In one sentence
A speed figure converts how fast a horse ran into a points rating that can be compared across different courses, distances and ground conditions.
How it works
A raw time means little on its own: six furlongs at Epsom on soft ground is not the same test as six furlongs at Newmarket on firm. Speed figures fix this by comparing each time with a standard time for that course and distance.
Next comes the going allowance, an estimate of how much the ground slowed or quickened every race on the card that day. It is usually worked out by comparing all the day's winning times with their standards. Subtract it and you have a time as if run on standard ground.
Finally, the time difference is turned into lengths, and lengths into pounds of weight, giving a figure in the same units as handicap marks. Beaten horses get the winner's figure minus their beaten distance in pounds.
The maths
- T: the winner's actual time in seconds.
- G: the going allowance in seconds for this distance on the day (positive on slow ground).
- S: the standard time for this course and distance.
- L: lengths per second (commonly taken as about 5 in sprints).
- W: pounds per length (larger at shorter trips; about 3 at six furlongs).
- B: the base figure given to a standard-time performance (100 here).
- d: beaten distance in lengths.
In words: work out how far the ground-adjusted time was from standard, convert it to lengths then to pounds, and subtract from the base.
Worked betting example
A six-furlong handicap. Standard time 71.80s. The winner clocks 72.60s. The going allowance is +0.50s (soft).
- Ground-adjusted time: 72.60 − 0.50 = 72.10s.
- Slower than standard by 72.10 − 71.80 = 0.30s.
- In lengths: 0.30 × 5 = 1.5 lengths. In pounds: 1.5 × 3 = 4.5 lb.
- Winner's figure: 100 − 4.5 = 95.5.
- A horse beaten 2.5 lengths: 95.5 − 2.5 × 3 = 88.0.
- Next time out, suppose the winner is the top-rated on figures after weight adjustments, and your own records suggest horses in that position in similar races win about 24% of the time. Betfair offers 5.0.
- £10 stake, 2% commission: a win pays £40 × 0.98 = £39.20. EV ≈ 0.24 × £39.20 − 0.76 × £10 ≈ +£1.81.
The 24% has to come from your own tested data; guessing it defeats the purpose.
Where it's good
- Flat sprints and all-weather racing, where pace is steady and times are reliable.
- Comparing horses that have run at different tracks or in different classes.
- Spotting improvers whose figures are rising faster than their handicap mark.
- As a strong input to a conditional logit or other win-probability model.
Limitations and pitfalls
- The going allowance is estimated from few races and is often the largest source of error.
- Tactical, slowly run races produce misleadingly low figures regardless of ability; pace adjustments help.
- Standard times need many races to set; new or rarely used course configurations are unreliable.
- Hand-timed or inaccurately measured races (rail movements change distances) corrupt figures.
- The lengths-per-second and pounds-per-length factors vary with trip and are approximations.
- Published figures are widely followed, so top-rated horses are often already short in the market.
- Jumps racing, with falls, obstacles and long trips, suits speed figures less well.
How to build it
- Python: pandas for the race data, statsmodels to estimate standard times and going allowances by regression.
- Data: sectional or finishing times, course and distance, going, beaten distances and weights carried.
- Estimate going allowance per meeting by regressing each race's time deviation on class and distance, not by eyeballing.
- Tip: store every figure with its race conditions, so later models can learn which figures were trustworthy.
Related methods
- Linear regression: used to set standard times and going allowances.
- Conditional logit: turns figures and other inputs into win probabilities.
- Mixed and hierarchical models: estimate course and going effects more robustly.
- Plackett-Luce and Harville: extend win chances to forecast and place markets.