Pair Selection & Regime Filters - Making Z‑Score Tradable

Pair Selection & Regime Filters - Making Z‑Score Rotation Tradable (Not Just Interesting)

Pair rotation looks simple on paper: two assets diverge, you rotate from the weak to the strong (or position for mean reversion). In practice, most “pair tools” fail because the user treats any divergence as tradable. The reality is harsher:

- Some pairs are not stable relationships; they are temporary coincidences.
- Some divergences are structural repricings, not mean‑revertable dislocations.
- Some regimes reward rotation; other regimes punish it with trend persistence.

This is where a Z‑score‑based rotation engine becomes valuable — not because it gives you more signals, but because it can help you enforce a disciplined selection and filtering process.

1) Pair selection is 70% of the edge
Before you touch settings, decide whether the pair makes sense. A “good” pair usually has at least one of these properties:
- Same sector / same narrative bucket (e.g., two L1s, two AI infra tokens, two DEX tokens).
- Similar risk profile (beta, volatility, liquidity).
- A clear reason to be compared (benchmark vs challenger, ecosystem peers, correlated exposure).

Red flags:
- One side is consistently thinner liquidity (slippage dominates the “edge”).
- One side is driven by frequent idiosyncratic news (correlation breaks often).
- Volatility profiles are wildly different (Z‑score extremes become normal).

2) Define what “dislocation” means for THIS pair
Z‑score is normalization, not truth. Your job is to pick settings that match the pair’s natural behavior:
- If the pair is stable and slow, you can use higher thresholds (fewer, cleaner extremes).
- If the pair is noisy, you must rely more on selectivity (quality filters, cooldown) to avoid churn.

Practical habit:
Spend time in Bar Replay and watch how long extremes persist.
If the pair often stays extreme for many bars, treat rotation as a slower process, not a quick scalp.

3) Regime filters: when rotation works and when it lies
Pair rotation is most fragile during:
- Strong trend regimes (one asset leads and keeps leading).
- Correlation breakdowns (the pair stops behaving like a pair).
- Volatility shock phases (fat tails; “rare” moves become common).

A rotation engine with pair health / quality context helps you avoid the worst of these traps by telling you when the environment is “less trustworthy” for the model.

4) How I read engine states in a regime-aware way
- Monitoring: no meaningful dislocation; avoid forcing action.
- Pending: dislocation forming; prepare a plan, don’t pre‑commit.
- Blocked: the engine is explicitly telling you the context is low quality (health/cooldown/config). Respect it.
- Eligible: dislocation + filters align; now you can evaluate execution with your own rules.

Key point:
Eligible does not mean “buy/sell now.” It means “this is worth a serious look.”

5) A practical pair rotation checklist (copy/paste)
Before acting on a rotation state:
- Does the pair still make structural sense (same narrative, similar liquidity)?
- Is the regime stable (no obvious correlation breakdown, no shock event)?
- Is the engine state Eligible (not Blocked)?
- Is cooldown satisfied (no flip‑flop risk)?
- Do I know my invalidation (where the idea is wrong) and my max risk?

If you cannot answer these quickly, you are not trading a process — you are reacting to a number.

Limitations:
All pair frameworks can fail in trend persistence and structural repricing. Use as decision support, not as a standalone strategy.

Risk disclosure:
Educational content only. Not financial advice.