RESEARCH · SIMULATED · NOT FOR SALE
The early-price finding.
At 07:00 on race day the model picks up to three horses whose Betfair price looks too long, and each bet runs to the result, win or lose, with no trading out.
HOLDOUT · FEB 2025 TO JAN 2026 · LEVEL STAKES · AFTER 5% COMMISSION
Walk-forward 2017–2024: 8,367 selections, 16.1% winners, +43.6% at 07:00, +45.5% at 08:00, -4.2% at SP. Longest losing run 42; deepest drawdown 77 points.
THE LEDGER IN POUNDS · SIMULATED
Bars: profit or loss each period after commission and costs. Line: running total. Shaded: the holdout (Feb 2025 to Jan 2026), run once. Each year is tested on a model built only from the years before it. Data ends January 2026.
Month-by-month ledger
§1 What it shows
The early market misprices some horses, and by the off it has corrected. Backed at 07:00 or 08:00 the selections return well; backed at Betfair SP the same horses roughly break even. The value is in the timing, not the horse.
§2 By start time
| Picked at | Walk-forward winners | WF return at that price | WF at SP | Holdout winners | Holdout return | Holdout at SP |
|---|---|---|---|---|---|---|
| 21:00 day before | 13.1% | +28.0% | -12.0% | 13.9% | +40.2% | -12.2% |
| 07:00 | 16.1% | +43.6% | -4.2% | 15.9% | +41.8% | +7.2% |
| 09:00 | 12.7% | +45.4% | -2.8% | 12.1% | +37.3% | -3.1% |
| 10:00 | 11.1% | +35.7% | -6.3% | 11.5% | +34.8% | -4.2% |
§3 By price (07:00 model, all years)
| Price at 07:00 | Selections | Winners | Return at 07:00 | At 08:00 | At SP |
|---|---|---|---|---|---|
| evens–4/1 | 906 | 41.4% | +42.5% | +42.3% | -2.3% |
| 4/1–7/1 | 1,754 | 25.5% | +54.2% | +48.7% | +3.4% |
| 7/1–11/1 | 1,993 | 15.5% | +41.4% | +44.7% | -4.0% |
| 11/1–19/1 | 2,651 | 10.1% | +46.0% | +43.8% | -0.6% |
| 19/1–29/1 | 2,017 | 5.8% | +31.0% | +40.8% | -10.2% |
§4 By year (walk-forward)
| Year | Selections | Winners | Return at 07:00 | At 08:00 | At SP |
|---|---|---|---|---|---|
| 2017 | 1,078 | 16.3% | +56.4% | +58.3% | +9.3% |
| 2018 | 1,011 | 16.1% | +40.9% | +48.3% | +1.5% |
| 2019 | 1,072 | 16.0% | +43.2% | +49.2% | -5.5% |
| 2020 | 876 | 16.9% | +55.7% | +59.1% | +5.5% |
| 2021 | 1,082 | 17.1% | +50.0% | +55.6% | -10.4% |
| 2022 | 1,079 | 14.7% | +23.7% | +32.9% | -21.0% |
| 2023 | 1,086 | 16.0% | +42.1% | +22.3% | -2.8% |
| 2024 | 1,083 | 16.1% | +39.2% | +38.6% | -8.4% |
§5 Why it isn't for sale
- The money isn't there yet. These horses trade a median £745 on Betfair before 11:00. On the one night we recorded the full order book, a typical runner offered about £3 at the best price at 07:00. Early markets are thin; quoted prices may not take a real stake.
- Followers would move the price. If many people back the same horse at 07:00, the early price disappears. A tip nobody can get at the quoted price is worth nothing.
- Simulated. Prices are the last trade on Betfair at each time, not a guaranteed fill.
§6 Next
- Record the full Betfair order book every night, to measure how much could really be staked.
- Compare with bookmakers' early prices.
- Then a live, timestamped test with real stakes, logged on the results page.
Past results, simulated or live, do not predict future results. Statistical analysis, not advice. 18+.