Generative Modeling and Data Engineering

This unit immerses students in translating legacy trading code to Python, engineering and featurizing market and textual datasets, and implementing advanced generative and hybrid models (VAE, flows, GAN/WGAN) for trading and asset-management tasks. Learners will fit, tune, quantify, and compare models (including model quantization and improved GAN training techniques), run experiments for strategies such as options term-structure arbitrage and sentiment-driven signals, and prepare results for deployment considerations.

Learning Objectives

• Demonstrate the ability to translate MATLAB trading and analytics scripts into reproducible Python implementations, validating functional equivalence with unit tests and benchmark comparisons
• Apply data engineering and feature-engineering workflows to ingest, clean, align, and label market, options, and textual sentiment data for model training, producing documented datasets ready for experiments
• Analyze and compare generative and hybrid model architectures (VAE, coupling flows, GAN/WGAN, hybrids) by training and tuning models, reporting quantitative experiment results using performance and risk metrics (e.g., MSE, AUC, Sharpe, drawdown)
• Evaluate model optimization and deployment trade-offs—including model fitting diagnostics, GAN training improvements, and model quantization—by measuring robustness, inference latency, and performance impact for asset-management workflows