Understanding the metrics

Cointegration

Engle-Granger test on five years of adjusted prices (in logarithm). “Yes” means the no-relationship hypothesis is rejected at the displayed threshold: the pair's spread has a statistically demonstrated restoring force.

P-value

Probability of observing such a stable spread if the two securities actually had no link. The lower it is, the more credible the cointegration. Usual threshold: 0.05.

Q-value (FDR)

P-value corrected for multiple testing (Benjamini-Hochberg procedure). Across thousands of tested pairs, some look cointegrated by pure chance: the q-value controls the expected share of false discoveries.

Z-score

Current position of the spread relative to its mean, in standard deviations (60-day rolling window by default). A Z of +2 means A is unusually expensive relative to B, a Z of -2 the opposite.

Live Z

Same computation as the Z-score, but using the latest intraday prices (quotes delayed by about 15 to 20 minutes) instead of the last close.

Signal

Directional reading of the Z-score beyond the threshold (2 by default): LONG A / SHORT B when the spread is abnormally low (buy A, sell B), SHORT A / LONG B when it is abnormally high. Beyond |Z| = 4 the relationship is considered possibly broken and the row is dimmed.

OOS p-val

Out-of-sample validation: the relationship is estimated on the first 70% of the history, then the spread's stationarity is tested (ADF) on the remaining 30%, never seen before. A low value indicates the relationship held outside the estimation period.

Reversion

Realized reversion rate: among historical signals where |Z| reached 2, the share that came back toward equilibrium (|Z| below 0.5) within the next 20 sessions. The number of events is shown in parentheses; a rate based on few signals is greyed out as unreliable.

Half-life

Average time needed for the spread's deviation from its mean to shrink by half, estimated with a mean-reversion model. The shorter it is, the faster a position unwinds.

Correlation

Correlation of the two securities' daily returns. Green if statistically significant, red otherwise. High correlation is not enough: only cointegration demonstrates a restoring force on prices.

Hedge

Hedge ratio (beta) estimated by regression: the number of units of B to trade for each unit of A in order to neutralize the spread.

20-session divergence

Performance gap between A and B over the last 20 sessions, expressed in standard deviations of its historical distribution. No hedge ratio, no long-term equilibrium: this is the correlation-and-divergence reading (one name takes off, its twin does not follow). It complements the spread Z-score, it does not replace it.

Beginner's explanation: over the last month of trading, one of the two securities pulled ahead of the other. This number tells you whether that lead is ordinary or exceptional compared with their shared history.

Common benchmarks, in absolute value:

• under 1 σ: ordinary gap
• 1 to 2 σ: notable lead
• beyond 2 σ: rare divergence (roughly one session in twenty)
• beyond 3 σ: exceptional, look first for a cause in the two names' news

Ratio (β = 1)

Raw ratio of the two prices, P(A) / P(B), tracked with the same ± 2σ bands as the spread. It is the most intuitive reading of pair trading; it assumes a hedge ratio of 1, whereas the scanner's spread estimates β by regression. When β is close to 1, both views tell the same story.

Beginner's explanation: simply divide A's price by B's price. When that number drifts far from its usual level, the gap between the two securities is unusual, and that is what its Z measures.

Common benchmarks, in absolute value:

• under 1: ratio near its usual level
• 1 to 2: noticeable stretch
• beyond 2: zone considered extreme in pair trading (usual entry threshold)
• beyond 4: gap so wide the relationship itself may be broken

Price A / Price B

Latest adjusted closes of the two securities (dividends and splits accounted for), as used in the spread computation.

Market beta (SPY)

Slope of a regression of the asset’s returns on SPY returns, with an intercept. Up to 252 shared returns, with at least 120 required. The calculation uses US closing prices in USD. R² measures the share of variance explained by SPY.

Beginner's explanation: This figure describes how closely the asset tended to follow the US market over the measured period.

Pair sensitivity

For weights +1 in A and −hedge in B: (beta A − hedge × beta B) / (1 + |hedge|). The denominator is gross notional. For two equal-dollar legs in opposite directions, the coefficient is (beta A − beta B) / 2. Reversing the trade reverses the sign. The 0.30 display threshold is a tool setting, not a statistical test.

Beginner's explanation: A figure near zero means the two legs offset part of their market-related moves. Other risks may remain.

SIC sector

Activity classification published by the SEC for the issuer. Pair Scanner groups stocks by the first two digits of their SIC code. Official descriptions remain in English. Codes refresh in the background; unknown codes are excluded by the Same SIC sector filter.

Beginner's explanation: The filter brings together companies from the same activity group. It does not tell you whether their prices will converge.

Cross-asset

A fixed selection of pairs across asset classes, tested with the same statistical functions and included in the multiple-testing correction. Both legs are quoted in USD and only shared dates are used. Their closing times may differ. Live Z is disabled across different exchanges.

Beginner's explanation: For example, gold can be compared with a fund of mining companies. An economic link motivates the comparison but does not validate the pair.