Do betting odds tell the truth?
18,238 settled two-outcome markets, five sports, November 2024 – July 2026 ·
Published 3 August 2026 · Companion dataset free under
CC BY 4.0
Teaching dataset: one row per finished game for six leagues, with the score, our
model's pre-game numbers and the closing line —
download the CSV (free, CC BY 4.0).
Betting odds are predictions with money on them. Strip out the bookmaker's margin and every
price becomes a probability — a claim about how often something should happen. We tested
those claims against what actually happened, across every settled two-outcome market in our
archive: both sides of 18,238 markets, scored against real results.
36,476
outcome observations (both sides of every market)
±2 pts
how closely outcome rates track vig-free probabilities, in every decile
+2 pts
how much longshots overperform their vig-free price
−2 pts
how much favourites underperform theirs
Finding one: the market is a very good forecaster
Across every probability band, real outcome frequencies land within about two percentage
points of the market's vig-free probabilities. A thing the market prices as a 65% chance
happens about 63% of the time; a 25% chance happens about 27% of the time. As a pure
forecasting machine, the betting market is impressively calibrated — which is exactly
why beating it is hard.
| Vig-free probability band | Observations | Mean implied | Actually happened | Gap |
|---|
| 10–20% | 2,144 | 17.5% | 18.8% | +1.4 |
| 20–30% | 5,263 | 25.2% | 26.7% | +1.4 |
| 30–40% | 4,909 | 34.7% | 37.0% | +2.2 |
| 40–50% | 5,515 | 45.7% | 46.4% | +0.7 |
| 50–60% | 6,077 | 53.8% | 53.2% | −0.6 |
| 60–70% | 4,945 | 65.2% | 63.0% | −2.2 |
| 70–80% | 5,251 | 74.7% | 73.3% | −1.4 |
| 80–90% | 2,157 | 82.5% | 80.9% | −1.6 |
Finding two: a small tilt toward the longshots
The residual pattern is systematic, not noise: outcomes priced below 50% happen
slightly more often than their vig-free probability suggests, and favourites slightly less.
Once the margin is taken out evenly, longshots are priced a touch too low and
favourites a touch too high — by about one and a half points on the less likely
side, roughly the same across the bands. That is the opposite direction to the classic
favourite–longshot bias seen in racetrack betting, where longshots are overpriced; a
reversed or flat bias has been reported before in fixed-odds markets. It is small: well inside
the cut the bookmaker takes (see our margin
measurements), so it is not something a bettor can profit from.
The tilt is roughly twice as large in player-prop markets (±5 points in the 30–40%
band) as in match-level markets — the same markets that carry the
highest margins.
What this means for a bettor, in plain English
- The market's probabilities are mostly honest once you remove the cut.
What happens is within about two points of what the price says.
- A multi costs more with every leg — each leg carries its own
cut, and the cuts multiply.
- "The bookies always know" is roughly true — to within about two
points. Which is why disciplined price comparison matters more than hunches.
Method
Every settled two-outcome market in our grading archive was included from both
sides: for each outcome at vig-free probability p, its complement enters
at 1−p with the opposite result. The dataset therefore nets to exactly
50/50 by construction and contains no information about anyone's selections — only
about the market. Probabilities were de-vigged across each market's outcomes; de-vig method
choice affects tail allocation, a caveat the dataset documents. No odds, prices, bookmaker
identities, teams or players appear in the data.
The full dataset (36,476 rows) is free under CC BY 4.0 — links to the archived copies
are in the citation block below as they come online.
Citing this study
Bet Better (2026). Do Betting Odds Tell the Truth? Market Calibration Across 18,238
Settled Markets. https://betbetter.world/studies/market-calibration
Working paper. The peer-review-format version of this research, with the full
method and results, is archived with a DOI:
Glush, E. (2026). Well Calibrated, Unevenly Priced: Market Calibration and the
Favourite–Longshot Tilt in Online Fixed-Odds Betting. Zenodo.
https://doi.org/10.5281/zenodo.21818845
See it on today's games
We run the same price checks on every game we cover. Today's picks, with the chance in plain words and the price, are here:
NFL picks · NBA picks ·
MLB picks · NHL picks ·
college football picks · AFL tips ·
NRL tips · Premier League picks ·
tennis picks · NBA player props ·
today's top picks across every sport.
Companion research: the Bookmaker Margin Study ·
How much do betting lines actually move? ·
Australian gambling statistics ·
the free model API.
18+. Nothing here is betting advice. If the
market is calibrated to within two points and carries a
4.78% margin, the average bet loses by construction
— that is the honest arithmetic of gambling. If it is causing you harm:
Gambling Help Online or
1800 858 858.