Paper Betting: Testing a Strategy Without Risking Money
The only test a strategy cannot cheat on: bets written down before kickoff, at real prices, with pretend money. How we run it, our results so far, and how to do it yourself.
Last reviewed November 3, 2026
Key takeaways
- >A paper bet is recorded before kickoff, so hindsight cannot leak in
- >Short favourites everywhere: 264 bets, −0.6%: priced about right
- >Premier League totals under 1.30: won 86% of 238 and still lost 5.5%
- >Our AI quant strategy started 5–1 this season: far too few bets to judge
- >Write the rule, the judging test and the kill rule down before the first bet
Paper betting means placing bets on matches that have not been played yet, at real prices, but with pretend money. It sounds like a beginner's exercise. It is actually the only test a strategy cannot cheat on, because nobody, not even the strategy, knows the results when the bet is placed.
Why a backtest is not enough
A backtest replays the past. However carefully you do it, you are looking at matches whose results you already know, and that knowledge leaks in: you pick the rule that worked (overfitting), you notice it because it started well (the hot-start trap), you tweak it once more after a bad week. A paper bet is written down before kickoff. It cannot be tweaked afterwards.
How we run it
We run our strategies as paper accounts:
- Every bet is recorded before kickoff, with the price, the bookmaker, the stake (a flat €10) and the reason.
- Only prices you could really get: Toto or Unibet, at the price shown at that moment.
- Settled automatically from the real result, and never edited afterwards.
- Judged with the same luck test as every backtest: under 30 settled bets, no verdict.
Some of our strategies are simple rules. Others are run by an AI agent: once or twice a week it reviews the upcoming matches through each strategy's lens, checks the clubs' recent records, and places paper bets where the evidence agrees. Real money is never involved; the agent can only place paper bets.
The results so far
Strategy | Type | Bets | Record | Profit | ROI | Verdict |
|---|---|---|---|---|---|---|
Home/away favourites under 1.80, all leagues | Rule | 264 | 182–82 | −€15.50 | −0.6% | Within luck: no edge (retired) |
Premier League totals priced under 1.30 | Rule | 238 | 205–33 | −€131.50 | −5.5% | Losing beyond luck (retired) |
Eredivisie: Over 2.5 on every match | Rule | 80 settled | 60–20 | +€60.70 | +7.6% | Within luck (±14%) (retired) |
"Quant": an expected-goals model | AI agent | 6 settled | 5–1 | +€24.10 | +40.2% | Too few to judge (±57%) |
What this table says, honestly:
- The simple rules all failed to show an edge. Backing short favourites everywhere finished almost exactly level over 264 bets: the market prices favourites about right. The Premier League's very short totals won 86% of the time and still lost, because at an average price of 1.10 you need about 91%.
- The Eredivisie overs rule made money, but not enough to rule out luck. +7.6% on 80 bets is inside the ±14% band.
- The AI strategies are too young to judge. The quant model started 5–1, which is exactly the kind of start that means nothing yet (see How many bets before a result means something?). We will keep publishing its record either way.
How to paper-bet yourself
- Write the rule down first: competition, market, price range, stake, and when you stop.
- Record every bet before kickoff, with the price you could actually get. A spreadsheet is enough.
- Never delete or "correct" a bet afterwards. A void is only for a match that was not played.
- Decide in advance how you will judge it: for example, at least 30 bets, and a result outside the luck band.
- Decide in advance what kills it: for example, losing more than 20% after 50 bets.
Most strategies die on paper. That is the point: it is far cheaper to find out there than with real money.
Source: Edge Terminal paper accounts, flat €10 stakes, settled from official results, as of 3 November 2026 (two quant bets were still open; test bets from the World Cup in June 2026 are not included). A second AI strategy, "contrarian", has not placed a bet this season. The three rule strategies were retired on 30 October 2026 and are now followed in the backtester. Luck band: ±2 standard errors of the per-bet return. Past results only; nothing here is a tip.