"My method works." Every bettor thinks so. But how do you tell a genuinely good strategy from a few lucky months? The answer is one word: the backtest. That is, replaying your strategy across years of past matches to see what it would have produced.

Trap number one: randomness in disguise

The problem is that you can always find a rule that would have worked in the past. Real and absurd examples: "back the home team in the Bundesliga when the moon is waxing" → +12% over five years… then −3% over the next five. Or "bet on teams starting with M on Tuesdays". Those rules capture no real pattern, only noise. On new data, they collapse.

The rules of an honest backtest

  • Test on data never used. Build the strategy on one period (say 2014-2021) and check it on another, set aside (2022-2024). If it works on the first but not the second, the strategy is worthless.
  • Simulate reality. Better still: retrain the model month after month and bet only the following month, exactly as you would use it for real. The results are less flattering, but honest.
  • Count enough bets. Over 50 bets, anything can look like a winner. You need at least a thousand to believe it.

The numbers to look at (not just the profit)

  • Return (ROI): between +3% and +10% is realistic over time. Above +15%, be suspicious — it is usually too good to be true.
  • Hit rate alone is misleading: winning 80% of your bets at odds of 1.10 loses money; winning 30% at 4.00 makes money.
  • The worst drawdown: if the strategy is 30% down at its lowest point, will you hold on without cracking? Often, no.

The hidden enemy: variance

Even a good strategy — say a +5% return — goes through long, painful losing runs. That is mathematically unavoidable. It is why you stake small (half Kelly): to survive the bad stretches without blowing up your bankroll, and without quitting at the worst moment while the method is sound.

How PROLIFICK validates its models

Our models are tested on a lot of data: around ten NBA seasons, several in football and tennis, tens of thousands of matches. We apply the strict "retrain every month, bet the next month" rule, check the numbers sport by sport, and even test the odd periods (disrupted Covid calendars, finals). Finally, we make sure that when the model says "70%", the event really happens about 70% of the time.

That is not a promise of profit — variance is real — but the assurance that the method holds up. Backtesting is exactly what separates the bettor who *believes* he has a method from the one who *knows* he does. The rest is value and discipline.