Methodology

Event Edge forecasts the share-price reaction to scheduled corporate events, starting with earnings announcements of S&P 500 companies, and scores every forecast in public once the outcome is known.

What is measured

The reaction is the close-to-close return around the announcement:

• for an announcement after the close, from that day's close to the next session's close;
• for an announcement before the open, from the previous session's close to that day's close.

Announcements during market hours are excluded unless the company has a known convention (for example, a release before the open filed with the SEC later that morning). Prices are split- and dividend-adjusted daily closes. Alongside the reaction we record the 5- and 20-session forward returns and the same quantities for the S&P 500 ETF (SPY), giving excess returns over the market.

Where the data comes from

• Events: SEC EDGAR 8-K filings with Item 2.02 (results of operations). The acceptance time in the filing header, in Eastern Time, classifies the announcement as before the open, after the close, or during the session. Filings whose press release is not a quarterly results release (supplemental schedules, presentations, scheduling notices) are flagged and excluded.
• Expected dates: confirmed from company press releases and investor-relations pages where a date is published; otherwise estimated from the company's prior-year pattern and usual weekday, moved to the nearest trading day.
• Prices: daily bars from Alpaca (IEX feed).
• Calendar: NYSE trading days and early closes.

Every document used is stored unchanged and identified by its hash; derived numbers point back to the documents they came from.

Typical move and baseline

The typical move of a company is the median absolute reaction over its previous eight earnings announcements (at least four are required). It is a stand-in for the options-implied move until options data is added, and it is the yardstick for the main yes/no question: will the absolute reaction exceed the typical move?

Forecasts

Three forecasters produce, for each upcoming event, the probability that the absolute reaction exceeds the typical move, the probability that the reaction is positive, and the 10th, 50th and 90th percentiles of the signed reaction:

1. a statistical baseline (logistic and quantile regression on past reactions, session, recent volatility and filing cadence), fitted on events before January 2026;
2. an underwriter model reading a fixed evidence bundle;
3. a challenger model, a different model with a different prompt, reading the same bundle without seeing the underwriter's answer.

Model forecasters see only evidence available two sessions before the expected date. Every forecast is timestamped; a forecast created after the actual filing is marked late and left out of the scorecard.

Scoring

Probabilities are scored with the Brier score and log loss; quantiles with the pinball loss; calibration is shown as reliability diagrams. Events before July 2026 overlap the models' training data and are reported as a backtest, separate from the live record.

Disclaimer

Event Edge is a research and education project. Nothing on this site is investment advice or a recommendation to buy or sell any security.