| name | pair-trading |
| description | Use this skill when the user asks about pair trading, statistical arbitrage, cointegrated pairs, spread trading, hedge ratios, or market-neutral strategies.
|
| license | MIT |
| metadata | {"author":"quantagent","version":"1.0"} |
| allowed-tools | find_cointegrated_pairs, compute_spread_metrics, run_backtest_tool, detect_market_regime |
Pair Trading / Statistical Arbitrage
Overview
A pair trade is long one leg, short the other, betting the spread
mean-reverts. The edge comes from cointegration — a stationary long-run
relationship — not from correlation. Two stocks can be 0.95 correlated
and still drift apart forever.
Methodology
- Scan within a sector (
find_cointegrated_pairs with a sector
filter). Cross-sector "pairs" that pass the test are usually
statistical accidents; economic linkage (same industry, shared
inputs, substitute products) is what makes the relationship durable.
- Validate the pair (
compute_spread_metrics):
coint_pvalue < 0.05 — the Engle-Granger gate.half_life_days — expected mean-reversion speed. 5–30 days is
tradeable; None or 60+ means the spread doesn't revert usefully.hedge_ratio — units of leg B per unit of leg A for a
dollar-neutral-ish spread. Recompute periodically; it drifts.
- Trade the z-score: enter at |z| ≥ 2 in the stated direction,
exit near z = 0 (|z| ≤ 0.5), hard stop at |z| ≥ 3–4 — a spread that
keeps widening past 3σ is telling you the relationship broke.
Risks to State Every Time
- Regime breaks kill pairs: mergers, guidance resets, index
add/drops, or one leg's fundamental story changing. A p-value from
the past year cannot see these coming. Check recent news on both legs.
- The p-value is in-sample; with ~1,700 pairs tested at 0.05, dozens
pass by chance. Prefer pairs that are also economically obvious.
- Shorting has borrow costs and squeeze risk that this analysis ignores.
- Size both legs by the hedge ratio, and treat the pair as one
position in journal/risk-gate terms (one thesis, one stop in z-terms).
How to Report
- Lead with the strongest 2–3 pairs: symbols, p-value, half-life, and
current z-score with the entry/exit signal.
- State the trade construction explicitly (long X shares-equivalent,
short hedge_ratio × Y).
- Note in-sample and regime-break caveats — always.
