Tag: ml-alpha-research
Concepts
- Beating the Linear Baseline: When ML Actually Helps
- Binary vs Continuous Labels for Alpha Models
- Capacity of Machine-Learning Signals
- Choosing the Embargo Length
- Choosing the Prediction Target: Return, Rank or Direction
- Clustered Feature Importance and Mean Decrease Accuracy
- Conditioning Variables and Interaction Features
- Converting Model Scores into Positions
- Cost-Aware Trade Thresholds and No-Trade Bands
- Deriving Regime Labels for Model Switching
- Ensembling Across Seeds and Refit Dates
- Feature Half-Life and Smoothing Choices
- Matching Forward-Return Horizon to Signal Decay
- Hyperparameter Search Budget and Multiple Testing
- Label Overlap and Effective Sample Size
- Lagging Features to Enforce Point-in-Time Safety
- Leakage Audits for Alpha Pipelines
- Orthogonalising Signals Before Blending
- Pooled vs Per-Asset Models on Panel Data
- Rank Targets and Cross-Sectional Normalisation
- Recency Weighting and Decay in Training Samples
- Refit Frequency for Cross-Sectional Models
- Residualised Return Targets After a Risk Model
- Train-Test Decay Curves for ML Signals
- Turnover Penalties in Model Training
- Volatility-Scaled Return Targets