Skip to reversion research
Mean Reversion StrategyReference / Excursion / Evidence

Working glossary

Current close minus the mean of earlier observations, divided by the declared earlier-window scale. It is a displacement statistic, not automatically a normal-tail probability.

Original editorial revision: 9 October 2026. Expanded edition: 14 October 2026.

Educational diagnostics, not a signal service. A standardized price displacement is not a calibrated probability; invented examples and user-supplied observations are not verified market performance.

Prior-window z-score

Current close minus the mean of earlier observations, divided by the declared earlier-window scale. It is a displacement statistic, not automatically a normal-tail probability.

Distance recovery

A later price is strictly closer to the same frozen reference in absolute price distance. A directional move can overshoot and fail this definition.

Expectation half-life

Under a specified stable positive AR(1) coefficient, the fractional number of observation intervals for conditional expected displacement to halve. Not a trade exit deadline.

Stationarity hypothesis

An assumption about the distributional behaviour of the selected process. The exact test specification and alternative matter; a rejection of one null does not certify a strategy.

Three endpoints and a terminal quantity

Strict closeness means a future absolute distance is smaller than initial distance to the same reference. Band arrival means distance is at most a declared tolerance. Touch/cross means a price reaches or passes the destination from the initial side. Terminal distance recovery compares initial and final inspected distances, not capital. These endpoints can disagree. A initial point inside the band has a step-zero band hit; a point at reference already touches at step zero. Positive-price diagnostics are not automatically definitions for signed residuals, valuation errors or executable two-leg strategies.

Conditional half-life and first passage

Expectation half-life solves the assumed conditional mean decay toward a fixed equilibrium for a positive stable AR(1) coefficient. First passage asks when a path first satisfies a boundary event. A bounded observed hit and an assumed expectation are different objects. The original worksheet fits no coefficient and the new lens estimates no passage distribution. A null endpoint inside a complete inspected horizon is an observed non-hit under that contract, not proof of eternal non-recovery. Irregular time intervals cannot be converted into an exact calendar deadline by simply counting rows.

Event population and overlap

An eligible observation satisfies the declared data and scale rules. A threshold event is selected from that population under the frozen event operator. A cluster can contain adjacent events belonging to one longer excursion. Independent samples require a design rather than a larger displayed count. Missing observations, zero scale and unresolved labels must be attributable. A price series entered by a user is not authenticated merely because a browser can parse it. The workbook's complete ledger has full teaching coverage while remaining fictional rather than genuine research evidence.

Reference identity and research completion

Reference identity joins process, source version, destination definition, fit scope and effective clock. A material destination or hedge-ratio change creates a compatibility question for old labels. A research master is an independently required completion artifact, not an experiment-log adjective. Deployed identity and subscriber certification concern separate cloud-serving objects. A LIVE filename cannot connect those stages by itself. A raw aggregate of percentages also does not establish weighted account equity. Documentation should keep these objects distinct before any marketing claim turns one of them into another.