Definition
Overfitting happens when a betting model learns the noise in past data instead of the signal that can hold up in future markets. It looks strong in a backtest because it memorized quirks: one season, one injury pattern, one referee trend, or one small sample split. When the model is used on new games, that memorized pattern breaks and the edge disappears.
No standard betting payout formula defines overfitting. The test is performance out of sample: results on games the model did not train on.
Worked Example
A bettor builds an NBA totals model and tests 1,000 past games. After adding many filters, it shows a 57% hit rate on overs at -110.
At -110 odds, the break-even rate is:
110 / (110 + 100) = 52.38%
On paper, 57% looks profitable. But the bettor then tests the same model on the next 300 unseen games. It hits 151 of 300, or 50.33%. At -110, that loses money because it is below the 52.38% break-even point.
Why It Matters
Overfitting helps a bettor spot fake edge before staking real money. A model that works only on the data used to build it is not a betting advantage.