No. 04Event analysis

Event Study.

How much did the stock really react to the print? Abnormal returns, properly measured.

Run it in browserRead the methodology
Event day · τ = 0CAAR 5.93%-10-5051015202530relative day τAAR 2.10%-0.12%CAAR 5.98%-0.15%Aggregate abnormal returns, N = 12
Fig. 03Aggregate abnormal returns around N = 12 earnings events. Bars are AAR; the line is CAAR. Solid bars are statistically significant at the 5% level; faded bars are not. Notice the muted run-up before τ=0 — information leakage in the days ahead of the print.
What it does

A microscope on the print.

Pick a stock and a list of event dates — earnings prints, FOMC days, M&A announcements, anything you can timestamp. The tool fits a market model on a clean pre-event estimation window, projects what each day in the event window should have returned, and reports the abnormal residual.

Aggregate the residuals across events and you get the publication-quality plot you see above: AAR per relative day, with the cumulative CAAR overlay and per-day t-stats telling you which days are statistically real. The bars and the line are the evidence; everything else is interpretation.

Methodology

The market model, then the average.

01
Estimate β on the pre-window

For each event date t*, OLS-fit r_stock = α + β·r_SPY on bars [t* − preDays − estimationDays, t* − preDays). Default: 120 estimation bars, ending 10 days before the event.

02
Compute the daily abnormal return

For every τ in [−preDays, +postDays], expected = α + β·r_SPY(τ); AR(τ) = actual(τ) − expected. CAR(τ) is the running sum.

03
Average across N events

AAR(τ) = mean over events of AR(τ). CAAR(τ) is cumulative AAR. Aligns every event at τ=0 regardless of calendar date.

04
Significance

t-stat at τ = AAR(τ) ÷ (SD across events of AR(τ) ÷ √N). Days with |t| ≥ 1.96 are flagged in the chart.

Reference: MacKinlay, A. Craig. “Event Studies in Economics and Finance.” Journal of Economic Literature, 35 (1997).
Questions

Frequently asked

What is an event study?
An event study measures whether a particular event — an earnings release, a rate decision, a takeover announcement — moved a security more than its normal relationship to the market would predict. The toolkit dates back to Fama, Fisher, Jensen and Roll (1969); MacKinlay's 1997 review is the standard reference.
What does AR, CAR, AAR, and CAAR mean?
AR is the per-day abnormal return for one event (actual minus model-predicted). CAR is its cumulative sum across the event window. AAR is the average abnormal return at each relative day τ across N events; CAAR is the cumulative AAR. The t-statistic at each τ tells you whether the average is meaningfully different from zero.
What model is used for the expected return?
The classic market model: r_stock = α + β·r_SPY + ε, estimated by OLS on a pre-event window — 120 trading days by default, ending ten days before the event. The estimated α and β are then used to project an expected return across the event window. Per-event β and R² are returned for inspection.
Where do the event dates come from?
Paste them yourself as ISO dates (most reliable, no lookback limit), or have the tool derive earnings or Fed-decision dates from our news feed. The news-derived path is constrained by the RSS lookback — best for last quarter, not for long-history backtests. For historical work, custom dates are the path.
What does the t-statistic mean here?
For each relative day τ in the event window, the t-stat is AAR(τ) divided by the cross-event standard error: SD(AR_τ) / √N. Bars with |t| ≥ 1.96 are drawn solid in the chart; below that threshold they're faded to signal noise. The convention follows MacKinlay's symmetric two-sided test.
Can I window-tune the analysis?
Yes. The estimation window (default 120 days), pre-event window (10 days), and post-event window (30 days) are all configurable. Shorter post-windows isolate the announcement effect; longer ones surface drift. Custom estimation windows let you cleanly skip earnings season for pre-earnings work.
Try

Quantify the next print.

Paste your event dates. Get publication-quality residuals in seconds.

Open the workspace