Excess Returns The Alpha No Human Can Find | David Wright on Machine Learning's Hidden Edge
15 snips
Dec 17, 2025 David Wright, Head of Quantitative Investing at Pictet Asset Management, specializes in machine learning for investment strategies. He discusses how AI and machine learning enhance stock forecasting with features and decision trees. David contrasts machine learning with large language models, highlighting the importance of interpretability. He explains the role of human judgment in portfolio management and the necessity of robust data selection. Ultimately, he shares insights on feature construction, preventing overfitting, and the ideal investment horizons for machine learning.
AI Snips
Chapters
Transcript
Episode notes
ML Finds Feature Interactions
- Machine learning excels at finding which features to emphasize and how features interact to forecast short-horizon stock returns.
- These nonlinear combinations produce the core predictive power beyond simple factor weightings.
Define Clear Training Targets
- Train models on clear inputs and a precisely defined output like one‑month forward relative returns.
- Use decades of point-in-time features and rolling forward returns to let the algorithm learn mappings from inputs to outputs.
Why Trees Over LLMs For Forecasts
- Large language models generate text and lack the stability and interpretability needed for numerical return forecasts.
- Decision-tree-based models offer more stable, interpretable numeric predictions for investing tasks.