SVI 404 · Tier 400 — Graduate — Specialization & Capstone · 3 credits

Machine Learning & Frontier Models

Machine learning for volatility, anomaly detection, surface PCA, rough volatility, path signatures and Hawkes jumps.

Prerequisites

Learning Objectives

  • Frame next-day volatility forecasting as a supervised learning problem with an explicit feature matrix and a leak-free target.
  • Frame anomaly detection as a density problem: an event is anomalous because the model assigns it low probability under the learned distribution of normal market states.
  • Derive principal components as the eigenvectors of a centered covariance matrix and equivalently as the right-singular vectors of the data matrix via SVD.
  • Explain what it means for volatility to be 'rough' and estimate the Hurst exponent of a realized-volatility series.
  • Explain why a Poisson process cannot capture clustered jumps and how a Hawkes process fixes that through self-excitation.

Module Schedule

01Module 1: Machine Learning for Volatility Forecasting20 min02Module 2: Anomaly Detection: VolGAN & the Obsidian Screener20 min03Module 3: Surface PCA & Eigenmodes20 min04Module 4: Rough Volatility & Path Signatures20 min05Module 5: Self-Exciting Jumps: Hawkes Processes20 min