Learn / How to backtest an NFL betting strategy
How to backtest an NFL betting strategy
A backtest answers one question: if you had followed this rule every week, what would have happened? Done well, it stops you betting real money on an idea that only sounds good. Done badly, it makes almost anything look like a winner. Here is how to do it well.
1. Write the rule down precisely
"Fade the public" is a feeling, not a rule. A backtest needs something a computer could follow without judgment: bet the home team against the spread when it's an underdog of 3 points or more. Every part of that sentence is checkable for every game: who was home, who was favored, and by how much.
In the backtester that rule is three choices: Market = Spread, Bet on = Underdog, Playing at = Home, then a team's spread of 3 or more.
2. Start broad, then add one filter at a time
Run the plain version first ("all home underdogs") and note the result. Then add a single filter, like divisional games only, and see what changes. Adding filters one at a time shows you which part of the idea is doing the work. If a filter cuts your sample in half and the results barely move, it probably isn't adding anything.
3. Read the results properly
- Record and win rate. Wins, losses and pushes. Win rate ignores pushes.
- Break-even. The win rate you needed at the odds you actually got. At standard -110 odds that's 52.4%, not 50%. A 51% system loses money. See break-even win rate.
- Units and ROI. Units is total profit when you risk 1 unit per bet. ROI is profit per unit risked: +3% means you made 3 cents for every dollar bet over the long run.
- Max drawdown. The worst peak-to-bottom drop. Even good systems have long losing stretches; this tells you how long a rough patch you'd have had to sit through.
4. Run the luck check
A system can show a profit purely by chance, especially over a few hundred bets. The luck check estimates how often a strategy with no real edge would do at least this well. 30% means "this happens to coin-flip strategies all the time". 1% means luck is a less likely explanation. It's explained in skill or luck?
5. Test it on seasons it hasn't seen
The most honest check is to find a rule on older seasons, then see how it did on later seasons you didn't look at while building it. If it only worked in the seasons you used to design it, you probably fitted the noise. The strategy finder does this automatically for thousands of rules at once, and most of the best-looking ones fade.
Common mistakes
- Stacking filters until it works. With enough filters, any sample can be carved into a "winning" subset. Each filter should have a reason you could explain before seeing the result.
- Ignoring the juice. Testing as if every bet paid even money hides the 4–5% the sportsbook keeps.
- Using lines you couldn't have bet. Snap Theory uses closing lines, the final number before kickoff, so every result reflects a price that really existed. See closing lines explained.
- Trusting small samples. 40–20 looks great and means very little. Aim for hundreds of bets before reading much into a result.
- Assuming the past repeats. Markets adapt. A pattern that worked in 2005 may have been noticed and priced in since.
Snap Theory is a research tool, not betting advice. Past results don't guarantee future results. 21+.