How good are these forecasts?
Every forecast scored on seasons the model never saw while its settings were chosen.
Held-back seasons
| Forecast | Forecasts | Log loss | Favourite won |
|---|---|---|---|
| The model | 2119 | 0.586 | 67% |
| Base rate | 2119 | 0.702 | 48% |
| Recent form | 2119 | 0.648 | 62% |
Against the betting markets
Each market's price before the start, scored on the same finished games as the model. A negative figure means the market was more accurate.
| Market | Since | Priced | Model log loss | Market log loss | Model vs market |
|---|---|---|---|---|---|
| Bookmakers (Pinnacle, DraftKings, bet365) | 7 Feb 2026 | 39 | 0.535 | 0.390 | −37.1% |
Every nation is rated on all its men's T20 internationals, Elo-style, with the margin of victory (a win by wickets counts about 8 runs a wicket).
A season here is a World Cup cycle: every international from the day after one final to the next. Bilateral series are forecast and rated but aren't in the event's table.
The 2028 Men's T20 World Cup is simulated in full in every run: four groups of five play each other once, the top two in each go through to two Super 8 groups (a group winner with the other pairs' runners-up), the top two of each Super 8 group play the semi-finals, then the final.
The field isn't settled yet, so this is a provisional one: the hosts and the nations that would qualify on today's ratings, seeded into groups by rating. It changes as results come in and will be replaced by the real draw once it's announced.
Australia and New Zealand host the 2028 World Cup. The ICC hasn't published its dates, draw or full field yet, so the dates here are placeholders and the field is provisional: the hosts and the other eighteen nations from the 2026 World Cup, best-rated first.
Every international is treated as neutral: home advantage isn't modelled, because the archive doesn't say who was at home. That flatters touring sides a little.
Squad selection (rested stars, part-time associate sides) is not in the model.
Associate nations play each other far more than they play the full members, so their ratings are only loosely tied to the top sides'.
The champion record replays past World Cups from the archive, so a match abandoned without a result (or one the archive is missing) isn't in the replay, and sides level on points are split by run margins rather than net run rate: a replayed group can finish in a different order from the real one. An edition whose knockouts the archive can't piece together is left out.
Questions and answers
How accurate are PodiumOracle's ICC Men's T20 World Cup predictions?
On 2,119 games from held-back seasons (Road to 2024 to Road to 2028), which the model never saw while it was tuned, the side it favoured won 67% of the time. Its log loss was 0.586, against 0.702 for the base rate and 0.648 for recent form. Against the betting markets on the same finished games, its log loss was 0.535 against the bookmakers' 0.390 over 39 games (lower is better).