Credit booms are not rare and usually precede financial crises. However, some end in a crisis (bad booms) while others do not (good booms). We document that credit booms start with an increase in productivity, which subsequently falls much faster during bad booms. We develop a model in which crises happen when credit markets change to an information regime with careful examination of collateral. As this examination is more valuable when collateral backs projects with low productivity, crises become more likely during booms that display large productivity declines. As productivity decays over a boom as an endogenous result of more economic activity, a crisis may be the result of an exhausted boom and not necessarily of a negative productivity shock. We test the main predictions of the model and identify the component of productivity behind crises.