← Blog · 📝 Article · 5 November 2026

Filter Betting Signals: How to Separate Strong Lay Signals From Noise

Filter Betting Signals: How to Separate Strong Lay Signals From Noise

To filter betting signals, set clear rules for race type, odds band and market behaviour. Then test those rules on past data you did not tune them on, and judge the results on a long run of published selections, not a few good days. A signal that survives all three steps is worth acting on. One that does not is a hunch with a label on it.

This guide covers how filtering works in UK horse racing, which factors are worth filtering on, how to test a filter honestly, and how to read strike rate, losing runs and sample size. It is written for over 18s who bet on racing, with a lay-betting angle, because that is what we do at DonkeyRadar.

Donkeyradar
Find Data-Led Lay Signals
DonkeyRadar analyses historical strike rates and live market prices to identify lay signals before races across the UK, Australia and the US.
Visit DonkeyRadar

TL;DR

Table of Contents

What a betting signal is, and what filtering does to it

A signal is a flag that a horse meets a pattern. It is not a prediction and not a promise. The Racing Post’s Signposts are a familiar example: small pattern flags beside a horse’s name, such as trainer in form, course specialist or first run after a wind operation. They are binary lookups that do not weigh how strong the pattern is against price or the rest of the field.

That is the gap filtering fills. One flag on its own says little. The same article points out that the trainer-in-form flag is the most common and least informative, because almost any active mid-sized yard triggers it during a hot patch. A filter turns a pile of loose flags into a rule: only act when these conditions, and no others, are true.

Filtering for UK racing means defining specific rules for race conditions such as class, going, distance, field size and track type, so you only act where the system is meant to work. Typical system filters also include trainer form over a recent window, age group and price bands, which remove races that do not fit the logic.

For a lay bettor the goal is the mirror image of a back bettor’s. You are not hunting a winner. You are looking for the runner the market has priced too short relative to its real chance of losing the race to others.

Which filters are worth using

The temptation is to filter on everything. Resist it. British Racecourses lists the inputs commonly used in racing models: recent form, speed ratings, official rating, class, distance, going, course, draw, pace, sectional times, weight, age, trainer form, jockey performance, days since last run, field size and market price. The same piece warns that adding more variables can make a model worse.

The academic side agrees. William Benter, in his well-known paper on computer-based handicapping, describes improving factors as educated guessing and trial and error. He also warns about ending up with a hodgepodge of highly correlated variables. Two filters that measure the same thing are one filter with extra noise.

Race-level filters

These decide whether you look at a race at all:

Runner-level filters

These decide which horse in the race you consider:

Market filters

The betting market holds a great deal of information. Mark Dooley of British Racecourses, writing in August 2026, says ignoring it completely can be a mistake. But if your aim is to find market errors, he adds, avoid building a model that just copies the market. One approach is to build your own fundamental price first, then compare it with the market.

Market movement is a filter in its own right. Betfair marks runners in UK and Irish win markets with Beacons when the price has moved by a set amount. A flame marks a heavily backed shortener and a snowflake marks a drifter, triggered at a 7.5% probability difference against an anchor price. For a lay bettor, that is useful context: a runner shortening hard is being backed by someone, and a drifter is being abandoned. Neither is a verdict, but both belong in a filter that checks the live price before you act.

The market also tells you where to look. Inform Racing’s long-run figure, as summarised by one racing-form guide, is that 75% to 80% of UK winners come from the top five in the betting. That is a reason a lay filter has to be specific about price: the further down the market you go, the less a horse’s price carries.

Market filters

How to test a filter without fooling yourself

Most bad signals are not frauds. They are filters that were tuned until they looked good on old results. Three habits prevent that.

1. Split your data

To test a filter without overfitting, divide the dataset into separate training and test segments. Build and tune the rules on the training set, then apply them unchanged to the unseen test set. If the filter only works on the data it was built from, it does not work.

2. Use only what you could have known

The same guide warns against look-ahead bias: the backtest must use only data available at the moment the decision was made. In practice, avoid closing prices, final fields after non-runners are removed, and ratings updated after the race. A lay signal that looked brilliant at the starting price may never have been available at that price an hour before the off.

3. Start from a reason, not a trawl

Betting systems are more robust when built around a recognised inefficiency, such as the favourite-longshot bias or draw biases at particular tracks, rather than trawling a database for patterns. The same workflow guide says to compare the available odds with a fair probability estimate and make sure your opinion is defensible before you bet.

That is the logic behind our own work. If you want the longer version, our guides to horse racing data models and horse racing data analysis walk through how past strike rates and live prices combine.

Reading strike rate, odds and sample size

A high strike rate is the number everyone quotes and the one most easily misread. For a lay bet the strike rate means how often the selected horse loses, so a high figure is expected when you choose horses with a low chance of winning. What matters is what you risk to get it.

Laying costs you liability, not stake. At a lay price of 4.0, a £10 lay wins you £10 if the horse loses and costs you £30 if it wins. A long run of small wins can be wiped out by one losing selection if liability is not capped. That is why strike rate and profit are different questions, and why we publish both.

To judge any signal service, look at four things together:

What to check Why it matters
Strike rate Shows how often the signal came right, but only in context of price
Odds band Determines how much one loss costs against a typical win
Profit or loss in units The result after liability, not just win count
Sample size Tells you whether the figures mean anything yet

Why a few results prove little

Losing runs are part of every method. Writing in August 2026, John Burke of Daily Punt noted that at a strike rate of about 23%, a 10-race losing run in past results does not mean 10 is the worst you will see: over 100 bets a run near 17 is possible, and over 500 bets a run of 20 or more is a real possibility. His conclusion is the practical one: your staking has to survive the losing runs.

Possible losing runs across different bet samples

That example is for a back-betting method with a low strike rate. The principle carries over to lay betting with the picture reversed: a high strike rate can hide a rare, expensive loss. Know the largest single liability in your staking plan before you place the first lay, not after.

How to check a signal’s record is complete

This is the question that separates a useful signal service from a marketing page. A strike rate means nothing if only the winners were kept. Before you trust any signal, check these:

  1. Published before the race. The signal should exist, with a timestamp, before the off. A selection posted afterwards is a result, not a signal.
  2. Every selection recorded. Losers should sit in the history beside winners, with the price and the date.
  3. A price you could have taken. Results should use prices from when the signal went out, not the starting price.
  4. A permanent, public history. You should be able to see the whole record, not a hand-picked month.
  5. A sample large enough to mean something. Hundreds of selections, not dozens.

This is how we run DonkeyRadar. All signals are published before races, every result is recorded, and the history is public and permanent. Losses sit there with the wins. It happens, and transparency is the whole point. Our own results show a strike rate of over 85%, and we would rather you check that against the record than take our word for it.

Our view on filtering lay signals

We build DonkeyRadar on one belief: a signal is only as good as the rules behind it and the record in front of it.

Our algorithm processes historical strike rates and live market prices to identify the runner most likely to lose in a race, across UK, Australian and US meetings. That is a deliberately narrow job. We do not try to rate every horse in every race. We filter down to the situations where the statistics are most useful, then publish the signal before the race.

We grade signals into staking tiers and link straight to the Betfair Exchange from the dashboard, so the liability decision is made before you click, not after. Our approach is data-driven and based on tested statistical parameters, not opinion.

The filter is the product. The published record is the proof. We keep both in the open.

Donkey

Responsible use: risk before reward

Laying is not the same as backing. The International Federation of Horseracing Authorities notes that lay betting lets bettors profit from negative information, and that lay-to-lose on an exchange is higher risk than back-to-win and should be monitored when turnover is high. The practical reading for you is simple: the downside on a single lay can be many times your stake.

Safe use comes down to a few habits:

Our guide to capping liability in the 2.0 to 6.0 odds band sets out one way to put numbers on this, and our lay the field guide covers liability caps in more detail.

Building a simple filter of your own

If you want to experiment before using any service, start small. Write the rules down before looking at results.

  1. Pick one race type. For example, flat handicaps only.
  2. Set a price band. Choose a band where your maximum liability is acceptable.
  3. Add one runner-level rule. One form or trainer condition, not five.
  4. Add one market check. For instance, skip runners the market is backing hard in the last stretch before the off.
  5. Test on a training set, then a test set. Do not change the rules after seeing the test result.
  6. Track forward. Record each signal with a timestamp and the price you could actually have taken.
  7. Judge after hundreds, not tens.

This takes patience, and it is a good way to understand what a signal service is doing on your behalf. If you would rather start from tested rules, the next section explains where we come in. To place the lays themselves, our step-by-step guide to laying a horse on Betfair covers the mechanics, and our UK lay tips guide and odds-on lay strategy cover the wider approach.

Try filtered lay signals with DonkeyRadar

DonkeyRadar applies this kind of filtering for you. The free tier gives you the daily lay signals. Pro is £29 per month and adds real-time alerts by email and Telegram, the full results history and API access for trading software, after a seven-day trial. Both link directly to the Betfair Exchange from the dashboard, and every signal is published before the race.

Donkeyradar

See how the signals and the public record work on the DonkeyRadar site.

FAQ

What does it mean to filter betting signals?

It means applying written rules so you only act on signals that fit conditions your method is designed for. In horse racing those rules usually cover race type, class, going, distance, field size, price band and market movement. The aim is fewer, better-defined bets rather than every flag on the card.

How many filters should I use?

As few as you can justify. Research on racing models warns that adding more variables can make a model worse, and correlated variables add noise rather than information. Start with one race-level, one runner-level and one market rule, then test before adding anything.

Is a high strike rate enough to trust a signal?

No. Strike rate shows how often a signal was right, but not what each result cost or paid. For lay bets, one losing selection at a short price can cost many times a typical win, so check odds band, profit in units and sample size together. A full, timestamped record matters as much as the percentage.

How can I tell if a signal service has shown me every result?

Look for signals published before the race, losers recorded beside winners, prices you could have taken at the time, and a permanent public history. At DonkeyRadar all signals are published before races and results are tracked and kept available. Pro includes the full results history.

Are profits from betting in the UK tax-free?

Betting winnings for individual bettors in the UK are not currently taxed, which is the position we reflect on our site. Tax rules can change, so check the latest guidance from HMRC if your circumstances are unusual.

Sources