Third-Party Research & Methodology Only

This section shares summaries of third-party academic research and descriptions of quantitative models. The content represents the findings of the original researchers, not the opinions or recommendations of Foxholm Financial. Foxholm Financial does not publish hypothetical or backtested performance metrics on its quantitative research pages. All content is restricted to methodology, signal construction, factor logic, and risk architecture. SEC rules require that investment advisers not present misleading performance data, and our methodology-only approach reflects that standard and the firm's fiduciary obligations.

Mean Reversion Strategy

Statistical Trading Strategy Signal Generation

Robert Stowe, AAMS® Investment Advisor

Mean reversion is the idea that prices tend to move back toward their long-term average after swinging too far in either direction. A mean reversion strategy looks for assets that have moved unusually far from their normal level and bets that they will snap back.

This approach rests on a simple observation: extreme moves in either direction tend to be temporary. A stock that drops 30% in a week on no fundamental news is more likely to bounce back than to keep falling at that pace. Mean reversion strategies try to identify these overreactions systematically using statistical measures of how far a price has strayed from its typical range.

Conceptual Framework

Mean reversion has roots in the work of statistician Francis Galton, who in the 1880s observed that extreme measurements (like very tall parents) tend to produce offspring closer to the population average. In finance, the concept applies to prices, valuations, and spreads that deviate from a stable long-run level and eventually return to it.

The key question is whether mean reversion is a genuine economic force or a statistical artifact. In many markets, there is a plausible economic mechanism: when a stock falls far below its fair value, bargain hunters step in and push the price back up. When a stock rises far above fair value, sellers take profits. These forces create a gravitational pull toward equilibrium.

Core Assumptions

Mean reversion models make several assumptions about how prices behave. Each assumption introduces risk if reality diverges from the model:

A stable mean exists: The model assumes that the price, spread, or ratio being tracked has a long-run average that stays roughly constant over the trading horizon. If the underlying value shifts permanently (due to a change in the business, the economy, or market structure), the "mean" the model targets may no longer be relevant. A stock that falls 50% due to a genuine earnings collapse is not reverting; it has found a new level.
Deviations are temporary: The model assumes that large moves away from the mean are caused by temporary forces (overreaction, liquidity shocks, short-term sentiment) rather than lasting changes. When this assumption fails, the strategy buys into a falling trend or sells into a rising one.
Speed of reversion is predictable: Most models assume that prices revert at a roughly consistent speed. In practice, reversion can take days, months, or years, and the speed often depends on market conditions. During calm markets, reversion tends to happen faster. During crises, dislocations can persist far longer than models expect.
Transaction costs are manageable: Mean reversion strategies often trade frequently, buying on dips and selling on recoveries. If the expected reversion is small, transaction costs (including bid-ask spreads, commissions, and market impact) can consume the entire expected profit.

Signal Construction

A mean reversion strategy follows a structured process from measuring the deviation to generating a trade signal.

Step 1

Define the Mean

Step 2

Measure Deviation

Step 3

Apply Filters

Step 4

Generate Signal

Step 5

Set Exit Rules

Defining the Mean

The first decision is what "average" the price should revert to. Common choices include a simple moving average (the average closing price over the last N days), an exponential moving average (which gives more weight to recent prices), or a fundamental anchor like a price-to-earnings ratio. The choice of lookback period matters: a 20-day average responds quickly to recent changes but generates more noise, while a 200-day average is more stable but slower to reflect genuine shifts in value.

Measuring Deviation

Once the mean is defined, the model measures how far the current price sits from it. The most common measure is the z-score: the number of standard deviations the current price is above or below the mean. A z-score of +2.0 means the price is two standard deviations above its average, which historically occurs only about 2.5% of the time if prices follow a bell curve.

Other deviation measures include Bollinger Bands (which plot bands at a fixed number of standard deviations around a moving average), the Relative Strength Index or RSI (a momentum oscillator that measures the speed and size of recent price changes on a 0-to-100 scale), and the distance between a short-term and long-term moving average.

Filters and Confirmation

Raw deviation signals produce many false positives. Filters reduce the noise by requiring additional conditions before triggering a trade. Common filters include:

Volume confirmation: Requiring above-average trading volume on the reversal day suggests that real buying or selling interest is driving the move, not just thin trading.
Regime filter: Checking whether the broader market is in a trending or range-bound state. Mean reversion works best in range-bound markets; during strong trends, what looks like an overreaction may actually be the start of a larger move.
Fundamental screen: Filtering out stocks with deteriorating earnings, high debt, or other signs that the price decline reflects a genuine change in value rather than a temporary overreaction.

Risk Architecture

Mean reversion strategies face a distinct set of risks that differ from trend-following approaches. The core danger is that the strategy systematically buys falling assets and sells rising ones, which can produce severe losses when trends persist.

Model Risk

The biggest risk is misidentifying a structural change as a temporary deviation. If a stock's earnings permanently decline, its price will settle at a new, lower level. A mean reversion model that continues to buy as the price falls will accumulate losses. This is often called "catching a falling knife." The model has no way to distinguish between a temporary dislocation and a permanent repricing unless additional fundamental filters are applied.

A second source of model risk is the lookback period. If the period is too short, the estimated mean is noisy and the model trades on random fluctuations. If the period is too long, the mean may reflect outdated conditions. There is no universally correct lookback; it depends on the asset, the market environment, and the time horizon of the strategy.

Known Limitations

Limitations to Consider

Regime dependence: Mean reversion works well in range-bound, low-volatility markets but performs poorly during strong trends or market crises. A strategy that thrived in 2017 could have produced large losses in March 2020.
Structural breaks: When a company, sector, or economy undergoes a fundamental shift, the historical mean becomes irrelevant. The model cannot detect structural breaks on its own.
Crowding: Because mean reversion signals are widely known and easy to compute, many participants may trade the same signals simultaneously. This competition can erode the expected profit or cause the reversion to happen before the signal fires.
Fat tails: Extreme price moves happen more often than a bell curve predicts. A z-score of 3.0 should be very rare under normal assumptions, but in real markets, moves of this magnitude occur regularly. This means the model may underestimate the probability of the deviation getting worse before it gets better.
Holding period uncertainty: The model predicts that prices will revert, but not exactly when. Capital can be tied up in losing positions for weeks or months before the expected reversion occurs, or it may never occur at all.

Practical Considerations

Where Mean Reversion Works Best

Mean reversion tends to work best in liquid, well-established markets where prices are anchored to fundamental values and temporary dislocations get corrected by active traders. Large-cap stocks, major currency pairs, and bond yield spreads show stronger mean-reverting behavior than small-cap stocks or emerging-market assets, where information is scarcer and mispricings can persist longer.

Pairs trading, a closely related approach, applies mean reversion to the spread between two related securities rather than to a single asset's price. This removes much of the market-direction risk because the bet is on the relationship between two stocks, not on the direction of the overall market.

Mean Reversion vs. Momentum

Mean reversion and momentum are opposing views of how prices move. Momentum strategies buy recent winners and sell recent losers, betting that trends will continue. Mean reversion strategies do the opposite, betting that trends will reverse. Research consistently shows that both effects exist in the data, but they operate at different time horizons. Momentum tends to dominate over 3-to-12-month horizons, while mean reversion is stronger over very short periods (days to weeks) and very long periods (3 to 5 years).

Because these two signals are negatively correlated (one tends to profit when the other loses), combining them in a portfolio can reduce overall risk. This complementary relationship is a key reason many systematic portfolios include both momentum and mean reversion components.

Position Sizing and Stop Losses

Because mean reversion strategies buy into losses, position sizing is critical. A common approach is to scale into a position as the deviation grows: buy a small amount at a z-score of 2.0, add more at 2.5, and more at 3.0. This spreads the entry across a wider range of prices. Hard stop losses (automatic exits at a predetermined loss level) are essential to prevent a single position from causing outsized damage when the expected reversion fails to materialize.