What Is Abnormal Return?

Michael Whitman · Trading since 2017, full-time since 2020 · YouTube creator, 137K subscribers · Published 2026-09-14 · Updated 2026-10-19

Abnormal return is actual return minus the return a model said to expect, so it measures what the model failed to account for rather than what was earned. Because it depends entirely on the model chosen, changing the model changes the abnormality without anything about the investment changing.

Abnormal return is the most commonly misread term in finance research, because the word “abnormal” means something specific and narrow that has nothing to do with how it sounds.

How it works

Take what actually happened, subtract what a model said to expect, and the remainder is abnormal. That is the entire definition.

So it has two inputs and one of them is an assumption. The actual return is a fact. The expected return is a model output, and models are choices.

A simple model leaves more unexplained. Compare a stock to the market alone and much of its movement is abnormal. Add size and value factors and a great deal of that abnormality disappears — into the model, not into the market.

Nothing about the investment changed. Only the explanation did.

What it is actually for

Event studies. The question is whether an announcement moved a price beyond what the market would have done anyway — a merger, an earnings surprise, a drug trial result.

The method is to model the expected return, then measure the gap around the event date. That gap, cumulated over a few days, is the event’s estimated effect.

This is genuinely useful and it is research rather than trading. By the time abnormal return around an event has been measured, the event has happened and the price has moved.

A worked example

Take the S&P 500 since 1950. Across those decades 59% of ten-day windows ended higher than they began, so a rising stretch was the normal outcome rather than a surprise.

Suppose a model predicts a 59% chance of a higher close ten days out. A stretch that rises is not abnormal — it is the base rate arriving.

A stretch that rises much further than the model allowed is abnormal — and because gains and losses cluster in real markets, long one-way runs turn up more often than a model assuming independent days expects, so they show up as abnormal while being an ordinary feature of prices.

So the abnormality may be the model failing to account for clustering, not the market doing something extraordinary. The residual is real; the interpretation is where the work is.

Flow diagram of abnormal return: actual return, minus the expected return from a model, leaves a remainder, then the question of whether the market or the model failed. — A diagram of how this page defines abnormal return: a remainder after a model's expected return is taken away, so the model is half of the answer.

The joint hypothesis problem

Any test of abnormal return is testing two things at once. Whether the market behaved unusually, and whether the model of normal was correct.

A significant abnormal return means one of those failed — and the method cannot tell you which.

That is not a technicality. It is the reason decades of research finding “anomalies” remains contested: every anomaly is either a market inefficiency or a missing factor in the model, and the data alone does not separate them.

The original data

On the S&P 500 from 1950 to 2026, 53.1% of trading days closed up and 59% of ten-day windows finished higher. Every round trip still pays the spread plus any commission.

That base rate is what a null model would predict, and any strategy result has to be measured against it rather than against zero. A long-only method that is right on 59% of ten-day holds has produced exactly nothing abnormal.

And costs are the reason abnormality rarely survives contact with an account. A measured abnormal return of a fraction of a percent is real in a study and can be gone once the spread and commission on each round trip are paid.

The models people actually use

The simplest is the market model: expected return is some constant plus a multiple of the market’s return. Everything not explained by the market is abnormal.

Adding factors shrinks the residual. Size and value were the first two widely adopted, and each addition explains more of what was previously called abnormal — which is progress in modeling and looks like anomalies disappearing.

Every added factor is also a researcher degree of freedom. With enough factors almost any return becomes explained, and the line between a genuine risk factor and a curve fit is not sharply drawn.

So the honest reading of an abnormal return is conditional. Not “this was unusual” but “this was unusual given a particular model of normal, which is itself a hypothesis.” Stated that way the finding is weaker and considerably more accurate.

How this connects

The model of normal usually starts from market beta. A stock that moves one and a half times the market is expected to, so only the part beyond that multiple counts as abnormal.

What the model leaves unexplained is often a risk premium in disguise. A residual that persists for years is at least as likely to be compensation for a risk the model omitted as it is a mispricing.

Its day-to-day cousin is tracking error. Active risk measures how far a portfolio wanders from its benchmark; abnormal return measures how far a single outcome wanders from a model, and both describe deviation rather than skill.

Fund managers are judged on the same logic, which is the reason the Sharpe ratio exists: a return has to be weighed against the risk that produced it before it means anything.

The classic setting is the earnings report. The event window around an announcement is where most published abnormal returns are measured.

And a backtest that searches many rules is overfitting by another name, which is how statistical arbitrage desks lose money on residuals that were only ever noise.

When it fails

The characteristic failure is reading a published abnormal return as a tradeable edge. A study identifies a statistically significant residual around some event, the finding is real within its model, and the effect is a fraction of a percent measured across hundreds of cases. Trading it means paying the spread on every one of those cases, so a result that is genuinely abnormal in the data is negative after costs. The research was not wrong; it was answering a question about markets rather than about accounts.

A second failure is not stating the model. An abnormal return without its model specified is an uninterpretable number.

A third is testing many events and reporting the significant ones, which manufactures abnormality from multiple comparisons.

A fourth is confusing it with active return, which is a plain benchmark difference with no model of expected return behind it.

The question: The stock beat the model. Skill or noise?

And a fifth is treating “abnormal” as praise. It is a residual. A perfectly explained excellent return is still an excellent return.

One practical note on the window. Event studies pick a period around the event — often a few days either side — and the choice of window changes the result. Too narrow and the effect is missed; too wide and unrelated news contaminates it. That choice is made by the researcher, before the answer is known, and it is rarely reported as the judgment call it is.

Active return covers the benchmarked version without a model. Absolute return covers the figure you actually spend. And expectancy covers how many observations are needed before any average means anything.

The practical check

Abnormal is a technical word that sounds like a judgment. It does not mean good or unusual in any everyday sense — it means a model did not predict it, and models are wrong constantly, which makes abnormality as much a statement about the model as about the market.

— Michael Whitman

This page is educational, not financial advice. Test every idea on your own charts before risking money.