Unpredictability Creates Alpha: Stock Picking

Key Takeaways

• AI predicts 71% of fund managers' stock trades
• Predictable, style‑aligned stocks tend to underperform
• Unpredictable stocks deliver higher future returns
• Crowded trades erode alpha for value and growth funds
• Embracing uncertainty can boost portfolio performance

Pulse Analysis

The integration of artificial intelligence into fund‑holding analysis marks a turning point for quantitative finance. Cohen and his co‑authors compiled a comprehensive dataset of quarterly positions for every U.S.‑domiciled equity fund over a 33‑year span, then trained a machine‑learning model on macro variables, fund characteristics, and past holdings. Achieving a 71% accuracy rate in forecasting buys, holds, or sells, the model demonstrates that fund behavior is far more systematic than anecdotal market lore suggests, offering a new lens for investors seeking data‑driven edge.

Beyond the technical feat, the paper uncovers a counterintuitive performance pattern: stocks that fit neatly into a fund’s stated style—clear‑cut value, growth, or quality picks—are the most predictable yet consistently lag behind Treasury bills and generate negative four‑factor alpha. This crowding effect reflects a classic market‑efficiency paradox where the collective conviction of many managers compresses expected returns. In contrast, the less predictable 29% of trades—often ambiguous, borderline cheap, or distressed securities—outperform, indicating that uncertainty itself can be a source of excess return. The results echo behavioral finance insights that investors overpay for familiar, low‑risk narratives while undervaluing complex opportunities.

For practitioners, the study suggests a strategic pivot: rather than chasing the safest, most obvious style matches, portfolio managers should allocate capital to stocks that elude easy classification. This may involve deeper fundamental research, alternative data, or contrarian positioning in sectors with higher information asymmetry. Risk controls remain essential, but the reward potential lies in the “unknown unknowns” that AI models flag as low‑predictability. As machine learning continues to refine its predictive power, the ability to identify and exploit these uncertain pockets could become a defining competitive advantage in active equity management.