← 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.