Why your ROI is probably wrong
Six ways a return-on-investment figure overstates performance, all of them accidental and all of them common: turnover versus stake, excluded voids, forgotten bonus money, survivorship in the record, unequal stakes counted equally, and the sample being far too small to mean anything.
The most common error is dividing profit by the wrong denominator. ROI on turnover (profit ÷ total staked) and return on bankroll (profit ÷ starting bank) answer different questions and differ by a factor equal to how many times you turned the bank over. Quoting one while implying the other is how a 3% edge becomes a "30% return".
The six errors
- Wrong denominator. £300 profit from £10,000 staked is 3% ROI on turnover. If the bankroll was £1,000, that is also a 30% return on bankroll. Both are true; neither describes the other.
- Voids excluded. Bets voided by rule changes, postponements or palpable errors are quietly dropped from many records. They are not random: they cluster around the mistakes and the outlier prices.
- Bonus money as profit. A matched free bet that lands adds to profit without any forecasting involved. Separating promotional returns from selection returns is the only way to know which one is working.
- Survivorship. Bets that were "not really part of the strategy" get excluded after they lose. Applied consistently, this converts any record into a winning one.
- Unequal stakes counted equally. Averaging the ROI of individual bets is not the same as the ROI of the portfolio. If the losers were larger, the average lies.
- Sample too small. See below. This is the largest of the six and the least visible.
How much sample does an ROI figure need?
The uncertainty in an ROI estimate scales with the odds and inversely with the square root of the number of bets. At average odds of 2.00 and flat stakes, the approximate 95% interval around a measured ROI is:
| Bets settled | Approx. 95% interval around measured ROI | What that means |
|---|---|---|
| 50 | ±28 pp | A measured +10% is compatible with −18% to +38% |
| 100 | ±20 pp | Still uninformative |
| 500 | ±9 pp | A measured +10% is probably positive |
| 1,000 | ±6 pp | Usable |
| 5,000 | ±3 pp | Reliable |
The intervals widen at longer odds. A record built on prices around 4.00 needs roughly three times the sample of one built at 2.00 to reach the same precision — which is why an impressive ROI on 60 bets at long prices tells you nothing at all.
This cuts against us as much as anyone. Our own published sample is young, and we say so on the front page rather than in a footnote: the record is a starting point, not a proof. Anyone quoting an ROI without a sample size beside it is either unaware of this or relying on you being unaware.
The version that is hard to fake
A defensible performance statement contains five numbers together: bets settled, average odds taken, profit in units, maximum drawdown, and the period covered. Any one of them alone can be arranged to flatter. All five together are difficult to make look good unless they are good.
Check any of this against our record
Every signal CONSENSUS publishes carries the bookmaker odds fixed before the event starts and the settled result afterwards — including the drawdowns and the losing runs. The running total is on the front page and every entry is in the log.