The Model

How ForzaPitch predicts football matches

01  Methodology

ForzaPitch uses a Poisson regression model enhanced with the Dixon-Coles correction to predict football match outcomes. For each match, we estimate the expected goals (xG) for each team based on their attacking and defensive form, Elo rating, home advantage, and league-wide scoring averages. The Poisson distribution then gives us the probability of every possible scoreline, from which we derive 1X2, over/under, and other markets.

02  Elo Ratings

Each team carries a global Elo rating updated after every match. For cup competitions where two teams from different leagues meet, we use Elo-weighted league baselines to correct for the difference in competitive level.
Weekly accuracy — last 8 weeks
40% 55% 70% 16 Mar 23 Mar 30 Mar 6 Apr 13 Apr 20 Apr
Period average: 58.2%
63.9%
Global Accuracy
Analysed: 901 matches
Calibration

A well-calibrated model predicts 60% probability events correctly about 60% of the time. The table below compares our predicted confidence bands against actual outcomes.

25% 50% 75% 100% 0% 25% 50% 75% 100% perfect 55% previsto → 58.4% reale (n=327) 65% previsto → 63.7% reale (n=342) 75% previsto → 68.3% reale (n=161) 85% previsto → 80.6% reale (n=62) 95% previsto → 77.8% reale (n=9) Predicted probability Actual frequency
model perfect calibration ● <7pp deviation: good / mid / far
Confidence band Matches Expected (mid) Actual
50–60% 327 55.0% 58.4%
60–70% 342 65.0% 63.7%
70–100% 232 85.0% 72.0%
Performance by League
72.7%
71.4%
61.1%
77.8%
64.5%
61.5%
100.0%
66.7%
66.7%
50.0%
61.1%
50.0%
25.0%
25.0%
100.0%
62.5%
80.0%
MLS
72.2%
72.4%
75.0%
76.5%
0.0%
HNL
66.7%
67.5%
42.9%
63.2%
100.0%
52.9%
41.7%
16.7%
100.0%
60.0%
100.0%
55.6%
100.0%
66.7%
100.0%
100.0%
71.4%
33.3%
58.3%
33.3%
55.6%
FNL
25.0%
50.0%
100.0%
0.0%
100.0%
100.0%
75.0%
80.0%
80.0%
66.7%
87.5%
50.0%
66.7%
75.0%
100.0%
Cup
0.0%
100.0%
100.0%
0.0%
50.0%
100.0%
33.3%
75.0%
100.0%
100.0%