Business Case
Algorithmic trading strategies have dozens of hyperparameters. Grid search is computationally expensive and doesn’t scale. We needed a smarter approach to find optimal configurations for our portfolio management algorithms.
Impact
- Contributed to 400%+ annual returns on a $2M portfolio
- Featured in major press: Cronista and CriptoNoticias
- Near-optimal configurations in far fewer evaluations than grid or random search
- Team: Part of the 5-PhD team at Xcapit
Approach
Applied Bayesian optimization using Gaussian Processes to efficiently search the hyperparameter space. Tested multiple strategies including VWAP vs SMA, Three Standard Moving Average, Chandelier Exit, and Bollinger Bands Volume.