Mean Reversion: Recovery, Overshoot and a Moving Reference
Define what returning toward a mean actually measures before treating a directional move as recovery or a statistical pattern as a trade.

Freeze the reference for the question
Write the mean calculation, input window, availability time and whether the comparison uses a frozen or updating reference. A price can approach a moving average because price moved, because the average moved or because both changed. Those are different observations. For a boundary-recovery question, a fixed prior reference makes the distance comparison explicit. For an updating-reference strategy, the update rule belongs in the strategy identity. Do not switch references after inspecting which version produces more apparent recoveries.
Measure distance as well as direction
A positive move after a low price does not necessarily leave the price closer to the old mean. Define signed displacement and absolute distance separately. Record what counts as recovery, equality, overshoot and missing observation. A boundary touch is another possible outcome with its own timing rule. Avoid treating these labels as interchangeable wins. A recovery label measures the chosen price relationship; it does not by itself specify an entry, exit, account return or probability suitable for position sizing.
Worked example: the price moves the right way but overshoots
Invent a frozen reference of 100 and a current close of 95, giving an absolute distance of five. A later close of 103 has distance three and is closer under the stated rule. A later close of 107 has distance seven: it moved upward but ended farther away. Classifying both as identical recovery would silently change the objective. These numbers are a definition fixture, not evidence of a market pattern. An executable strategy would additionally need decision availability, order assumptions, costs and an exit rule.
Build the eligible comparison population
Declare which observations can enter the study, including source coverage, missing values, session boundaries and the availability of the full outcome window. Overlapping windows can create dependent observations. Keep the baseline population comparable to the selected-boundary population rather than comparing a curated subset with an unrelated average. Time-aware evaluation documentation supplies software context, not a universal financial test. Preserve the candidate registry and the history of thresholds considered so an attractive pattern is not reported without its selection background.
Keep outcome research separate from trade economics
A causal recovery relationship can be useful to study while a particular execution implementation remains unprofitable or unavailable. Conversely, a selected trade return can hide a weak or differently defined entry objective. Freeze the entry question before applying an execution or exit screen. A low-price observation is not automatically a fill at that low. SEC order guidance helps identify the execution questions that remain outside the distance label. Research references and real broker transactions must retain their distinct identities.
Publish the definition and unresolved assumptions
A report should provide reference values, input availability, current and later distances, eligible population, dependencies and calculation code. Describe the result under its declared definition rather than promoting a generic mean-reversion edge. If the underlying observations or evaluation history are missing, publish a protocol, not a numerical finding. The worksheet lets another reader reproduce the classification on invented or properly sourced values. Passing this unit-level exercise confirms consistent arithmetic; it cannot certify that a strategy works, determine a market scope or replace independent validation.
Decision record
| Question | Record to retain | Decision boundary |
|---|---|---|
| Reference | Frozen window or explicit update rule | Price and mean movement must not be conflated |
| Recovery | Absolute distances before and after | Expected direction can still end farther away |
| Population | Eligible observations and overlap policy | Many dependent windows are not independent trades |
| Trading claim | Separate entry, exit, cost and fill convention | A label relationship is not account profit |
Blank working record
The worksheet contains column headings only. Enter your own dated observations, keep absent evidence marked unknown, and do not treat the file as certification.
Primary references
These documents support mechanisms and source definitions, not a tested trading result or a recommendation. Accessed 9 October 2026. The decision workflow and invented examples above are editorial exercises.
Keep reading
Publication methodology | Working glossary | Source and correction policy
Equality and initial-band cases need explicit operators
A future point at the same absolute distance fails strict closeness even when it crosses the destination. Band equality counts as a hit under the inclusive operator here. A starting point already in the band has a step-zero hit, while a start exactly at reference has undefined percentage distance recovery. These edge cases are now downloadable with full declarations. They do not identify a real event probability or a financial exit. Preserve these meanings if the outcome contract is used in later genuine research.