Why mean reversion matters
It underpins relative-value, statistical-arbitrage, rebalancing, and valuation strategies, but apparent deviations can instead represent permanent structural change.
How it is applied
Researchers define the variable and equilibrium, estimate stability and speed, control for trends and regime breaks, establish entry and exit rules, and include costs, liquidity, and stop conditions. Define the variable, equilibrium estimate, horizon, entry threshold, and exit rule, then test stability out of sample. Strategies often trade deviations in prices, spreads, valuations, or relative relationships while controlling structural breaks, transaction costs, and crowded positioning.
Portfolio example
Two historically related securities diverge after a temporary flow. A strategy buys the cheaper and sells the richer, expecting their spread to narrow. A spread historically averages 2% with 0.5% standard deviation and widens to 3.5%. A reversion strategy may buy the cheap leg and sell the rich leg, expecting convergence. Continued widening to 5% can create severe loss.
How to interpret it
A large deviation can offer opportunity only if the estimated mean remains relevant and convergence occurs within the investor’s horizon and financing capacity. Mean reversion assumes deviations are temporary around a reasonably stable relationship. Faster estimated reversion supports shorter holding periods. Economic explanation is essential because a changing fundamental can justify a new mean.
Limitations and common misconceptions
Means move, relationships break, and prices can trend farther before reversing. Short samples, data mining, leverage, crowded exits, and transaction costs can turn convergence trades into large losses. Historical averages shift, trends persist, and leverage can force exit before convergence. Backtests often select relationships that worked by chance. Borrow, funding, market impact, and tail losses can overwhelm many small gains. Position limits should account for the possibility that the observed relationship reflects a hidden common factor or financing condition. A stop based only on further deviation can crystallize loss just before convergence, while no stop can permit ruin. Governance should define the evidence for a structural break. Half-life estimates are uncertain and should not dictate leverage mechanically. The strategy needs a predefined response when the assumed equilibrium no longer appears economically credible.
Sources and further reading
- Quantitative MethodsCFA Institute