The app for independent voices
Deep Momentum just crushed 45 stock markets.
A new paper reveals that momentum stocks across the globe share a hidden flaw: a bimodal return distribution. Past winners don't just stay winners — they either soar or crash. Same for losers. This "bimodality" is why momentum fails in Asia and is the primary channel linking behavioral biases to momentum profits.
The fix? Deep Momentum — a machine learning strategy that accounts for bimodality by predicting the full return distribution and reclassifying stocks by expected return, not the modal class.
Tested across 45 countries (2010–2023):
Traditional momentum: 20.9% return, 0.98 Sharpe
Naïve XGBoost: 31.1% return, 2.13 Sharpe
Deep Momentum: 41.4% return, 2.49 Sharpe — positive returns in every country tested
Korea: MOM –0.2% → Deep Momentum +46.1%
China: MOM +2% → Deep Momentum +31.5%
The kicker? A global model trained on all 45 countries simultaneously — a joint learning approach — achieves a Sharpe ratio of 8.98 (equal-weighted) and 2.81 (value-weighted). It outperforms country-specific models in 38 of 43 countries, solving the small-data problem in emerging markets.
Behavioral biases drive everything: low individualism and weak disposition effects → high bimodality → momentum crashes. Machine learning steps in exactly where human behavior fails.
