If HDFCBANK rose 4.2% in the five days after a results announcement and the Nifty 50 rose 3.6% over the identical five days, the announcement plausibly explains 0.6 percentage points, not 4.2. Abnormal return is that residual.
AR = actual return - expected return
The whole substance of the metric sits in how you define the expected return, and that is a modelling choice you make rather than a fact you look up. Change the model and you change the answer.
The models, and what each assumes
| Model | Expected return | What it assumes |
|---|---|---|
| Raw | Zero | That drift and market moves do not matter over the window |
| Market-adjusted | The index return over the same window | That the stock moves one-for-one with the index |
| Market model | alpha + beta x r_index, fitted over a window before the event | That the stock’s historical beta is stable |
| Factor model | Fitted on market, size, value and other factors | That the chosen factor set is the right one |
The differences are not academic. Take a smallcap announcing an order win during a week when the Nifty 50 was flat but the Nifty Smallcap 250 rose 5%. Market-adjusted against the Nifty 50 credits the announcement with the whole move. Against the smallcap index, most of the abnormal return disappears. Neither is wrong, and you have to say which one you used.
Abnormal returns are usually summed across the days of a window into a cumulative abnormal return, then averaged across all events of that type to give the figure a research screen displays.
Caveats
The estimation window used to fit a market model must sit clear of the event. If the market already suspected the announcement, the pre-event days are contaminated and your beta absorbs part of the effect you were trying to measure.
Beta is unstable for exactly the securities where events matter most. A stock in a corporate action, a merger or a delisting process has a beta from the last 250 days that describes a different company.
An abnormal return still needs a base rate and a t-statistic. A mean cumulative abnormal return of 1.3% across 40 events looks like a finding until you see the standard deviation of 6% across those events, which puts the t below 1.4.
Sign conventions matter for shorts and for events that are bad news. Read the direction the study defined before interpreting a negative number as a failure.