Outright bank failures without prior indication of nancial instability are very rare. Supervisory authorities monitor banks constantly. Thus, they usually obtain early warning signals that precede ultimate failure and, in fact, banks can be regarded as troubled to varying degrees before outright closure. But to our knowledge virtually all studies that predict bank failures neglect the ordinal nature of bank distress. Exploiting the distress database of the Deutsche Bundesbank we distinguish four dierent distress events that banks experience. Only the worst entails a bank to exit the market. Weaker orders of distress are, rst, compulsory notications of the authorities about potential problems, second, corrective actions such as warnings and hearings and, third, actions by banking pillar's insurance schemes. Since the four categories of hazard functions are not proportional, we specify a generalized ordered logit model to estimate the respective probabilities of distress simultaneously. Our model estimates each set of probabilities with high accuracy and conrms, rst, the necessity to account for dierent kinds of distress events and, second, the violation of the proportional odds assumption implicit in most limited dependent analyses of bank failure.Keywords: Bank, failure, distress, generalized ordered logit JEL: C35, G21, G33, K23, L50 Non-technical summary Bank insolvencies are dierent from failures of non-nancial rms because they may disrupt trust in the banking system as a whole. This explains why banks are subject to more regulatory scrutiny compared to most other industries. Supervisors therefore monitor banks on an ongoing basis to ensure the nancial safety and soundness of individual banks and the system. To this end, they use among other things o-site monitoring systems, which rely mostly on bank-specic nancial data reported to the authorities. These so-called hazard rate models predict the probability of default of individual banks on the basis of historical nancial and default data.Obviously, the denition of default is crucial to such analyses. On the one hand, banks are subject to clear-cut failure denitions, such as minimum capital requirements stipulated in laws and regulations. On the other hand, the systemic dangers of bank defaults in conjunction with an ongoing supervision process suggest that regulators, and practitioners, may regard a bank as troubled much earlier than ultimate default in the form of (forced) market exit occurs. We argue in this paper that most bank hazard studies neglect this ordinal nature of what we prefer to dene as distress rather than default.We address the issue by using data on a number of distress events that German universal banks from all three pillars experienced between 1994 and 2004. We categorize these events into four groups of ascending degrees of distress, ranging from early warning information issued by banks to regulators to absorbing events such as closure and restructuring mergers. To account for this pecking order of events, we specify a genera...