Elite Pass research

The rating was right. The price already knew.

We tested our own rating against the market across 30,088 races. It ranked horses well. It did not tell us anything the price had not already worked out.

Start with the instinct, because it is a good one

If one horse in a race is clearly top rated, surely that tells you something.

It does. That part is not in dispute and never was. A rating summarises a great deal of information into one number, and a good one orders a field sensibly. Better-rated horses do win more often.

The harder question, and the one this page is about, is whether it tells you anything the price does not already know.

Useful and mispriced are different things

Imagine a rating correctly identifies a horse with a real four-in-ten chance of winning. That sounds like a powerful tool, and in one sense it is.

But if the market has also priced that horse at about four in ten, the rating has identified the horse correctly and identified no value at all. It was right and it was late. Everything it knew was already in the price you are being asked to take.

That example is illustrative, not a measurement.

This is the distinction that decides whether an idea is worth anything: not does it predict winners, but does it predict them better than the price does. Almost every racing angle that looks exciting passes the first test. The second is where they go to die.

So we tested our own

Not a competitor's marketing claim. The rating Elite Pass builds, which is the one we would have had to rely on.

The study covers 30,088 races and 298,444 runner records: every race with a single clear top-rated horse, a complete exchange book, at least four runners and a sane market. Exchange starting prices are only covered in that archive from August 2020, so the study runs from there, and it is split by time, trained on the earlier years and tested on the later ones so that the test is a forecast rather than a memory.

We asked two questions. Does being our top-rated horse carry information beyond the price? And does it matter how far clear that horse is — a wide gap to the second-best being the case where a rating ought to be shouting loudest?

What came back

On the first question: our top-rated horses won at very close to the rate the market implied they would. Measured against a de-vigged exchange price — the market's own probability with the bookmaker-style margin stripped out — the ratio of actual winners to expected winners came in at 0.998, where 1.000 is exactly par.

On the second: separation did not rescue it. A horse being a long way clear of the second-best on our own numbers did not produce reliable extra information once the price was accounted for.

Both effects pointed very slightly in the helpful direction and neither was strong enough to be distinguished from nothing. That is the honest summary: no reliable information beyond the price.

The ratio above is descriptive. It is not a registered endpoint and does not support a claim of edge in either direction.

The rating was right. The price already knew. A concept diagram, not a measurement. Two overlapping fields: what our rating knew, and what the Betfair Starting Price already reflected. The sliver of the rating lying outside the price is where any incremental information would show up. After controlling for price, no reliable incremental effect was detected there in the studied population. The shapes carry no measured quantity and no exact containment is claimed. ELITE PASS · RESEARCH The rating was right. The price already knew. WHAT THE PRICE REFLECTED OUR RATING After controlling for the price, no reliable signal was detected out here. A rating can identify the strongest horse without identifying anything the market has missed. 30,088 RACES 298,444 RUNNERS 0.998 ACTUAL TO EXPECTED · 1.000 IS PARITY PREDICTIVE ≠ INCREMENTAL elitepassracing.com
A concept diagram, not a measurement: the shapes carry no quantity and are not drawn to scale. The overlap is not a claim that the price contained everything the rating knew. What was tested is narrower — our rating's rank and its separation from the second-rated horse, against a de-vigged Betfair Starting Price baseline — and after that control no reliable incremental effect was detected in this population. The actual-to-expected ratio is descriptive and is not a registered endpoint.
Being straight about it

What this does not say

It does not say ratings do not work. One rating was tested: ours.

It says nothing about Timeform, Racing Post Ratings, Topspeed, official ratings or speed figures. None of them was in this study, and naming them here would be inventing a result we do not have.

It also does not say top-rated horses lose more than they should, and it does not say they win more than they should. At 0.998 they finished in front almost exactly as often as the price said they would. That is parity, and parity is the absence of an effect in either direction rather than evidence for one. The market simply had it right.

Elite Pass has no validated betting edge and does not claim one.

What it changed

This is one of the reasons Elite Pass stopped trying to build a better winner-picker. If the strongest number we could produce was already inside the price, then producing a slightly stronger number was not the work.

Knowing which horse looks best is only part of a race. The rest is what the race is going to ask of it: how it will be run, what role the horse is likely to get, how much has to go right. A rating describes the horse. It does not describe the problem the horse has to solve.

So the question we ask each morning changed. Not which horse has the best number, but what is happening around this race that the number and the price may not fully describe.

Elite Pass Gold · early access

The part a rating cannot see

Gold is the daily read for races where the evidence converges, and an honest No Read everywhere else.

Early access

Join Gold early access

Be first in when Gold Race Intelligence opens. No daily spam, no tipster noise.