Racing intelligence guide
How horse racing predictions work.
A useful racing model does not hunt for one magic statistic. It combines many imperfect signals, estimates each runner’s chance and stays honest about uncertainty.
Last updated 6 September 2026The factors a serious model should consider
Horse: recent performances, speed and finishing effort, consistency, class movement, distance and surface record, days since last run, recent workload, age, weight and improvement or decline.
Race: field strength, race type, pace shape, likely running position, distance, surface, going, course configuration, draw and the interaction between these conditions.
Connections: trainer and jockey form, their record together, course experience and performance with comparable runners—not reputation alone.
Market: prices at consistent pre-race times can add valuable collective information. They must be time-stamped and kept separate from results so a model never learns from information that was unavailable when the prediction was made.
Why context beats isolated statistics
A strong recent finish means something different after an easy lead than after racing wide in a fast-run contest. A favourable draw can matter greatly at one course and distance and barely at all at another. Good modelling captures these interactions instead of treating every race as identical.
Different race types also carry different levels of uncertainty. Lightly raced novices may have less dependable history than exposed handicappers, while small samples can make trainer, jockey or course percentages look more certain than they are.
How to judge a racing predictor honestly
- Require predictions to be recorded before the advertised cutoff—not recreated after the result.
- Count every eligible race and keep losing days visible.
- Use a large forward-looking sample; one day is entertainment, not evidence.
- Compare the top selection with simple baselines such as the favourite and public predictors.
- Check probability calibration: runners shown near 30% should win roughly three times in ten over a large comparable sample.
- Review performance by race type, course, going and confidence instead of trusting one headline percentage.
What FormSignal shows
FormSignal provides one pre-race top selection for each listed UK and Irish race in its feed, a projected finishing order, relative model strength and recorded results. It is a prediction and analysis product: probabilities are not guarantees.
Apply for the founding beta or see how the Android app presents a race.
Frequently asked questions
What information matters most in a horse racing prediction?
No single statistic is decisive. A useful model combines recent form and speed, race class, distance and surface suitability, going, course and draw effects, weight, pace shape, trainer and jockey patterns, field strength and time-specific market information.
Does the favourite always have the best chance?
The market favourite usually has the highest implied chance, but favourites still lose regularly. A prediction should express relative probability and uncertainty rather than promise a winner.
How should prediction accuracy be measured?
Judge predictions recorded before the race over a meaningful sample. Check top-selection win rate, probability calibration, performance by race type and whether all wins and losses remain in the record. Avoid judging a model from one unusually good day.