Dimon Global Group — Predictions
This is the desk’s scored-forecast scoreboard. The desk makes a small number of specific, checkable forecasts about public events, writes each one down before the event — with a probability and a rule for deciding it — and then publishes the grade, good or bad, when the event resolves.
Three inaugural forecasts, entered in the ledger on November 2, 2026.
None of the three has reached its resolution date yet. All three remain open.
Brier score: no score yet — a scoreboard with no games played says so. The first score appears only when the first forecast resolves.
Forecasts are logged before publication. The three inaugural forecasts below were logged to the desk’s ledger file on 2026-11-02 at 03:30 CST, before this page existed. Entries are shown exactly as logged; the ledger is append-only and is never edited after logging.
“SpaceX Starship Flight 15 launches (liftoff from Starbase) on or before November 30, 2026.”
“At least one additional named storm forms in the Atlantic basin between November 2 and the official end of the 2026 season on November 30, 2026.”
“The S&P 500 closes above 7,580.07 — its close on Friday, October 30, 2026 — on the last trading day of November 2026 (Friday, November 27, 2026).”
Every forecast is logged BEFORE the event, with a probability and a resolution rule. A forecast that was not written down in advance does not count, and does not appear here.
The ledger is append-only — never edited after logging. Wording, probabilities, and resolution rules stay exactly as first recorded, including the forecasts the desk gets wrong.
Resolved forecasts are scored with the Brier score. Lower is better. The score is published with the resolution, whether it flatters the desk or not.
Forecasts are also checked for calibration: in each probability bucket, the observed frequency should approach the stated probability over time.
Brier score = mean of (probability − outcome)² — where the outcome is 1 if the forecast came true and 0 if it did not. Lower is better: a perfect forecast scores 0, and always saying 50% scores 0.25.
Calibration is the second test. Group forecasts by their stated probability — all the 30% calls, all the 75% calls — and, in each probability bucket, the observed frequency should approach the stated probability. A desk that says 75% should be right about three times in four, no more and no less.
The desk follows a narrow set of public, forecastable signals — events with clear outcomes, public reporting, and a date by which the answer is known. That discipline is the point: if an event cannot be resolved cleanly, it is not forecast here.
The desk publishes forecasts and their grades, nothing else; no security is ever recommended here.