The SPX strategy uses normalized momentum and drawdown features with cubic polynomials, predicting returns via a sigmoid model to rebalance daily, investing only if returns exceed a 5% threshold.
ASSET CLASS: CFDs, ETFs, futures | REGION: United States | FREQUENCY:
Daily | MARKET: equities | KEYWORD: Logistic Regression, Momentum
I. STRATEGY IN A NUTSHELL
The investment strategy focuses on the SPX Index, using normalized momentum (30–360 business days) and drawdowns (15–120 business days) to predict profitability. Features are scaled to -1 to +1 to generate a profitability vector indicating if future returns exceed 5% annually. Cubic polynomials capture non-linear patterns. Daily predictions use weighted combinations of features through a sigmoid function. Weights are optimized on historical data, and predictions are made three days ahead. If returns are predicted below the threshold, the strategy holds cash; otherwise, it invests in SPX at market close. The strategy is rebalanced daily for dynamic adaptation to market conditions.
II. ECONOMIC RATIONALE
Momentum is a well-known factor, but machine learning enhances its application by adapting to changing market conditions and handling complex factor interactions. Polynomials allow non-linear relationships between features, enabling dynamic weighting based on market conditions. This method improves classical momentum strategies, including logistic regression-based models, and is flexible enough to apply to US or international equity indices, offering robustness and predictive power across diverse market environments.
III. SOURCE PAPER
Applying Machine Learning to Trading Strategies: Using Logistic Regression to Build Momentum-Based Trading Strategies [Click to Open PDF]
Beaudan, Patrick and He, Shuoyuan, Northern Trust Corporation; Emotomy, San Francisco State University
<Abstract>
This paper proposes a machine learning approach to building investment strategies that addresses several drawbacks of a classic approach. To demonstrate our approach, we use a logistic regression algorithm to build a time-series dual momentum trading strategy on the S&P 500 Index. Our algorithm outperforms both buy-and-hold and several base-case dual momentum strategies, significantly increasing returns and reducing risk. Applying the algorithm to other U.S. and international large capitalization equity indices generally yields improvements in risk-adjusted performance.
IV. BACKTEST PERFORMANCE
| Annualised Return | 8.6% |
| Volatility | 14% |
| Beta | N/A |
| Sharpe Ratio | 0.61 |
| Sortino Ratio | N/A |
| Maximum Drawdown | -45% |
| Win Rate | N/A |
