What is Mean Reversion? definition, context and sources

01What is Mean Reversion?

Statistical mean reversion and a trading rule are not the same thing. Historical negative autocorrelation of returns in a particular market and horizon does not mean that an individual Bitcoin price must return to a specific moving average or that a strategy can earn money after costs.

02Distinguish a return hypothesis from a target price line

Poterba and Summers studied whether long-horizon stock returns partially reverse; they did not claim every price must touch a fixed level. A later reappraisal by Kim, Nelson and Startz showed how sensitive that conclusion is to the sample and period. Reversion of returns, price, a two-asset spread and deviation from a moving average are different hypotheses that require separate tests. [1] [2]

03Define the measured variable and its reference first

State the market, exchange, time resolution, quote currency, price type and historical window length. A simple average of already known closes is one transparent reference, not a natural or immutable value of Bitcoin. A fixed average from an earlier window and a continuously recalculated SMA are different constructions; both must use only data available at decision time. [3] [5]

04Distance from an average alone is not an entry signal

A deviation can be expressed in price units or as a percentage of a reference calculated in advance; the rule must state which direction and threshold trigger an observation or order. Also specify in advance how to treat equality, recalculation of the average, repeated signals and reaching the target. A threshold selected retrospectively for the best-looking chart distorts the test. [3] [5]

05The average can move and a trend can continue

A price jump may reflect new information or a market-regime change; the old reference may then be a poor estimate of the future center. The NBER reappraisal found weaker reversal evidence than the earlier paper, and CFA Institute explicitly warns of structural breaks in historical data. A longer deviation by itself increases neither certainty nor identifies a safe time for a contrarian position. [2] [5]

06Entry, exit and invalidation belong together

A trading implementation must specify whether a deviation leads to a long, short or no position; when it exits partly or fully; and what happens if the deviation keeps growing or the reference regime no longer holds. Merely saying buy below the average and sell above it says nothing about position size, a time limit, liquidity or losses if a trend continues. [5] [4]

07A completed Close cannot also be assumed to be the fill price

If a deviation is calculated from a completed Close, the signal is not known before that close. TradingView's default model fills a market order on the next available tick, often at the following Open. A test must separate signal creation from possible execution and include fees, spread and Slippage; with a small deviation, costs can change the sign of the result. [4] [5]

08A stock-market study does not establish Bitcoin price reversion

The NBER papers describe stock returns in particular countries and periods, not a guaranteed property of Bitcoin, a specific exchange or a short trading interval. Bitcoin requires separate tests of liquidity, market regime, historical-data availability and price source. Even a visible return on a chart does not establish an honestly attainable fill or positive expectancy. [1] [2] [5]

09Evaluate results outside the data used to choose the rule

Fix the window, reference, threshold, entry, exit, position size, execution and costs before testing. Separate the period used to choose parameters from later evaluation and examine trending intervals and abrupt regime changes. CFA Institute warns about look-ahead and survivorship bias and the need for sensitivity and scenario analysis; a single retrospectively selected winning chart is insufficient. [5]

ExampleEXAMPLE · MEAN

A deviation and later return toward a fixed reference

A fictional sequence of five completed Close values in USD/BTC for t = 0 through t = 4 is [100000, 102000, 98000, 104000, 99000]. The fixed reference known after t = 2 is the average of the first three closes: (100000 + 102000 + 98000) / 3 = 100000 USD/BTC. At t = 3, Close is 104000, or +4000 USD/BTC and +4 % relative to this fixed reference. At t = 4, Close is 99000, or −1000 USD/BTC and −1 %; the absolute distance from the reference narrowed from 4000 to 1000 USD/BTC. This is one fictional sequence described in hindsight, not a prediction or validation of a strategy. Index t denotes the order of completed observations, not days. At t = 3, the future Close at t = 4 is unknown. The next Open, actual fill, spread, fees and position size are missing, so a trade profit cannot be inferred from the difference between closes. The reference is deliberately fixed; a rolling SMA would have different values at later points.

Mean Reversion is a testable hypothesis for a particular variable and horizon, not an obligation for price to return to a line.
The historical average must be known when the signal occurs; fixed and continuously recalculated references differ.
A regime change can invalidate an old average and widen the deviation further.
A trading result requires executable entry and exit, costs and evaluation outside the selected data.

Read in context

For the clearest picture, read this entry together with Moving average, Backtesting, Trend Following, Range Trading, Risk Management. The reverse links also lead from Range Trading.

?Practical questions

01Must price return to its historical average after a large deviation?

No. A historical average is a calculation from a selected window, not a binding future price. A trend or regime change can extend the deviation or move the reference itself.

02Does a Close returning toward the average prove a profitable trade?

No. A signal using a completed Close exists only after that close, and the trade may execute at another price. Results depend on entry and exit rules, position size, spread, fees and slippage.