北京雁栖湖应用数学研究院

This course introduces deep generative models and their applications in quantitative finance, focusing on their ability to generate realistic, high-dimensional market data where traditional parametric models often fail—particularly in the tails and in data-limited stress scenarios. Emphasis is placed on combining modern generative AI with core financial mathematics, ensuring consistency with probabilistic foundations, stochastic processes, and no-arbitrage principles.

The course reviews essential probability, asset-pricing, and deep-learning concepts before covering the main generative model families—generative adversarial networks, normalizing flows, and diffusion/score-based models—and the practical challenges of training, conditioning, and evaluation in financial settings. These tools are applied to key problems including return and scenario generation, portfolio construction, risk-neutral density estimation, derivatives/surrogate pricing and calibration, stress testing and risk management, and market microstructure and order-book simulation.

By the end, students will be able to design and assess deep generative models as financially consistent predictors, simulators, and surrogates for modern quantitative finance applications.