This study analyzes systematic and non-systematic credit risk in mortgage portfolios given US loan-level information by controlling for time-varying observable information in relation to the borrower, the collateral and the macro economy. The total risk in relation to rating class default rates is decomposed into systematic and class-specic non-systematic risk by a state space model. The paper finds that the total risk relates to credit quality in a smile-shaped pattern: systematic risk is negatively related and non systematic risk is positively related to average default rate levels. In addition, total risk increases during and after the Global Financial Crisis. The impact of the crisis on systematic risk is persistent whereas the impact on non-systematic risk appears to be temporary. The analysis of regulatory capital suggests that mortgage risk models in conjunction with periodic updating warrant a sucient level of regulatory capital given the current regime. These findings are relevant to prudential regulators who are currently discussing the implementation of a monotone relationship between default probabilities and asset correlations under Basel III.