.