Generative Models for Financial Time Series

This unit covers advanced generative modeling techniques applied to financial time series, focusing on VAEs, autoregressive flows, continuous normalizing flows, and GAN extensions. Students will combine computed next‑day returns with Fama‑French factor data, customize models and LLM components for asset management workflows, and evaluate model performance and deployment considerations.

Learning Objectives

• Analyze and compare the performance of VAEs, autoregressive flows, continuous normalizing flows, and GAN extensions for modeling and simulating financial time series using quantitative evaluation metrics
• Apply continuous normalizing flows and autoregressive flow models to construct and train generative pipelines that produce realistic next‑day return scenarios using uploaded Fama‑French factor datasets
• Demonstrate integration of computed next‑day returns and Fama‑French factors into end‑to‑end generative model workflows and assess their impact on trading signal generation and backtested strategy outcomes
• Evaluate model customization, taxonomy, and deployment choices — including LLM adaptation, ethical/operational considerations, and model selection tradeoffs — to justify recommendations for asset management applications