What does the data say about the things racing people believe?
Beliefs, tested against the record
Favourites. Form. Market moves. Pace. Draw. Racing is full of rules people learn to trust.
Elite Pass puts those ideas through the data, and keeps the result even when the answer is inconvenient. Some survive. Some disappear. Some only work under very particular conditions.
Read the studies ↓Every question we can answer, and what we found
Ten findings, each with the sample behind it. Where the honest answer is that an idea did not survive testing, that is what the row says.
- QuestionAnswerScaleWhat it means
How often does the favourite win?
35.0%
60,671 races · 95% CI 34.6% to 35.3%
Most likely does not mean likely. The favourite loses about two races in three.
MeasuredView research →
Does a favourite's price predict how often it wins?
Yes, in every band
60,671 races · five BSP bands
Actual tracks what the price implied, band by band, across the whole range.
Market already accounted for itView research →
Do favourites win less often in big fields?
43.6% → 22.4%
60,671 races · four field-size bands
More runners, more ways to be beaten — and the price already reflects it.
Market already accounted for itView research →
Do horse racing ratings work?
Ranked well, added nothing
30,088 races · 298,444 runners · actual-to-expected 0.998
In the rating population tested, rank and separation added no reliable information beyond Betfair SP.
Market already accounted for itView research →
Do racing systems work?
0 of 227
227 systems, rules and theories examined
None has cleared the validation bar. The count includes work done before this archive existed.
No reliable effect foundView research →
How accurately does the exchange price favourites?
Slope 0.974
18,529 runners paired · 95% CI 0.925 to 1.023
Close to fair. The bookmaker price is a different object: slope 1.095, and a 19.2% margin against 0.2%.
MeasuredView research →
How much does the market still move before the off?
4.48pp → 1.43pp
784 races · 60 minutes out to 3 minutes out
Most of the day's information is in the price well before the stalls open.
MeasuredView research →
Does recent form predict winners?
It finds better horses
2,487 races
Form, measured as the top-rated runner on our morning rating, is real information. The market has already absorbed it by the off.
Market already accounted for itView research →
Do steamers win more often?
No effect beyond the price
Market-movement study
Money moving for a horse did not survive being measured against the price it moved to.
No reliable effect foundView research →
Can the draw alone beat the market?
No, on a held-out test
n = 477 held-out races
Draw bias is real in the right conditions. Taken on its own as a way to beat the price, it returned a clean null on races the method had never seen. A small sample, and not a claim that the draw does not matter.
No reliable effect foundView research →
Every figure here is the same governed number the research page carries; nothing is rounded or restated for this table. The exact technical status of each finding is on its own page and in the evidence room.
The belief · “The favourite is safe.”
The favourite is the most likely winner.
It still loses most of its
races.
Across 60,671 races, favourites won 35.0%, and failed to win 65.0%.
The favourite is usually the horse the market considers most likely to win. Most likely does not mean certain: across every race in our historical corpus with a complete Betfair book and a single clear favourite, the favourite lost nearly twice as often as it won.
n = 60,671 · unique BSP favourites; 300 joint-favourite races reported separately
What we found
Across this sample, the favourite failed to win 65% of races. Most likely does not mean certain.
Why it matters
The interesting question is never “is it the most likely winner?”. It is “what does this race ask that horse to survive?”
The caveat
This measures how often favourites win. It says nothing about whether backing them made or lost money. Joint favourites (300 races) are excluded and counted separately.
Measured
The belief · “A short favourite is a good favourite.”
Short-priced favourites win more often.
That doesn't automatically
make them better value.
Shorter favourites win more often, almost exactly as often as the market expected.
Are all favourites equal? No, a 6/4 favourite and a 5/1 favourite are different propositions, and the win rates show it. But at every price band, the actual win rate lands within a percentage point of what the market's own prices implied. The shorter the favourite, the more often it wins. The market knows that too.
n = 60,671 · five BSP bands
What we found
Favourites at 2.00 or shorter won 61.4% of their races; favourites over 6.00, barely 15%. But look at the ivory bars: at every band, the market's own prices predicted almost exactly what happened. The information in “short favourites win more” was already in the price.
Why it matters
A horse winning more often does not automatically make it better value. What matters is whether it wins more often than its price already implied.
The caveat
Bands are price ranges, not value judgements. This chart makes no claim about which band was profitable to back. None of them showed a meaningful gap between actual and expected.
Measured
The belief · “Favourites are solid in any race.”
Where does favourite confidence become fragile?
Favourites are more vulnerable in big fields, and the market prices that in too.
In small fields the favourite wins nearly half the time; with sixteen or more runners, barely a fifth. That is a real, pre-race, structural difference, and at every field size the market's expectation tracks it. Field size alone tells you where favourites fail more often. It does not, on its own, tell you anything the price hasn't already said.
n = 60,671 · pre-defined field-size bands
What we found
Field size changes everything about a favourite's chance: from 43.6% in small fields down to 22.4% with sixteen or more runners. It is a real, pre-race, structural effect, and the market's expectation tracks it band for band.
Why it matters
This is the shape of every finding on this page: the structure genuinely matters, and the price genuinely knows. The question Elite Pass cares about is whether any structural read exists that the price hasn't already absorbed.
The caveat
Field size was chosen in advance as the cleanest pre-race split, not selected after looking for the most dramatic chart.
Measured
The belief · “If the money is coming, somebody knows.”
A shortening price looks informative, until you compare horses at the same final price.
What we tested
Do horses whose price shortened materially before the off outperform their final market expectation? Raw comparison first, then the control that matters: comparing steamers only against other horses at the same final price.
What we found
Raw, steamers look slightly poor value against final expectation. Price-matched, the effect collapses to statistical noise. In the same study, drifters did not underperform their final price either. The move was real information about the horse, and by the off, the market had finished absorbing it.
Why it matters
The market move is information. The harder question is whether it contains information not already reflected in the final price, and in this sample it did not.
The caveat
One study window, drawn from the 2,772 races carrying a full pre-off price ladder. A different capture window could behave differently; the finding has not yet been replicated on an independent period.
Tested · not supported in this sample
The belief · “The best form wins.”
Good form finds better horses.
The market often knows they're
better too.
Good form finds better horses. The market already knows they're better.
Runners our morning rating ranks top win far more often when the market also likes them, a gradient from 41.8% down to 11.0%. Convincing, until you ask what the price alone predicted: 40.7% down to 12.0%. The gradient is the price. In a separate half-million-runner benchmark, no structural predictor we tested beat the market either.
n = 2,487 · races with a unique top-rated runner; gradient measured against BSP
What was measured
“Form” here means the top-ranked runner on our morning ability rating: one defined, pre-race measure, not a blend. Its winners were then compared with what the Betfair starting price alone predicted.
What we found
The rating's gradient is real. Its top pick wins four times as often when the market agrees with it. But price alone predicted almost the identical gradient. And in a separate benchmark of 531,115 runner-race records, no structural predictor we tested beat the market's own probability.
Why it matters
A factor can be highly predictive and still offer no independent advantage once the price is taken into account. That single sentence explains more failed betting systems than any other in racing.
The caveat
This tests one rating family. It does not prove that no form measure anywhere adds information, only that this one, measured properly, did not.
Tested · priced by the market
The belief · “A good draw is an edge.”
The draw matters. “Good draw” on its own may be the wrong question.
What we tested
Whether a draw-based rule, frozen in advance, could beat the market on unseen races, the last win-side question our historical programme had open.
What we found
On the held-out sample (477 races), the draw rule showed no advantage over the market. Badly drawn horses genuinely underperform, but as with everything else on this page, prices already reflect it. Draw as a stand-alone positive signal did not survive.
Why it matters
Elite Pass now treats draw as one input in a bigger question: draw plus expected position plus pace plus the shape of the track. On some courses that interaction is where the real story lives, which is the interaction our current work is trying to express.
The caveat
This closes the question for draw alone, at the sample tested. Course-specific draw-and-pace interactions remain open research.
Not supported as a stand-alone edge
The short version.
“The favourite is safe.”
Most likely is not certain: favourites failed to win 65% of 60,671 races.
Measured
“A shortening horse is value.”
Compared at the same final price, the advantage disappeared in our sample.
Not supported
“The best form wins.”
Strong form is predictive, and the price predicted almost the same thing.
Priced by the market
“A low draw is an edge.”
Draw alone showed no advantage over the market on unseen races.
Not supported alone
“Front runners have an edge.”
A pace profile alone showed nothing; whether the track changes that is open research.
Still researching
Five ideas that looked convincing, until we tested them.
Racing systems often begin with something that is genuinely predictive.
The difficult question is whether that information still adds anything once price, sample size and unseen races are allowed into the test.
These are representative examples from the Elite Pass research programme. The names are editorial and proprietary implementation detail is withheld, but every figure below is reproduced from a recorded Elite Pass record. Where a number could not be reproduced, it is absent rather than estimated.
The Form Stack
FAILED TO CLEAR VALIDATION“Combine the strongest conventional signals: ratings, recent form, pace, course and distance, and where they agree you should find runners the market has underestimated.”
- Why it looked convincing
- Every component carries genuine racing information. When several agree, the horse looks compelling, and the agreement itself feels like corroboration.
- What we tested
- A deterministic settle-and-join of the richer daily sheets against Betfair win starting prices, then a like-for-like comparison against the market's own price.
- What happened
- The fundamentals stayed predictive of racing outcomes. They did not beat the starting price. Adding them to the market model made it slightly worse, and the difference sits inside the margin of error.
- Where it broke
- The mistake was treating good information as though it were information the market had failed to price.
- Races settled
- 317
- Runners
- 2,828
- Combined model
- 1.7772
- Market benchmark
- 1.7261
Lower is better. The gap is −0.051 with a 95% confidence interval of −0.113 to +0.009. It crosses zero, so on this sample the difference is not distinguishable from noise.
Knowing more about the horse did not automatically mean knowing more than the market.
Eight race-days is a small sample, well below the size the programme's own power analysis calls for. This is a failure to clear validation, not a demonstration that form is worthless.
Market priced · Overfitting Out-of-sample chronological validation, pre-registered · 73 held-out races, from 317 settled races / 2,828 runners over 8 race-days · as at 2026-07-14
Follow the Money
EFFECT DID NOT SURVIVE PRICE CONTROL“If a horse is shortening before the off, somebody knows something. Follow sustained market support.”
- Why it looked convincing
- In raw results, horses that shorten do appear to perform better. Looking backwards, the pattern is extremely persuasive.
- What we tested
- Market movers compared against horses that finished at equivalent final prices, rather than against the field as a whole.
- What happened
- Once the comparison was made within matching final-price bands, the advantage did not hold. Horses that overtook the favourite in the market finished behind favourites that held, and structure-aligned reversals produced nothing.
- Where it broke
- A shortening horse usually finishes at a shorter price. That final price already contains most of the information the move represented.
- Market overtakers
- −12.9pp
- Structure-aligned
- 0 / 4
- Late drift and CLV
- No gain
The late-drift result is a known market shape: shorter prices are slightly under-bet and longer prices over-bet, rather than an exploitable signal. The structure-aligned figure rests on about four opportunities inside an eleven-race graded set — enough to show the pattern did not appear, not enough to settle it either way.
The move was real. The independent advantage wasn't.
This tests market movement as a stand-alone selector. It does not claim that market behaviour is uninformative; Elite Pass still records the market at fixed times before the off.
Price confounding · Market priced Market-movement backtest and late-drift lane review · 335 races (2,962 ladder rows) over 9 race-days; overtaker arm 61 flips, structure-aligned arm about 4 opportunities — not powered · as at 2026-07-26
The Draw System
NOT SUPPORTED AS A STAND-ALONE EDGE“Identify the historically favourable draw at a course and distance, and concentrate selections there.”
- Why it looked convincing
- Draw patterns are visually striking. At some courses, historical winners genuinely cluster in particular stalls.
- What we tested
- The threshold was frozen on development data and then reported on races the rule had never seen, with the result revealed only at settlement.
- What happened
- On the held-out races the rule did not beat the market. The lane was closed rather than retuned.
- Where it broke
- Draw is not acting alone. Its effect depends on pace, expected position, field shape, course configuration and distance: the things a stall number cannot carry by itself.
- Held-out races
- 477
- Win rate
- 15.5%
Reported on races the rule had never seen, with the threshold fixed in advance. Separately, draw did carry information as a negative filter, as a reason to avoid a runner rather than to select one.
The draw can matter. A draw number on its own was not enough.
This is one powered held-out test of a stand-alone draw rule. It is not a finding that draw is irrelevant to racing.
Context dependent · Sample illusion Held-out validation, threshold fixed before the test · 477 held-out races · as at 2026-07-07
The Top-Rated Horse
PREDICTIVE, BUT MARKET PRICED“Rank every runner by the strongest pre-race ability measure and back the one at the top.”
- Why it looked convincing
- The ranking genuinely separates stronger runners from weaker ones. The top of the list wins substantially more often than the bottom.
- What we tested
- The top-rated runner was split by what the market thought of it: cases where the price agreed, was neutral, and disagreed, and each group was compared against what price alone already implied.
- What happened
- A steep gradient appeared, from 41.8% down to 11.0%. Price alone predicted almost exactly the same gradient.
- Where it broke
- Predicting winners better than random is not the same as identifying mispriced winners.
- Market agrees
- 41.8%
- vs 40.7% implied
- Market neutral
- 20.9%
- vs 23.6% implied
- Market disagrees
- 11.0%
- vs 12.0% implied
Actual against what the price alone already implied, on the same races.
Good form found better horses. The market usually charged more for them.
These three groups come from one adjudication of 2,487 races. No larger corpus is quoted alongside them: figures from other Elite Pass studies measure different populations and cannot be joined to this gradient.
Market priced recorded market-conditional adjudication · 2,487 races · as at 2026-08-19
The Favourite System
HIGH STRIKE RATE, NO INDEPENDENT ADVANTAGE ESTABLISHED“Concentrate on the market's strongest opinion. The shorter the favourite, the safer the selection.”
- Why it looked convincing
- Short favourites genuinely do win far more often. The strike rate feels like control.
- What we tested
- Every race in the corpus with a single strictly-lowest starting price, scored against what that price implied, overall and inside each price band.
- What happened
- Favourites won 35.0% of the time. The market's own prices implied 21,285 wins. They produced 21,215.
- Where it broke
- A high strike rate can feel like an edge while simply reflecting an expensive price.
- Races
- 60,671
- Favourite wins
- 35.0%
- Expected wins
- 21,285
- Actual wins
- 21,215
| Starting price | Won | Price implied |
|---|---|---|
| Up to 2.00 | 61.4% | 61.5% |
| 2.01 – 3.00 | 40.7% | 41.1% |
| 3.01 – 4.00 | 29.8% | 30.0% |
| 4.01 – 6.00 | 23.0% | 22.4% |
| Over 6.00 | 15.0% | 15.9% |
Actual against expected in every price band. The favourite wins far more often at 2.00 than at 6.00, and almost exactly as often as each price implied.
A horse winning more often does not automatically make it better value. What matters is whether it wins more often than its price already implied.
A convenience corpus weighted toward all-weather and major courses. Descriptive only; no claim of value or edge is made from it.
Market priced Favourite research report, historical corpus recomputed 2026-08-21 · 60,671 unique-favourite races · as at 2026-08-21
The interesting part isn’t that systems fail. It’s why.
Some racing ideas are wrong. Others are genuinely predictive but already priced. Others work only under particular conditions. And some disappear the moment they meet data they weren’t built on.
Elite Pass keeps those distinctions, because they are not the same failure, and they do not lead to the same next question.
- Market priced
- The signal predicts racing outcomes, but the price already reflects it.
- Sample illusion
- The apparent effect does not survive a larger or unseen sample.
- Context dependent
- The effect exists only under particular course or race conditions.
- Price confounding
- A result looks powerful until horses at equivalent prices are compared.
- Overfitting
- The system explains the races it was built from, but not unseen races.
- Data dependence
- The effect needs information that cannot be reliably obtained before the race.
The failures narrowed the question.
Instead of looking for another isolated variable, the research increasingly turned toward the interaction between the race, the horse and the market.
Have a system of your own? Elite Pass will try to break it before you trust it: independent system audit and consultancy.
The open questions.
Why favourites fail
Can pre-race structure identify when the favourite faces more resistance than its price suggests?
ResearchingPace × position
When several runners want the same part of the race, what changes?
ResearchingCourse geometry
When does running style become more important because of the track?
ResearchingEP Rank
Can these competing views be compressed into a stable, validated within-race hierarchy?
Research previewThree terms, quickly.
BSP
Betfair Starting Price: the exchange price calculated at the off. It is the price this research measures everything against, because it is the last and best-informed price a race produces.
Implied probability
The chance represented by a price. Decimal odds of 2.00 imply a 50% chance, before the book's margin is accounted for. This research normalises across the whole field so the implied chances in a race sum to 100%.
Percentage point
The direct difference between two probabilities: a move from 30% to 34% is four percentage points.
The questions, one at a time
- Why do favourites lose? Across 60,671 races the favourite lost roughly two times in three, which is about what the price implied. That makes the useful question a different one.
- How often does the favourite actually win? About 35% of the time, and how that moves with price and field size.
- Does the draw matter? Sometimes enormously. As a standalone way of beating the market it returned a clean null on held-out races, so we killed the idea and kept the record.
- Do steamers win more often? They do. The advantage disappears once you compare them with drifters at the same final price.
- Do horse racing ratings work? We tested our own across 30,088 races. It ranked horses well and told us nothing the price had not already worked out.
- What is Betfair Starting Price? The benchmark every number on this site is measured against, and the paired test where the exchange price came out indistinguishable from perfect and the bookmakers' did not.
- Do horse racing systems actually work? What happens to a racing system when it is tested against the price that was actually available, rather than against chance.
- Five racing beliefs, put under pressure Five things racing people repeat, tested against the market that had already priced them. Three did not survive.
The ideas behind the questions
Guides to the concepts, not validated Elite Pass findings.
- What is race shape? The problem the whole field creates: who wants the lead, who needs cover, who gets forced wide, and what that costs. The idea this whole thing is built on.
- What is pace, and what is pace burn? Why two horses that both need the lead can ruin each other, and why an uncontested lead changes a race that looked settled on paper.
The market tells us which horse is expected to win. Elite Pass is interested in the race that expectation has to survive.