Mean reversion research lab
Explore prior-window displacements and next-close distance using your own positive closing-price series or a clearly labelled synthetic sample.
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.
Paste a series of closes, one positive number per line. The lab asks a narrow question: after price stretches away from its prior rolling mean, did the next close become strictly closer to that unchanged mean more often than the eligible baseline?
Runs locally in your browser. No account, payment, price feed or upload. A close-to-close diagnostic is not a backtest or tradable strategy.
Closing-price series
At least 30 prices. Use one consistently adjusted series; do not mix assets or scales. Dates and trade fills are not inferred.
Predeclare the test
At bar i, the mean and standard deviation use only closes before i. The current close defines a stretch; close i + 1 is used only to measure the next move. Changing settings after seeing results is exploration, not untouched validation.
Observed next-bar behavior
Most recent stretch observations
| Bar | Close | Prior mean | Z-score | Next close | Closer to prior mean? |
|---|
Interpretation limits: bars overlap, so observations are not independent. A move toward the mean can still be too small to trade after spread, slippage and fees. Close-to-close measurements do not specify an executable entry, exit, stop or fill. Data quality, regime shifts and out-of-sample validation matter.
Distance, not merely direction
Recovery here means that the next close is strictly nearer to the unchanged prior-window mean. A next close can move in the expected direction but overshoot so far that the absolute distance increases; that observation is not counted as recovery. Overlapping windows and selected thresholds do not supply independent trades or a tested edge.
One-step descriptions and full paths answer different questions
The original lab compares a next close with a frozen prior-window mean and separately reports a directional move. The new lens follows an entered frozen reference across multiple observations and records band and touch/cross endpoints. Their populations are not identical. A positive directional move can overshoot and fail the lab's strict closer rule. A full path can hit a band and later end farther away. Neither tool includes executions or costs. Retain both contracts rather than silently broadening a next-close descriptive statistic into a completed trading record.
User-entered sources remain unverified
The lab accepts positive numeric closes under its existing bounds. The synthetic sample is clearly authored, while an entered list is merely user supplied. Neither proves original source, arrival clock, adjustment policy or complete eligible market population. Prior population standard deviation is the stated convention; constant windows are excluded as undefined rather than assigned a convenient score. Exportable teaching packets from the new workbook add inspection value without converting the lab sample into a genuine chronological validation set.