Methodology — how to read our numbers

We publish a strike rate and an ROI. Both are easy to misread. This page explains exactly what they mean, what they don't mean, and what they translate to in pounds and time.

In one paragraph

Our backtest of model v1.3 returned +8.0% ROI per pound staked over 1,252 picks across 5,877UK & Irish races (2025-11-18 → 2026-05-17). That is not a 8.0% annual return on capital. It means every £1 you stake returned an average £1.8 — not that your bankroll grew by 8.0% over a year.

ROI is not APR

These two abbreviations look similar and mean very different things.

ROI (return on investment)

In betting, ROI = total profit ÷ total amount staked. It's a per-stakefigure, not a per-year figure. There's no compounding. Every £1 lives for a few hours, gets settled, and is gone or returned with profit.

APR (annual percentage rate)

A finance concept. APR is the annual return on capital that stays put — a savings account paying 5% APR turns £1,000 into £1,050 after a year, compounding optional. Tipping numbers cannot honestly be reported as APR.

What +8.0% ROI actually looks like

To translate ROI into pounds and time, you need two things: how many tips you act on per year, and how big each stake is relative to your bankroll (typically 1–2%).

Strategy	Tips / yr	Pts staked / yr	Pts profit @ +8.0% ROI
Public tip-of-the-day (Racing Alpha)	365	365	+29.2
Every v1.3-eligible race	~2,500	~2,500	+200

Translated to pounds with a £1,000 bankroll:

Stake size (1pt =)	% of bankroll	Tip-of-day profit / yr	Equivalent annual %
£10	1% (conservative)	≈ £292	≈ 29%
£20	2% (standard)	≈ £584	≈ 58%
£50	5% (aggressive — high risk)	≈ £1460	≈ 146%

These figures assume the +8.0% backtest edge persists live, that you can take advised prices, and that you do not increase stake size during winning runs. None of those are guaranteed.

What these numbers don't mean

Variance is brutal

At a 22.4% strike rate, you'll have 6+ losing tips in a row regularly. That's ordinary noise, not a broken model. A good year of tipping at this strike rate could still land anywhere between roughly −20 and +60 points purely due to variance — the +8.0% is the long-run expected value, not the outcome you'll see in any single year.

Edge can erode at the prices you actually take

Backtests settle at SP (starting price), which is the market's final consensus. By the time a race goes off, prices on AI-favoured horses have usually tightened. If you can grab early prices (Best Odds Guaranteed bookmakers, exchange tops), you may match or beat the backtest. If you wait until the off, you're settling at SP and the live edge is likely a little lower.

The selection rules were tuned on the same data

v1.3's segment filter and price band were derived from the same 6-month backtest we report against. There's a modest overfitting risk — true live ROI is typically 1–3pp lower than backtest ROI. A realistic live expectation is +5 to +8% ROI, not a guaranteed +8.0%.

6 months is not 12

Our window (Nov 2025 → May 2026) is heavy on winter racing — more jumps, more soft/heavy ground. Summer racing (turf flat, faster going) is under-represented. We will re-backtest each quarter and re-validate every model assumption.

Betting is not investing

ROI does not compound. Drawdowns of 10–20% of bankroll happen even on profitable systems. You can lose money. This is not a savings account, an index fund, or any kind of guaranteed return. If you cannot afford to lose what you stake, do not stake it.

What we promise

Every tip is published before the off. No retro-editing. No quietly dropped losing runs. See the live performance ledger.
Every backtest is public. Full segmented breakdown, every pick stored. See /backtests.
The model is versioned. When we change the algorithm, the old version stays queryable. You can verify our claims at any time.
No “guaranteed winners” language. Anywhere. Past performance is not a guarantee of future results.

Going deeper

Racing Alpha is information and entertainment for adults aged 18 and over. We do not accept bets and we are not a betting intermediary. Past performance does not predict future results. Free help: BeGambleAware · GamCare · GAMSTOP.