Generative AI Can’t Generate Alpha… But Machine Learning Can | Pictet’s David Wright

Summary

David Wright explains Pictet's machine learning investment process using decision trees and gradient boosting, contrasting it with generative AI. He details the two ETFs (PQNT, PQUS) that aim for 1–2% annual alpha with beta-1 risk. The conversation also touches on AI infrastructure spending expectations being overblown due to efficiency gains.

• Pictet uses gradient-boosted decision trees on 400+ features to forecast 20-day stock returns.
• Generative AI is rejected for alpha generation due to hallucinations and look-ahead bias.
• PQNT (MSCI EAFE benchmark) and PQUS (S&P 500 benchmark) are designed as passive replacements with 1–2% target outperformance.
• The model is retrained every three months and uses features from price trends, accounting changes, analyst revisions, and calendar effects.
• Portfolio construction maintains beta of 1 and risk-controlled active positions.
• Wright expresses skepticism about the huge expected spend on AI chips and cloud, citing efficiency gains in model training.
• The strategy has been running in Europe for 2–3 years before the US ETF launch.