Generative Models for Financial Time Series

This capstone unit teaches advanced generative modeling techniques for trading and asset management, focusing on VAEs, GANs, autoregressive and continuous normalizing flows applied to financial time series. Students will integrate Fama–French factors with computed next‑day returns, customize models (including LLM adaptation), run coding exercises, and evaluate model performance and deployment considerations.

Learning Objectives

• Analyze and compare generative model architectures (VAEs, GAN extensions, autoregressive flows, and continuous normalizing flows) for modeling and simulating financial time series, identifying strengths and limitations for next‑day return prediction
• Apply data integration and preprocessing workflows to upload Fama–French factors, compute next‑day returns, and create training datasets and features suitable for flow‑based and GAN‑based models
• Demonstrate the implementation and customization of flow‑based and GAN/ VAE models (including continuous normalizing flows and autoregressive flows) and adapted LLM components by completing practical coding exercises and producing reproducible model outputs
• Evaluate model performance and trading implications through backtesting, statistical assessment, and discussion of risk, governance, and operational considerations for deploying generative AI in asset and portfolio management