Constructing pair spreads with z-score triggers

Three locked pieces

The editorial reading treats a two-name spread as a constructed instrument, not as three separate judgments about entry, size, and unwind. The locked pieces are a local fair ratio, a residual expressed in its own recent volatility, and a precommitted hedge size with a next-session exit.

Pairs-trading is a two-name procedure that holds one issue long and the other short on the premise that their prices share a recurrent relationship that can be estimated and faded when it stretches. Statistical-arbitrage is the matching rule: temporary deviations of a constructed spread from a local baseline are treated as tradable residuals, subject to hedge sizing and a defined holding period.

Assembling a candidate pair

Candidate two-name books can be assembled from a regularly recomputed correlation screen. A shared recurrent price pattern is treated as sufficient to start construction even when the corporate income link is left unused.

A pair can have a charter-style income split of 60 and 40 while the trading procedure still estimates the relative value from prices as if that split were unknown. The editorial point is that the income charter is not an input to the ratio. Relative value is read from prices.

A local estimate of fair value

The pair-ratio is the observed price of one name divided by the other. It is the raw series whose local average stands in for a stable relative value.

A 30-session moving average of the observed price ratio is used as that local estimate of the pair's stable value. The window is chosen as short enough to adapt and long enough to damp day-to-day noise.

Residuals in volatility units

A raw cash gap versus that estimated ratio is treated as an incomplete trigger because a fixed dollar difference does not mean the same thing in quiet and volatile regimes.

Z-score-normalization scales a spread residual by subtracting a lookback mean and dividing by the matching lookback standard deviation so thresholds are stated in recent volatility units rather than cash. Here the residual is converted into a z-score by subtracting its 30-session mean and dividing by its 30-session standard deviation, so mispricing is stated in recent volatility units.

Under an implicit normal assumption the scaled residual is treated as mean-zero and unit-variance. A 1.5-standard-deviation band then corresponds to about 6.7 percent of observations, or 17 days in a 250-session year.

One procedure for entry and unwind

Entry is defined when the scaled residual leaves a 1.5-standard-deviation band. The expensive name is sold and the cheap name is bought.

The hedge-ratio is the share quantity of the second name chosen to match the local fair-ratio estimate rather than a one-for-one or dollar-neutral book. That size uses the same 30-session average ratio used as the fair-value estimate.

The book is closed at the next session's close regardless of residual path. That forced-exit is a holding-period rule that closes the two-name book at a precommitted later print, independent of whether the residual has already returned to the baseline.

In the editorial reading, staying inside the band is abstention by the same rule, not a second decision. Entry, abstention, and unwind are then one testable procedure.

Filters that stay inside the construction

Further construction filters prefer names that share an industrial sector and a comparable business model, on the grounds that mixed revenue mixes weaken the case for a stable ratio. A more restrictive residual band reduces the number of accepted signals.