Skip to reversion research
Mean Reversion StrategyReference / Excursion / Evidence

The statistics of mean reversion

A rolling displacement does not supply a normal-tail probability. Mean-reversion evidence needs a specified process and conditional relationship, not just a price below a mean. Stationarity diagnostics, distance recovery and executed profit must be examined separately.

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.

The research question

A rolling displacement does not supply a normal-tail probability. Mean-reversion evidence needs a specified process and conditional relationship, not just a price below a mean. Stationarity diagnostics, distance recovery and executed profit must be examined separately.

An inspectable record

For a reversion claim, request the frozen reference definition, observation availability, complete eligible event population, treatment of crossing and overshooting, exit clock and cost model. An alert tool can express a condition without proving that a seller has tested it. Separate a model summary from a subscriber account record.

Next check

Write the price series, adjustment policy, prior lookback, standard-deviation convention, reference update clock, event threshold and recovery definition. Save a version before inspecting later outcomes. Count every eligible observation, including zero-variance exclusions, missing bars and events that never return.

Source documents

Documentation establishes the stated mechanism, not tested trading performance. Checked 9 October 2026; a later version can differ.

Detailed research

State the process and null before the score

An autoregressive fit, unit-root test and standardized displacement address different questions. The exact deterministic terms, lag choice, process transformation and sample matter. A small reported p-value does not describe a future trading probability or remove structural-change concerns. This site fits no real stationarity test in the new lens; it calculates finite supplied distances only. A scalar positive price path cannot establish the statistical properties of a signed residual, changing spread or valuation estimate.

Descriptive completeness still leaves inference open

The full authored ledger permits exact reconstruction of seven examples, including equality and overshoot failures. Its rows are not independently sampled market observations. Repeated adjacent events in a real series can share information, and threshold selection can consume validation. Preserve those boundaries before calling a displayed count a strategy win rate. Statistical documentation supports its mechanisms, while actual model selection and independent financial master retain their own evidence requirements. No fitted process or calibrated grade is invented to close that gap.

An absolute band has no inferred distribution

Band one in the authored packet is one EDU price unit, not one standard deviation or a confidence interval. The lens estimates no scale and calls no distribution function. A larger entered band changes a descriptive hit, not a fitted probability.