Research project · preseason forecast
Can 2025 data predict the 2026 MLB season?
Every 2026 game was forecast before Opening Day using only results through 2025, then checked against what happened.
Regular-season games where the forecast picked the winner
55.3%
1,343 of 2,429 games
- 62% when it gave a team 60% or more (389 games)
- 70% when it gave a team 65% or more (101 games)
- 52.9% if you always pick the home team
- 56.4% in the postseason (22 of 39 games, by series)
- Win totals
- ±6.6
- wins off, per team on average
- Playoff teams called
- 8 of 12
- Inside 80% range
- 26 of 30
- teams; about 24 expected
- World Series (in progress)
- MIL leads 3–1
- vs TB; MIL given 10.0% preseason
In brief
- Data: final scores of 61,628 games, 2000–2025, plus the 2026 schedule. No 2026 results.
- Model: four candidates tested on 2010–2025; the best, an Elo rating, made the forecast.
- Simulation: the season was played 100,000 times to get records and playoff odds.
- Why not higher? Baseball is close to a coin flip — even the best team loses 40% of the time — and a preseason rating can’t see trades, injuries or starting pitchers.
Projected vs. actual wins
Team80% range
Biggest surprises
| Team | Projected | Actual | Diff |
|---|---|---|---|
| Tampa Bay Rays | 79.1 | 98 | +18.9 |
| Chicago White Sox | 71.2 | 84 | +12.8 |
| Atlanta Braves | 82.1 | 94 | +11.9 |
Biggest disappointments
| Team | Projected | Actual | Diff |
|---|---|---|---|
| San Francisco Giants | 81.3 | 65 | −16.3 |
| Athletics | 78.8 | 64 | −14.8 |
| Kansas City Royals | 83.2 | 69 | −14.2 |
Model comparison
| Method | 2026 accuracy | 2026 wins off | 2026 log loss | Backtest log loss |
|---|---|---|---|---|
| ● Elo, carried over | 55.3% | ±6.6 | 0.6845 | 0.6832 |
| Team Marcel (3-yr Pythagorean) | 55.4% | ±6.9 | 0.6848 | 0.6843 |
| Last season's W% | 54.4% | ±8.9 | 0.6927 | 0.6903 |
| Coin flip + home field | 52.9% | ±8.5 | 0.6917 | 0.6909 |
Preseason title favourite: Milwaukee Brewers (10.0%). Full evaluation →