Diffusion Models for Structured Financial and Economic Data: From Factors to Adaptive Sequences

RENYUAN XU – STANFORD UNIVERSITY 

ABSTRACT

Generative AI is becoming increasingly central to finance and economics, advancing applications such as stress testing, risk analysis, and robust portfolio construction in high-dimensional settings. Among the many approaches, diffusion models offer a principled and tractable framework whose theoretical guarantees guide efficient implementation. This talk explores how diffusion models can be adapted to address two fundamental challenges in financial and economic data generation: high-dimensional factor structures and adaptive sequential dynamics. A unifying theme across both works is the integration of domain-specific statistical structures into diffusion frameworks in order to overcome classical limitations—namely the curse of dimensionality in cross-sectional data and the preservation of temporal adaptiveness in time-series settings.