Discrete-Time Survival Model Based on Dynamic Industry Graph for Mortgage Credit Risk

To meet the modeling needs of cross-regional dependence in mortgage credit risk, the study proposes measuring economic proximity by industrial structure similarity and constructing a discrete-time survival model with a dynamic industry graph. The dynamic industry graph is reconstructed annually, with topology and edge weights adjusted as regional industrial structures evolve. This model combines a conditional autoregressive process and a first-order autoregressive process, estimated using integrated nested Laplace approximation. Backtesting based on US mortgage loan-level data shows that its in-sample fit and out-of-sample discrimination are superior to static industry graph models.