Please note:

Speaker change for the upcoming November 17th seminar

A Kernel Machine for Large-Scale

Corporate Default Modeling

A Seminar by Chao Zhang

Tuesday November 17, 2026

6:00 PM Seminar Begins

7:30 PM Reception

Hybrid Event

Fordham University

McNally Amphitheater

140 West 62nd Street

New York, NY 10023

Free Registration!

For Virtual Attendees: Please select virtual instead of member type upon registration.

Abstract:

We present a scalable kernel logistic regression framework for estimating forward-looking, real-world corporate default probabilities using features derived from financial variables and structural credit-model outputs. This kernel machine is used in the Bloomberg Corporate Default Risk Model (DRSK). The model learns a globally defined nonlinear log-odds function in a reproducing kernel Hilbert space. The chosen hybrid linear–Gaussian radial-basis-function kernel is universal and capable of approximating a broad class of nonlinear dynamics. Under standard conditions and an appropriate regularization scheme, the associated learning procedure is universally consistent, with prediction risk converging to the Bayes risk as the sample size increases.

The application involves large-scale, highly imbalanced default data spanning multiple regions, sectors, and credit cycles. To address the computational cost of conventional kernel machine training, we combine majority-class down-sampling with selection-bias correction and select model complexity through cross-validation and Bayesian optimization. Empirical results demonstrate improved calibration and out-of-sample predictive performance relative to conventional logistic regression. The results show that universal kernel methods can be successfully deployed for large-scale, rare-event corporate default modeling.

Bio:

Chao Zhang is a Quantitative Researcher on Bloomberg’s Quant Risk Analytics team. His work focuses on the development of machine learning and quantitative models for a wide range of applications in finance and risk analytics, including credit, market, and climate risk. He earned his PhD in Mathematical Statistics from the University of California, Santa Barbara, where his research encompassed functional and distributional data analysis, as well as statistical and machine learning.