Recency Bias and Mean Reversion

Recent results are vivid, but a historical average is not a forecast. The useful question is whether the reference level still applies and over what horizon.

Two ideas are often fused too quickly

Recency bias is a judgment tendency: recent, available, and emotionally vivid information can receive more weight than older evidence, even when the older evidence is relevant. Mean reversion is a statistical description: a measure that has moved far from a reference distribution may later move back toward it. One concerns how people form expectations; the other concerns what the data sometimes does.

The Tversky and Kahneman heuristics paper describes availability and representativeness as sources of judgement error. It does not say that recent information is always misleading. A new product, regulation, technology, or customer loss can make older data less relevant. Recency becomes a problem when vividness substitutes for testing whether the underlying process changed.

Mean reversion has a horizon and a mechanism

A margin may revert because competition enters, customers resist high prices, capacity responds, or costs normalize. A commodity price may attract production when high and curtail supply when low. An asset return may show statistical reversal over one horizon and momentum over another. Without the mechanism and horizon, “mean reversion” is only a label for a chart pattern.

De Bondt and Thaler’s long-term return-reversal study is a historical finding in a defined sample and period. It is not evidence that every stock below its past average is cheap or that a reversal will occur soon. By contrast, Jegadeesh and Titman’s momentum research documented continuation over another horizon. The contrast is important: the same price history can support different behaviour at different time scales.

Business metrics can revert—or reset

• Margins. High margins can attract competition or invite customer pushback, but a patent, network, regulation, or process advantage can move the sustainable level higher.
• Growth. Rapid growth can slow as the base expands, but a new market or product can create a new growth regime rather than a return to the old average.
• Valuation. A high multiple can fall as expectations normalize, but the correct reference multiple changes with rates, cash-flow duration, and business quality.
• Commodity prices. Supply response can pull prices toward a cost curve, but investment delays, geopolitics, and depletion can keep the cycle extreme for years.
• Returns. Past winners can reverse over long horizons while continuing to trend over shorter horizons. The sample, benchmark, and transaction costs matter.

Recency bias makes the current regime feel permanent. At a peak, recent growth and margin strength seem like the normal future; at a trough, recent weakness seems like proof of permanent decline. The analyst must ask whether competitive entry, demand, capacity, or regulation is actually moving the measure or merely creating a temporary phase.

A concrete diagnostic sequence

First, define the measure and reference: operating margin versus a company’s history, commodity price versus a cost curve, or return versus a market benchmark. Second, specify the horizon over which reversion is expected. Third, identify the force that would cause it. Fourth, look for evidence that the force is operating: new capacity, price concessions, inventory normalization, customer switching, or changed capital allocation.

Finally, test the alternative explanation that the old average is obsolete. A business with a new distribution model may have a different cost structure. A commodity with constrained supply may have a different equilibrium. A market with changed participation may have a different return process. A mean-reversion trade made before this question is answered is a bet on an untested reference level.

What the combined error looks like

An investor sees three years of expanding margins and projects the peak indefinitely. The company then adds capacity, customers negotiate, or input prices normalize. The forecast was exposed to recency bias, and a mean-reversion mechanism was visible. The opposite error is to reject a company’s new economics solely because its margins exceed the old average. If the process, customers, and competitive position have changed, the old average may no longer be the right comparison.

Price can also move before the accounting evidence. A stock may continue rising as investors update expectations, then reverse after the new growth rate fails to justify the valuation. The price path is not proof of either bias or reversion; it is an observation that needs a defined benchmark and later evidence.

What investors can test

• Put current growth, margins, valuation, and returns in a distribution with an explicit time period and peer set.
• Separate short-run momentum from long-run reversal and account for transaction costs, taxes, and liquidity.
• Write the mechanism that should pull the measure back and the evidence that would confirm or weaken it.
• Check for structural changes in customers, products, regulation, capacity, technology, and capital intensity before using a historical average.
• Compare management forecasts and market expectations with base rates, but do not treat the base rate as a ceiling.
• Review the position when the mechanism changes, not merely when the price crosses the historical average.

Recency bias and mean reversion are most useful together as a discipline against unexamined extrapolation. They do not turn an extreme into a forecast. They force the investor to decide whether the present is a temporary state, a momentum phase, or a new normal—and to keep the uncertainty visible.