Racing Odds Calibration — How Accurate Are Australian Racing Prices?
Every price is a probability claim. We tested 45,334 of them against actual results — the shortest prices are priced almost to the decimal, while runners at 21.00–33.99 win 32% more often than the market implies.
Updated · 8 min read
Key takeaways
- The archive covers 62,543 Australian races and 503,406 runner records from 1 June to 31 October 2026.
- 53,503 runners carried an exchange starting price and 322,166 carried a finishing position; the band table uses the 45,334 runners with both.
- Runners priced 1.01–1.99 won 64.05% of the time against a field-normalised implied 64.94% — effectively perfectly calibrated.
- Runners priced 21.00–33.99 won 5.64% against an implied 4.28%: understated by 32% in relative terms on a sample of 4,453.
- Field size moves the favourite more than price does: the shortest-priced runner wins 49.54% in 4–6 runner fields and 31.68% in 13+.
What we measured
We took every completed Australian thoroughbred race in our archive — 62,543 races, 503,406 individual runner records, from 1 June to 31 October 2026 — and asked a single question: when the market prices a horse at X, how often does that horse actually win?
Of those 503,406 records, 53,503 carry an exchange starting price (the price the exchange settles at when the race jumps) and 322,166 carry a finishing position. The band table below uses the 45,334 runners where both are present and the race had at least four priced runners, so every horse in the sample is competing in a genuinely priced market.
Why the raw implied probability overstates the market
Implied probability is simply 1 divided by the price. Price a horse at 4.00 and you are claiming a 25% win chance. Add up every horse in a race and, on a fair book, the total should come to exactly 1.00.
It does not. Across 7,566 fully priced races the field summed to 0.9401 — about 6% under even money. That means every raw implied figure is 6% too low before we even start, and a naive comparison will make the market look generous in every price band.
So we normalised. For each runner we divided its implied probability by the total of its own race, which removes the under-round and leaves a clean per-horse claim. The "implied" column below is that field-normalised figure, and it is the honest one to compare against.
Actual win rate by price band
45,334 priced runners, 1 June – 31 October 2026. "Actual" is how often they won; "implied" is the field-normalised market claim; "gap" is the relative difference.
Price band · sample · actual vs implied
- 1.01–1.99 · 1,822 runners · won 64.05% vs implied 64.94% · gap −1.4%
- 2.00–6.99 · 13,981 runners · won 29.11% vs implied 28.04% · gap +3.8%
- 7.00–11.99 · 7,375 runners · won 13.88% vs implied 12.28% · gap +13.0%
- 12.00–20.99 · 6,789 runners · won 7.57% vs implied 7.20% · gap +5.1%
- 21.00–33.99 · 4,453 runners · won 5.64% vs implied 4.28% · gap +31.8%
- 34.00 and longer · 10,914 runners · won 1.87% vs implied 1.46% · gap +28.1%
Read the top and bottom of that table together and the pattern is clear: the market is near-perfect at the short end and progressively soft as prices lengthen. This is the classic favourite–longshot bias, and it is still alive on Australian racing in 2026.
The short prices are priced to the decimal
The 1,822 runners priced between 1.01 and 1.99 won 64.05% of the time. The market claimed 64.94%. That is a gap of 0.89 of a percentage point, or 1.4% in relative terms — inside the noise you would expect from a sample of that size.
For a punter this is the important negative result. Backing short-priced runners is not a hidden edge and it is not a hidden trap: over 1,822 runners the price is doing exactly what it says on the tin. Any strategy built on "favourites are overbet" has to beat a market that is currently calibrated to within a percent at the short end.
Where the market is softest: 21.00–33.99
The 21.00–33.99 band is the standout. 4,453 runners in that range won 5.64% of their races against an implied 4.28%. In relative terms the market understates this band by 31.8% — the largest gap in the table.
The longest band is close behind: 10,914 runners at 34.00 and longer won 1.87% against an implied 1.46%, a 28.1% relative understatement. Note that this band is the second-largest sample in the table, so it is not a small-sample artefact.
The 2.00–6.99 band — where the overwhelming majority of money sits, 13,981 runners — is priced almost exactly right, at +3.8%. And 7.00–11.99 shows a genuine +13.0% on 7,375 runners, which sits between the two regimes rather than contradicting them.
Field size moves the favourite more than price does
The same archive lets us separate price from arithmetic. We grouped races by how many priced runners contested them and asked how often the shortest-priced runner in the race won.
Favourite strike rate by field size
- 4–6 runners · 4,013 races · favourite won 49.54%
- 7–8 runners · 2,052 races · favourite won 40.69%
- 9–12 runners · 1,055 races · favourite won 38.29%
- 13 or more runners · 101 races · favourite won 31.68%
A favourite in a six-horse field wins roughly half the time. The same favourite in a 13-horse field wins less than a third. That is an 18-point swing driven purely by how many rivals it has, and it dwarfs the calibration gaps in the band table.
It also explains why "the favourite" is not a strategy. Two runners both priced at 3.00 are not the same bet if one is in a five-horse field and the other is in a 16-horse field, and any staking plan that treats them identically is ignoring the single biggest variable in the data.
What this means for a bet
Three practical readings. First, do not look for an edge in short-priced runners — the market is calibrated there to within a percentage point, and you are paying commission on top. Second, the softest part of the book is the 21.00-and-longer region, which is also where bookmaker margins are widest, so an edge on paper has to survive a much bigger over-round before it is real money. Third, field size should be an input to any model or staking decision: it moves a favourite further than anything else in this dataset.
The honest caveat is that this is a measurement of the market, not a licence to bet longshots. A band that wins 32% more often than implied can still be a losing band once the bookmaker margin, commission and variance are applied. The number to take away is the shape: sharp at the top, soft at the bottom.
Method and limits
Sample: 45,334 runners with both an exchange starting price above 1.01 and a recorded finishing position, drawn from 62,543 Australian races between 1 June and 31 October 2026. Races with fewer than four priced runners were excluded.
Normalisation: each runner's implied probability was divided by the sum of implied probabilities for its own race (mean field sum 0.9401 across 7,566 races), removing the under-round before comparison.
Limits: because the band table requires both a price and a finishing position, it describes the subset of runners where both are recorded rather than every runner in every race — 53,503 of 503,406 records carry a price. Treat the figures as conditional on that coverage. Win rates are per-runner, so races with more runners contribute more rows. Figures are as at 30 October 2026 and are recomputed from the archive, not hand-picked.
FAQ
Do favourites win more often than their odds suggest?
No — the opposite. Runners priced 1.01–1.99 won 64.05% of the time against a field-normalised implied 64.94%, a gap of about 1.4% in relative terms. Short prices on Australian racing are calibrated to within roughly a percentage point.
What is the favourite–longshot bias?
It is the tendency of markets to understate longshots and, at the extreme, overstate the very shortest prices. In our archive the 21.00–33.99 band won 5.64% against an implied 4.28% (+31.8% relative) and the 34.00+ band won 1.87% against 1.46% (+28.1%), while the 1.01–1.99 band came in slightly below its implied rate.
How often does the favourite win a horse race?
It depends almost entirely on field size. The shortest-priced runner won 49.54% of 4–6 runner races, 40.69% of 7–8 runner races, 38.29% of 9–12 runner races and 31.68% of races with 13 or more runners in our 1 June – 31 October 2026 sample.
Why do the implied probabilities not add up to 100%?
Because a book carries an under-round or over-round. Across 7,566 fully priced races the field summed to 0.9401 — about 6% under even money — so we normalise each runner by its own race total before comparing implied probability to actual win rate.