The strategy uses anomaly features, risk premia, and biased expectations to rank stocks, creating monthly value-weighted long-short portfolios via machine learning models retrained on a rolling three-year window.
ASSET CLASS: stocks | REGION: United States | FREQUENCY:
Monthly | MARKET: equities | KEYWORD: Stock
I. STRATEGY IN A NUTSHELL
Combines 27 stock anomalies, risk premia betas, and analysts’ earnings forecasts using machine learning models (Random Forests, SVMs, Gradient Boosted Trees) to predict one- to three-quarter-ahead earnings. Stocks are ranked by predicted probabilities and assigned to monthly value-weighted long-short portfolios, retrained on a rolling three-year window.
II. ECONOMIC RATIONALE
The strategy exploits mispricing from behavioral biases, particularly analysts’ overoptimistic forecasts, alongside well-known anomalies like momentum and book-to-market. Integrating anomalies, fundamentals, and macro factors via ML systematically identifies mispriced stocks, enabling contrarian or arbitrage opportunities beyond traditional factor models.
III. SOURCE PAPER
Risks versus Mispricing : Decomposing Asset Pricing [Click to Open PDF]
Han, X, City University London – Bayes Business School
<Abstract>
I use classification-based machine-learning methods to decompose 32 anomaly payoffs
into risk exposures and mispricing. The component driven by risk earns statistically insignificant
returns, despite its efficacy in explaining the time-series variation in anomaly payoffs.
The mispricing component is driven by biased expectations and earns significant returns that
also subsume anomaly payoffs. These findings indicate that the unconditional averages of
anomaly returns can be fully explained by biased expectations, whereas risk exposures play
an important role in explaining the time-series variation in anomaly returns.
IV. BACKTEST PERFORMANCE
| Annualised Return | 23.87% |
| Volatility | 17.38% |
| Beta | N/A |
| Sharpe Ratio | 1.37 |
| Sortino Ratio | N/A |
| Maximum Drawdown | N/A |
| Win Rate | N/A |
