Learning Program — Domain
Machine Learning in Market Volatility
Analysis
Statistical markets generate noise. Understanding which signals persist under volatility pressure — and which dissolve — is what this program is built around.
Program Structure
What the program covers
Participants work through applied problems using real equity and index data. The program focuses on how regression models, gradient boosting, and recurrent architectures behave differently as volatility regimes shift.
Identifying regime transitions using GARCH family models. Rolling window statistics, stationarity testing, and their practical limits on equity time series.
Constructing lagged return features, realized volatility measures, and cross-asset spread signals. Avoiding look-ahead bias in pipeline design.
Why linear models degrade in high-volatility windows and how tree-based ensembles handle distributional shift — with worked examples on S&P 500 intraday data.
LSTM and attention-based models for volatility forecasting. Walk-forward validation versus k-fold — when each applies and why the distinction matters for production readiness.
Measuring model quality beyond accuracy: Diebold-Mariano tests, calibration under stress, and the gap between backtested and live performance.
Program Instructors
Worked in risk modelling for institutional asset managers. Teaches the volatility regime and feature engineering modules with emphasis on replicable methodology.
Background in applied ML at a Canadian financial data firm. Leads the sequence models and evaluation modules, focusing on the gap between research conditions and live deployment.