The same amount every time, on our numbers. And the interesting part is that we tested the alternative properly and it came out worse.
The idea behind variable staking is intuitive: put more on when the bet looks good, less when it looks marginal, and you should end up ahead of someone betting flat. Our model produces an estimated edge on every runner, so we could test exactly that.
ROI by how much the model “likes” the horse
Flat-stake ROI for top picks in each model-edge band. If confidence predicted profit the bars would climb as edge grows, but instead they don't, and every band stays negative.
| Model edge | Bets | Strike | Profit / loss | ROI |
|---|---|---|---|---|
| below 0 | 5,900 | 33.6% | -£6,864 | -11.6% |
| 0–3% | 1,341 | 17.2% | -£1,950 | -14.5% |
| 3–6% | 1,268 | 14.2% | -£2,161 | -17.0% |
| 6–10% | 1,446 | 12.5% | -£2,812 | -19.4% |
| 10%+ | 1,353 | 14.2% | -£1,532 | -11.3% |
Flat stakes vs sizing by confidence (±50%)
The fair comparison is ROI (return per £ staked); absolute profit/loss just tracks how much was staked in total.
| Strategy | Staked | Profit / loss | ROI | Worst drawdown |
|---|---|---|---|---|
| Flat £10 | £113,080 | -£15,318 | -13.5% | -£16,088 |
| Sized by edge (±50%) | £108,086 | -£15,275 | -14.1% | -£16,188 |
Sizing by confidence staked £4,994 less overall and lost £43.29 less in cash, and per-£ ROI came out 0.6 points apart (-13.5% vs -14.1%), both losing. The strong signal is the bucket pattern above, not the staking rule. The 6–10% band returned the lowest ROI and the 10%+ band the least-bad. Every band finished negative, and ROI does not climb as edge grows. See the full track record for the underlying ledger.
Varying the stake made it slightly worse, and the reason is worth understanding, because it applies to every staking plan ever sold.