Why sports
betting doesn't
work.
We placed 22,900 simulated $100 bets across six football leagues and five seasons, tested 12 betting strategies and built 2 real statistical models, all walk-forward, out-of-sample, and against real market odds. Every league lost. Both models lost everywhere. And no rule that went green in one league survived the other five. This page is the receipts.
No tilt, no chasing, no bad luck.
Just the margin, compounding.
Bet on everyone. Lose anyway.
The dumbest possible strategy, as a control: a flat $100 on every team, in every match, at market-average odds. If odds were fair prices, this should roughly break even. It doesn't, because the bookmaker's margin (the “vig”) is baked into every single price. Pick any league, any season: the result never changes sign.
Net profit / loss by team · Premier League 2025-26
380 matches, 760 bets of $100. A handful of teams finish green. The rest bleed far more than the winners make.
Full per-team breakdown
| Team | Bets | Wins | Win % | Staked | Returned | Net | ROI |
|---|---|---|---|---|---|---|---|
| Aston Villa | 38 | 19 | 50.0% | $3,800 | $5,492 | +$1,692 | +44.53% |
| Sunderland | 38 | 14 | 36.8% | $3,800 | $5,191 | +$1,391 | +36.61% |
| Man United | 38 | 20 | 52.6% | $3,800 | $4,972 | +$1,172 | +30.84% |
| Everton | 38 | 13 | 34.2% | $3,800 | $4,263 | +$463 | +12.18% |
| Arsenal | 38 | 26 | 68.4% | $3,800 | $3,818 | +$18 | +0.47% |
| Bournemouth | 38 | 13 | 34.2% | $3,800 | $3,779 | −$21 | −0.55% |
| Brentford | 38 | 14 | 36.8% | $3,800 | $3,774 | −$26 | −0.68% |
| Fulham | 38 | 15 | 39.5% | $3,800 | $3,590 | −$210 | −5.53% |
| Man City | 38 | 23 | 60.5% | $3,800 | $3,551 | −$249 | −6.55% |
| Nott'm Forest | 38 | 11 | 28.9% | $3,800 | $3,436 | −$364 | −9.58% |
| Brighton | 38 | 14 | 36.8% | $3,800 | $3,424 | −$376 | −9.89% |
| West Ham | 38 | 10 | 26.3% | $3,800 | $3,074 | −$726 | −19.11% |
| Newcastle | 38 | 14 | 36.8% | $3,800 | $3,010 | −$790 | −20.79% |
| Crystal Palace | 38 | 11 | 28.9% | $3,800 | $3,003 | −$797 | −20.97% |
| Leeds | 38 | 11 | 28.9% | $3,800 | $2,956 | −$844 | −22.21% |
| Liverpool | 38 | 17 | 44.7% | $3,800 | $2,862 | −$938 | −24.68% |
| Tottenham | 38 | 10 | 26.3% | $3,800 | $2,733 | −$1,067 | −28.08% |
| Chelsea | 38 | 14 | 36.8% | $3,800 | $2,625 | −$1,175 | −30.92% |
| Burnley | 38 | 4 | 10.5% | $3,800 | $1,579 | −$2,221 | −58.45% |
| Wolves | 38 | 3 | 7.9% | $3,800 | $1,273 | −$2,527 | −66.50% |
12 strategies, 12 ways to lose
Every piece of betting folk wisdom you've heard (back the favorite, home teams win, ride the big clubs, take a flyer on longshots), run as a mechanical rule with flat $100 stakes. Browse any league and season; the occasional green rule in a single season never survives the ALL view.
ROI by strategy · Premier League 2025-26
1 rule green in this slice. Check the ALL view before getting excited.
Full comparison
| Strategy | Bets | Win % | Staked | Net | ROI |
|---|---|---|---|---|---|
| Back the draw, every match | 380 | 27.4% | $38,000 | +$2,473 | +6.51% |
| Back strong favorites only (odds <= 1.50) | 69 | 71% | $6,900 | −$294 | −4.26% |
| Back home team only when home is favorite | 245 | 52.7% | $24,500 | −$1,562 | −6.38% |
| Back home team, every match | 380 | 42.6% | $38,000 | −$3,627 | −9.54% |
| Back the favorite, every match | 380 | 49.5% | $38,000 | −$3,723 | −9.80% |
| Back every team (baseline) | 760 | 36.3% | $76,000 | −$7,595 | −9.99% |
| Back the underdog, every match | 380 | 23.2% | $38,000 | −$3,872 | −10.19% |
| Back away team, every match | 380 | 30% | $38,000 | −$3,968 | −10.44% |
| Back big club only when favorite | 182 | 53.3% | $18,200 | −$2,408 | −13.23% |
| Back away team only when away is favorite | 135 | 43.7% | $13,500 | −$2,161 | −16.01% |
| Back home team when odds 1.50-2.00 ('home banker') | 104 | 48.1% | $10,400 | −$1,701 | −16.36% |
| Back longshots only (odds >= 5.00) | 112 | 10.7% | $11,200 | −$3,658 | −32.66% |
The green bars are a trap
“Back the draw, every match” was the only strategy in the green in the 2025-26 Premier League (+6.51%). It had been green the season before too (+2.27%). Two profitable seasons running: that's exactly how betting systems get believed. Flip through the seasons and leagues, then hit ALL, and watch the “edge” refuse to show up twice in the same place.
Same strategy, five seasons · Premier League
Green seasons keep the dream alive; the 5-season aggregate for Premier League sits at −3.55%.
Season by season
| Season | Matches | Draws | Draw % | Staked | Net | ROI |
|---|---|---|---|---|---|---|
| 2021-22 | 380 | 88 | 23.2% | $38,000 | −$3,330 | −8.76% |
| 2022-23 | 380 | 87 | 22.9% | $38,000 | −$3,048 | −8.02% |
| 2023-24 | 380 | 82 | 21.6% | $38,000 | −$3,701 | −9.74% |
| 2024-25 | 380 | 93 | 24.5% | $38,000 | +$864 | +2.27% |
| 2025-26 | 380 | 104 | 27.4% | $38,000 | +$2,473 | +6.51% |
Fine. We built real models.
Forget folk wisdom: this is how quants actually attack a market. An Elo rating system (chess-style strength ratings with home advantage and margin-of-victory) and a Dixon-Coles Poisson goal model (the academic standard for football scorelines). Trained walk-forward on a decade of matches in each of the six leagues, refit before every season with zero lookahead, betting only when the model sees value beyond the edge threshold.
Higher threshold = the model must see a bigger mispricing before it bets. ALL aggregates the five out-of-sample seasons.
Elo value model
2025-26Dixon-Coles goal model
2025-26ROI by season at 5% minimum edge · Premier League
An occasional green season keeps hope alive. No league's 5-season aggregate ever goes positive, for either model.
Elo vs the market: log-loss
Lower = sharper predictions. The de-vigged market wins in 59 of 60 league-season comparisons across both models.
Dixon-Coles vs the market: log-loss
The lone exception: Bundesliga 2022-23, where Dixon-Coles edged the market, its best season anywhere. It proves the rule.
The model isn't stupid. That's the point
Dixon-Coles attack/defense ratings entering the latest Premier League season (home advantage 0.138, ρ -0.004). It ranks the giants and the relegation fodder about where you'd put them. A model can be a perfectly sensible description of football and still know nothing the market hasn't already priced.
“Just bet win-or-draw then”
The safest-feeling bet in football: double chance, where your team only has to not lose. Your hit rate rockets from roughly 38% to 62%. And the money? Derived double-chance odds across every league say: you lose slower, not less surely.
Baseline ROI by season: win-only vs win-or-draw · Premier League
Flat $100 on every team, both formats.
Does it rescue the models? No.
Elo and Dixon-Coles hunting value in the 1X/X2 double-chance markets instead: Premier League aggregate ROI at every edge threshold.
| Min. edge | Elo · win-only | Elo · win-or-draw | DC · win-only | DC · win-or-draw |
|---|---|---|---|---|
| 0% | −9.31% | −7.83% | −9.71% | −9.17% |
| 3% | −10.94% | −11.89% | −13.76% | −12.77% |
| 5% | −9.93% | −13.13% | −15.64% | −14.00% |
| 8% | −10.42% | −13.27% | −17.99% | −14.32% |
The best loser: Asian Handicap
Asian Handicap gives one team a virtual goal head-start and removes the draw entirely. The Dixon-Coles model prices the real historical handicap lines from its full scoreline grid, correctly handling pushes and quarter-lines, against genuine market prices. This is the closest anything on this page gets to break-even. It still loses, in every league that has the market.
ROI by edge threshold · Premier League 2025-26
One season in isolation can go green. That's the trap; hit ALL for the honest picture.
Every season, every threshold · Premier League
| Season | Min. edge | Bets | Wins | Staked | Net | ROI |
|---|---|---|---|---|---|---|
| 2021-22 | 0% | 304 | 147 | $30,400 | +$711 | +2.34% |
| 2021-22 | 3% | 268 | 127 | $26,800 | +$213 | +0.79% |
| 2021-22 | 5% | 244 | 117 | $24,400 | +$435 | +1.78% |
| 2021-22 | 8% | 208 | 98 | $20,800 | −$1 | +0.00% |
| 2022-23 | 0% | 311 | 136 | $31,100 | −$3,699 | −11.89% |
| 2022-23 | 3% | 262 | 113 | $26,200 | −$3,472 | −13.25% |
| 2022-23 | 5% | 243 | 107 | $24,300 | −$2,672 | −11.00% |
| 2022-23 | 8% | 213 | 95 | $21,300 | −$2,164 | −10.16% |
| 2023-24 | 0% | 305 | 142 | $30,500 | −$699 | −2.29% |
| 2023-24 | 3% | 276 | 128 | $27,600 | −$893 | −3.23% |
| 2023-24 | 5% | 250 | 115 | $25,000 | −$756 | −3.02% |
| 2023-24 | 8% | 211 | 92 | $21,100 | −$1,743 | −8.26% |
| 2024-25 | 0% | 285 | 131 | $28,500 | −$1,145 | −4.02% |
| 2024-25 | 3% | 252 | 117 | $25,200 | −$851 | −3.38% |
| 2024-25 | 5% | 233 | 107 | $23,300 | −$1,000 | −4.29% |
| 2024-25 | 8% | 197 | 85 | $19,700 | −$1,632 | −8.28% |
| 2025-26 | 0% | 281 | 131 | $28,100 | +$41 | +0.14% |
| 2025-26 | 3% | 229 | 106 | $22,900 | +$284 | +1.24% |
| 2025-26 | 5% | 194 | 89 | $19,400 | +$453 | +2.33% |
| 2025-26 | 8% | 153 | 73 | $15,300 | +$1,305 | +8.53% |
“Okay, but corners are beatable, right?”
The eternal hope: skip the efficient headline market and beat a sleepy side market instead. Before you can beat corner betting you need to predict corners, so we tested that prerequisite directly, on Premier League data. The result is the most instructive dead end in this whole project.
Test 1: is corner-winning a real team trait?
Correlation between each team's corners-per-match in the first vs second half of a season. Yes: strongly persistent.
Test 2: can you predict a match's total?
Model-vs-actual correlation for single-match corner totals, walk-forward out-of-sample. Effectively zero, every season.
Prediction error (RMSE, corners): lower is better
A fitted Poisson attack/defense model for corners against two dumb baselines. The model never beats “guess the league average every time.”
| Season | Avg total corners | Guess league avg | Team form | Poisson model | Model ↔ actual corr. |
|---|---|---|---|---|---|
| 2021-22 | 10.42 | 3.379 | 3.407 | 3.884 | 0.006 |
| 2022-23 | 10.11 | 3.273 | 3.292 | 3.660 | 0.044 |
| 2023-24 | 10.84 | 3.543 | 3.545 | 4.081 | 0.003 |
| 2024-25 | 10.30 | 3.470 | 3.453 | 3.883 | 0.045 |
| 2025-26 | 10.00 | 3.291 | 3.288 | 3.320 | 0.080 |
It's not the league. It's the math.
Maybe the Premier League is just too well-studied? Same baseline, flat $100 on every team in every match, across six leagues on three continents, five seasons each: 22,900 bets from La Liga to MLS. Thirty league-seasons. Thirty losses. Flip through the seasons below and try to find a green one.
ROI by league · 2025-26 (MLS: 2025)
Different football cultures, different bookmaker mixes, identical outcome: the margin wins.
Ligue 1
−6.47%Premier League
−7.27%Bundesliga
−7.88%MLS
−8.28%La Liga
−8.93%Serie A
−11.08%Think you'd have beaten it?
Pick a league, a season, and a rule. We'll settle it instantly against the real odds — flat $100 stakes, no hindsight, no mercy. Find a payout, then check whether it survives ALL seasons and the other five leagues.
Place your bet
Every combination is precomputed from the same dataset as the rest of this page: real market-average odds, every match of every season, settled exactly as a bookmaker would have.
Across 5 seasons this rule is green in 1 of 6 leagues — and still fails the replication test. The house margin doesn't care which rule you pick.
“Yeah, but...”
Every counter-argument we've heard, taken seriously and answered from the data.
My mate is up $2,000 this season. Explain that.+
This page can. Back-the-draw was up +6.51% over a full Premier League season, and Asian Handicap was green for an entire year across every threshold. Individual seasons produce winners constantly; that's what variance does over a few hundred bets. The question is never "is anyone up right now?", it's "does it survive five seasons and six leagues?" Nothing here did. Also remember who you hear from: winners post screenshots, losers post nothing.
What about arbitrage and matched betting?+
Both are real, and both are something other than betting. Arbitrage margins are fractions of a percent, require capital spread across many books, and bookmakers limit or close accounts that do it well. Matched betting extracts sign-up promotions; it works exactly until the promotions or your accounts run out. Neither is "picking winners", which is the thing this site tests, and neither scales into a living for the vast majority who try.
Don't professional sharps beat the market?+
A tiny number do, and they're part of why you can't. Sharps beat the closing line using speed, private information, and models far beyond public data, and books respond by limiting their stakes and copying their information into the odds. By kickoff, the line you bet against already contains the sharps' opinions. Betting the closing line means betting against everyone smarter and faster than you, after their edge has been priced in.
Is live / in-play betting easier to beat?+
It's worse. In-play prices update in seconds from data feeds you don't have, margins are typically wider than pre-match, and the format is engineered for impulse decisions. Everything that makes the pre-match closing line hard to beat is amplified when the market re-prices every few seconds and you're deciding during a corner.
Couldn't a better model than yours win?+
The bar isn't "a good model", it's "a model better than the closing line by more than the ~5-6% margin, persistently." Our Elo and Dixon-Coles models ranked teams sensibly and still out-predicted the de-vigged market once in 60 league-season comparisons — and that one model lost money anyway. Beating it consistently takes information the market doesn't have yet: injury news seconds early, private tracking data, syndicate-scale infrastructure. That's a job, not a bet.
This is football. What about the NBA, NFL, or tennis?+
Same structure, same math. A standard US spread at -110/-110 carries about a 4.5% hold, and margins on props and parlays run far higher than anything in this study. The leagues differ; the vig doesn't. Six football leagues on three continents all clustering at −6% to −11% is the point: it's not the sport, it's the pricing.
The house isn't lucky.
It's priced.
The vig is a toll you pay on every bet
Odds sum to ~105-106% implied probability. Before skill enters the picture you are down ~5-6% per bet. Any edge you find must clear that bar first, and compounding works against you with every wager.
The odds already know what you know
Two fundamentally different statistical models, trained on a decade of data in six leagues, out-predicted the de-vigged market line once in 60 tries, and that one model still lost money overall. Public information is priced in before you can bet on it.
Hot streaks are variance wearing a costume
Back-the-draw in 2025-26 (+6.5%), favorites in La Liga, away-favorites in Serie A: test 12 rules in 6 leagues and a few cells come up green, exactly as chance predicts. Not one rule stayed green across all six leagues. If your edge doesn't replicate, it isn't an edge.
There is no soft market waiting for you
Six leagues, three continents, double chance, handicaps, corners: the least-bad model market (Asian Handicap) still lost ~3% everywhere it exists. 'Less bad' is the ceiling for a public-data bettor, and less bad still means broke, only slower.
22,900 bets. 30 league-seasons.
2 models, 6 leagues, every market.
All negative.
Sports betting isn't a skill market for the public. It's entertainment with a ~6% cover charge on every ticket. The only winning move is not to play.