Bay Area Data Science Summit · First Place
MLP for Optimal Equity Derivative Hedging
Feb 2025 · UC Berkeley
An exposure-aware neural pipeline for reducing the premium cost of equity derivative hedging.
Problem
Equity derivative hedging must balance premium cost, changing market conditions, and a required exposure target. Static rules struggle when the feature space is high-dimensional and volatility changes over time.
Pipeline
The project combined three layers:
- causal time-series alignment, missing-value treatment, and PnL/exposure construction;
- smooth rank-preserving targets derived from exposure-adjusted returns; and
- a multilayer perceptron trained with stratified cross-validation.
A modular trading engine then used model scores to prioritize instruments and adapt sell orders while tracking the portfolio’s exposure constraint.
Results
The final strategy reduced test-set premium cost by 64.5% relative to the baseline while maintaining the $10M exposure target. The project won first place at the UC Berkeley competition endorsed by Wells Fargo.
My role
I led the international team, coordinated modeling and evaluation, and developed the learning and decision pipeline that connected market features to executable hedging actions.