Carolina Ballroom
Sponsored by Bank of America
Varying-coefficient Regression Models with Application to Stress Test (303429)
*Qiyi Lu, Wells FargoIn financial institutions, stress test models are built to translate different economic scenarios into losses, expenses or revenues, and then used to determine capital requirements under baseline and stressed economic conditions. In some stress test models, the dynamics being modeled could vary over multiple dimensions of time, such as vintage, calendar time and account age (month on book). It is of interest to account for these time dimensions simultaneously in order to produce a better model compared to traditional techniques that consider only one dimension. We propose a general framework of varying-coefficient model for pool level credit data. The proposed model framework allows coefficients to change smoothly over a time dimension such as loan age for each loan vintage, thus provides better interpretation on the maturation effect and macroeconomic conditions. It is a nonparametric framework and existing nonparametric smoothing techniques can be used to construct an estimation procedure. Empirical analyses on Indirect Auto stress test is used to illustrate the proposed method and to show the usefulness and advantage of the approach.