“Nine of the last ten winners fitted this trend.” It is a convincing sentence. It sounds like a shortcut through a difficult race, which is precisely why it deserves another question.
How many of the beaten horses fitted it as well? Without that number, you know something about the winners. You do not yet know whether the trend helped to identify them.
9 WINNERS. NOW COUNT THE RUNNERS.
Both groups won at 5%. Capturing 90% of winners meant capturing 90% of runners too.
Ten races, 200 starters and ten winners. Illustrative numbers; not Cheltenham results.
The missing number changes everything
Imagine ten races with twenty runners in each: 200 starters altogether. Nine of the ten winners had a particular characteristic. So did 171 of the horses that lost.
That gives the trend group nine winners from 180 runners, a 5% win strike rate. The group without the characteristic had one winner from twenty runners: also 5%. The headline sounds impressive because it contains most of the winners. It also contains most of the runners.
These are invented numbers for a worked example, not historical Cheltenham results. They demonstrate why a winners-only statistic cannot tell you how selective or useful a filter was.
Even a higher strike rate is only part of it
Now change the example. Suppose the same nine winners came from just forty qualifiers. Their 22.5% strike rate would look much more interesting. You would still need to know the prices.
A group consisting largely of short-priced favourites should win more often than a group of outsiders. That does not automatically make the favourites profitable. The odds determine how often they need to win to cover the losers.
At decimal odds of 2.00, a win-only selection needs to win more than half the time to produce a positive expected return before any applicable costs. At 5.00, the corresponding break-even rate is 20%. A strike rate is meaningful only when you know the price context.
Beware the filter that becomes too neat
Age, official rating, days since last run, previous Festival form, country of training, number of starts: there are plenty of ways to slice a race’s history. Try enough combinations and something will eventually look remarkable.
The warning sign is a rule that keeps acquiring extra conditions until it fits the past perfectly. Why must a horse be seven rather than eight? Why does the cut-off sit at exactly that rating? Why exclude one particular year? There may be good answers, but they should exist independently of making the winning column look tidy.
A useful explanation has a racing reason behind it. An experience measure might relate to the pressure of a novice chase. A stamina measure might relate to the distance. That still needs testing, but it is a better starting point than an unexplained collection of numbers.
Conditions can change while the race name stays familiar
Historical comparisons need consistent race definitions. If the distance, eligibility or handicap status changes, older renewals may describe a different contest.
The Festival’s recent novice-chase changes are a good example of why this matters. The former Turners Novices’ Chase was replaced by the Jack Richards novice handicap in 2025; the National Hunt Chase also changed to a novice handicap. Treating the whole history as one unchanged population would blur a significant distinction.
Keep a visible break in any chart where the conditions change. Explain what is included and why. Sometimes the honest answer is that there are too few renewals under the current format to support a strong trend.
Sources: 1
Give the rule a test it has not already seen
If you devise a filter using ten years of results, those same ten years cannot provide an independent test of it. You have already used their answers to shape the rule.
Set the rule in advance, record it and examine a later period without changing the conditions after every defeat. Better still, compare several sensible time windows and show how sensitive the results are to one big-priced winner.
This will not remove uncertainty. It does stop a back-fitted idea being presented as if it had predicted the past. At Puntrr, that distinction belongs beside the numbers, where people can actually see it.
Five questions before using a Cheltenham trend
First, how many runners qualified, including the losers? Second, what prices were available and what were the returns? Third, were the race conditions comparable? Fourth, was the rule chosen before or after looking at the results? Fifth, does it still make racing sense for today’s runners?
A trend that survives those questions can help organise your research. It might highlight a concern or suggest a horse deserves closer attention. It should not save you from reading the form.
The aim is not to ban statistics. It is to insist that they earn the confidence we place in them. A modest, fully explained finding is more useful than a spectacular headline with half the record missing.
Your questions, answered
Are Cheltenham trends useful?
They can help you organise evidence, provided the analysis includes all qualifying runners, prices, the period studied and any changes to the race conditions.
Why can “nine of the last ten winners” be misleading?
If nearly all runners matched the trend, capturing nearly all winners may offer little discrimination. You need the number of qualifiers and their results, not just the winning cases.
What is a back-fitted racing trend?
It is a rule shaped using outcomes already known. It may describe those results well without working on later races. An independent later test helps expose that problem.
Sources & editorial notes
Reviewed 11 November 2026. Interpretation is Puntrr’s; factual results and conditions are linked below.