Memo 003: Regime detection and failure modes

Many strategies work in one regime and fail in another. Mean reversion works in range-bound markets and gets run over in strong trends. Carry works in risk-on and unwinds in risk-off. Momentum works in trends and whipsaws in chop.

In the database we don’t just list “what works” ; we list where it works (regimes) and failure modes. That’s because we want the database to be a practical reference: before you implement, you should know when the strategy is likely to lose money.

See Regime Detection for how we think about regimes, and Mean Reversion Russell and FX Carry for examples of regime-dependent edges.

What would change our mind? If we find a robust, low-lag regime indicator that doesn’t overfit, we’d add it as a standard filter across strategies. So far we use simple vol and trend strength; more complex ML-based regime models are easy to overfit.