Event Pre/Post Mortem

THE EVENT STUDY is one of the foundational tools of empirical finance, introduced in its modern form by Fama, Fisher, Jensen and Roll (1969) and set out methodologically by MacKinlay (1997). Its structure is worth learning in full because this analysis is a simplified version of it and the simplification is the lesson.

A PROPER EVENT STUDY HAS FOUR PARTS.

• AN ESTIMATION WINDOW before the event, used to fit a model of the security's NORMAL return - most commonly a market model regressing the security on an index.
• AN EVENT WINDOW around the date.
• ABNORMAL RETURN, defined as actual return MINUS the return the model predicted given what the market did. This is the crux: it isolates what was specific to the event from what was simply the market moving.
• CUMULATIVE ABNORMAL RETURN over the window, tested for significance - ideally across many events, since a single event is one observation.

WHAT THIS ANALYSIS COMPUTES. Raw price drift over the three days before, and the raw move over the three days after.

SO THE GAP IS THE ABNORMAL-RETURN STEP, and it matters concretely. If the whole market rose 4% in the three days before an event, a stock that rose 4% shows a 'run-up' that had nothing to do with its event. In crypto the problem is worse, because co-movement with Bitcoin is high and most of any single asset's three-day move is the market. Without subtracting a benchmark, a pre-event drift measures the market plus the event and cannot separate them.

WHAT THE UNDERLYING QUESTION IS. Under semi-strong efficiency, public information is impounded in prices as it arrives, so an anticipated event should be largely priced BEFORE it happens and the reaction on the day should reflect only the surprise. That is the reasoning behind 'buy the rumour, sell the news': a run-up followed by a muted or negative reaction is what pricing-in looks like, not a paradox.

THREE DAYS IS A CHOICE with real consequences. Too short misses drift - post-earnings-announcement drift, documented since Ball and Brown 1968 on annual data and refined by later quarterly work, continues for weeks and months. Too long admits unrelated news. The window should be justified, and it is a convention here.

AND ONE EVENT IS ONE OBSERVATION. Any conclusion about whether 'the market' prices this KIND of event correctly needs many events; a single case is an anecdote with arithmetic.