Generative Models for Financial Time Series

This capstone unit integrates advanced generative architectures (VAEs, GANs, autoregressive and flow-based models) with practical trading workflows, including computing next-day returns and conditioning on Fama–French factors. Learners will customize LLMs for asset-management tasks, implement end-to-end pipelines for sequential financial data, and evaluate model fitness for trading and risk decisions.

Learning Objectives

• Analyze and compare VAE, GAN, autoregressive, and flow-based (including continuous normalizing) architectures for sequential financial time series, identifying assumptions, strengths, and limitations for next-day return simulation.
• Apply flow-based and generative model extensions to build a reproducible pipeline that uploads Fama–French factors, computes next-day returns, conditions models on factor inputs, and generates synthetic scenario data.
• Demonstrate customization of LLMs and prompt/model-adaptation techniques to automate trading workflows, integrate model outputs with trading features, and complete applied exercises documenting the process.
• Evaluate model outputs and trading implications using quantitative metrics (predictive accuracy, distributional similarity, tail-risk measures) and produce evidence-based recommendations for model selection, validation, and risk-aware deployment in asset management.