Regimes Implementation
| ← Regimes Hypothesis | Team 1 Execution |
Hypothesis
From TwoSigma competition: solution benefited from basing models and weightings on different market regimes (relying more on Ridge than tree in volatile periods).
Application: Detect where/in which regimes models perform best.
Approach Options
| Option | Description |
|---|---|
| Regimes vs Groupings | Compare significance of regimes vs crypto groupings |
| Regime Features | Use regime changes for feature detection |
| Weighted Spearman | Features predicting price spikes benefit competition score |
Implementation Pipeline
Data → Coin Dummies → Train model(s) per version → Regime Dummies → Execute with weighted outputs
Possible to have regimes within large groups.
Key Insight
May make sense to use regimes for feature detection: spot features that perform well across all regimes when samples might be dominated by some specific regime.