We investigate how provisioning models a¤ect bank regulation. We study an accuracy vs. timeliness trade-o¤ between an incurred loss model (IL) and a current expected credit loss model (CECL). Relative to IL, CECL improves e¢ ciency by enabling timely intervention to curb ine¢ cient ex post asset-substitution even though the imprecise information of CECL entails false alarms. However, from a real e¤ects perspective, our analysis uncovers a potential cost of CECL: banks respond to timely intervention by originating riskier loans so that timely intervention induces timelier risk-taking. By appropriately tailoring regulatory capital to information about credit losses, the regulator can improve the e¢ ciency of CECL. In particular, we show that regulatory capital under CECL would be looser when early estimates of credit losses are su¢ ciently precise and/or risk-shifting incentives are not too severe. From a policy perspective, our analysis suggests that better coordination between standard setters and bank regulators could enable the latter to relax capital requirements in order to spur lending.