Most portfolios are built on assumptions that data can test
Traditional allocation models treat markets as static. They don't account for shifting correlations, non-linear risk, or the compounding effect of small rebalancing errors.
These workshops give you hands-on experience with ML-driven approaches - from factor modelling in Python to reinforcement learning agents that rebalance under real constraints. The goal isn't a certificate. It's a working skill you can apply to an actual portfolio.
What the workshops cover
Each workshop focuses on a specific problem area within portfolio management. They're structured around exercises you complete during the session - not slides you watch and forget. You leave with notebooks, code, and a clear picture of where the technique fits.
Factor Model Construction
Build and validate multi-factor equity models using scikit-learn and Pandas. You'll work through data cleaning, feature selection, and cross-validation on real market data - not toy datasets.
Intermediate · PythonCovariance Estimation Under Noise
Sample covariance matrices are unreliable at high dimensions. This session covers shrinkage estimators, the Ledoit-Wolf method, and how estimation error propagates into allocation weights.
Advanced · StatisticsReinforcement Learning for Rebalancing
Set up a portfolio environment using OpenAI Gym conventions, then train a simple RL agent to make rebalancing decisions under transaction cost constraints. Practical and surprisingly tractable.
Advanced · RLAlternative Data Integration
Earnings call transcripts, satellite imagery, web traffic - alternative data is messy and lagged. This workshop walks through signal extraction, lag adjustment, and how to avoid look-ahead bias.
Intermediate · NLPRisk-Aware Backtesting
Most backtests flatter strategies. You'll learn how to stress-test using historical drawdown scenarios, apply walk-forward validation, and interpret results without overfitting to the past.
Foundational · All levelsCollaborative Portfolio Review
Bring a portfolio you're working on - real or hypothetical. Small group sessions with guided critique from peers and facilitators. Structured feedback, not open-ended discussion.
Peer-led · AppliedSession structure
Problem framing (30 min)
Each session opens with a specific, realistic problem - not a motivational overview. You'll see real data, real failure modes, and a clear question the rest of the session is built to answer.
Guided build (90 min)
Work through a notebook alongside the facilitator. Participants run code, hit errors, and fix them. The facilitator explains decisions as they're made - not in retrospect.
Independent variation (60 min)
You modify a parameter, swap a dataset, or extend the model in a direction of your choosing. This is where the learning actually sticks - when you make it break and then fix it yourself.
Group debrief (30–60 min)
Participants share what they changed and what happened. Facilitators highlight edge cases and connect the session's technique to broader portfolio management decisions.
Post-session materials
You receive the completed notebook, a short reference document, and access to a shared channel where participants continue the conversation between sessions.
