By Dr David Marlin & Prof Jane Williams
The “Maths” – Prof Jane Williams (Hartpury University) and Prof Tim Parkin (University of Glasgow) published a paper where we used a statistical approach to analyse factors associated with success in the English Epsom Derby. Our model predicted that the chances of winning were highest for horses with more 2yo wins (in any races, not just recognised Derby trial races), that were foaled in Ireland, that were ridden by the same jockey in all races and of course, not surprisingly, being high in the betting. Overall, this gave us a 78% chance of picking the winner.
The 2010 Epsom Derby is Nearly Won by an Outsider
The first time we applied this model was in the 2010 Epsom Derby. Four horses stood out: ‘Jan Vermeer’, the race favourite, ‘At First Sight’, a 100-1 outsider universally dismissed as only being a pacemaker, ‘Rewilding’ and ‘Midas Touch’. The winner of the 2010 Epson Derby was ‘Workforce’ (starting price 6-1), a British Thoroughbred racehorse ridden by Ryan Moore in a course record time. In 2009, he had won one race, and in 2010, he also won the Prix de l’Arc de Triomphe, both times ridden by the same jockey. ‘At First Sight’ led the race from the start and was only eventually overtaken just over 1 furlong from the finish to be beaten into 2nd place by just 7 lengths. ‘Rewilding’ was 3rd (starting price 9-2), the favourite ‘Jan Vermeer’ was 4th and ‘Midas Touch’ was 5th. So whilst we didn’t predict the winner, we did predict an “outsider” as the 2nd placed horse at 100-1 – well worth an each-way bet!
What Does Our Model Predict for The 2026 Epsom Derby?
Prof Jane Williams and I haven’t run our model for a few years, but decided to have a go for this year’s Derby. The critical tension in this analysis is between the model and the market on the item. The betting public rates Item as a ~20% chance; the model rates it at ~3-7%. If the model is well-calibrated for 2026, Benvenuto Cellini (IRE) is an outstanding value at 7/4, and Item is a poor value at 4/1. If racing has evolved since the model’s training period – if, for instance, the Dante Stakes has become a more reliable Derby predictor than it was in 1988-2009, or if Irish-bred advantage has diminished – then Item could outperform the model’s projection.
The strongest non-favourite selections by the model are Ancient Egypt (IRE) and Christmas Day (IRE), both Irish-bred with 2 two-year-old wins, at 16/1 and 33/1 respectively. These represent excellent each-way value if the model’s Irish-bred coefficient still holds: at 16/1, Ancient Egypt (IRE)’s model-implied probability (8-19%) exceeds its market-implied probability (~6%), suggesting potential each-way value.
Ultimately, the epidemiological approach reduces a complex handicapping problem to four powerful, empirically validated variables. Benvenuto Cellini (IRE) satisfies all four. The model says: to win back the favourite, but for epidemiological, not sentimental, reasons. For an each-way bet back Ancient Egypt (IRE) and or Christmas Day (IRE).
