Methodology
What Expected Value Really Means in a Synthetic Market
EV is the most quoted and least understood number in betting mathematics. Here is what it measures, what it ignores, and why a positive-EV lab result is not a promise.
Key takeaways
- EV is the centre of a distribution, not a forecast of any single result.
- The break-even probability of a price is exactly its implied probability — clear that, or EV is negative.
- Overround is a cost paid regardless of skill; a model edge must clear it before EV turns positive.
- EV is only as good as the probability estimate feeding it; small estimation errors flip the sign.
- Positive EV and a long losing run are entirely compatible over any sample you will actually observe.
EV is an average, not an outcome
Expected value is the average profit per unit staked if you could replay the same bet infinitely under fixed assumptions. At +100 with a true 60% chance, EV is 0.6 × 1 − 0.4 × 1 = +0.20 per unit. That does not mean you make 20% — it means that is the long-run centre of a very wide distribution of results.
In a synthetic market we can compute EV exactly because we set the true probability. That is a feature for learning and a trap for intuition: real markets hide the true probability, so real EV is always an estimate.
The formula is worth writing out once, because everything else follows from it. For a price with net decimal payout b and true win probability p, EV per unit is p × b − (1 − p). Setting that to zero and solving gives the break-even probability p = 1 / (b + 1) — which is precisely the price's implied probability. A price is not a prediction; it is a threshold.
The vig moves the goalposts
Prices carry a margin. When the implied probabilities of a market's outcomes sum to more than 100%, that excess is the overround, and it is a cost you pay regardless of skill. Novus Odds derives the real overround from the priced-versus-fair lines rather than echoing the margin you dialed in, so the vig you see is the vig the prices actually contain.
The practical consequence: a model edge has to clear the vig before expected value turns positive. Raise the margin in the Odds Lab and watch how much true-probability advantage it takes just to break even.
A concrete two-way market: both sides priced at -110 imply 52.38% each, summing to 104.76%. The overround is 4.76% of the book, and the fair probabilities after removing it are 50/50. To break even you need to be right 52.38% of the time; to have any edge at all you need better than that on a genuinely even proposition. The margin is not a fee taken from winnings — it is a shift in the threshold you have to beat.
EV inherits every error in your probability
EV is a two-input function, and one of those inputs is known exactly while the other is estimated. The price is a fact. The probability is a belief. So the sign of your EV is entirely at the mercy of how good that belief is.
Return to the -110 example. Suppose your model says 55% and you compute an EV of 0.55 × 0.909 − 0.45 = +0.05 per unit, a healthy 5%. Now suppose the model is two points optimistic and the truth is 53%. EV falls to 0.53 × 0.909 − 0.47 = +0.012. Three points optimistic and it is roughly zero. Four and you are losing money while your spreadsheet still reports a positive number.
This is why calibration work matters more than edge-hunting. A model that is honest about being 53% is more useful than one that confidently claims 58% and is wrong about it, because the first one can be corrected and the second one cannot even be diagnosed.
Why positive EV is not a promise
Even with a genuine edge, variance dominates the short run. Positive-EV bets lose all the time; negative-EV bets win all the time. The lab makes this visible: run the same positive-EV configuration under different seeds and compare expected ROI to observed ROI. The gap between them is variance, and it is the reason bankroll management exists.
Put numbers on it. Flat-staking a 5% edge at even money, the standard deviation of a single unit outcome is close to 1. Over 400 selections your expected profit is 20 units and the standard deviation of that total is about 20 units as well. A one-sigma bad run lands you at zero after four hundred decisions — with the edge fully intact the entire time.
Nothing about that is a failure of the model. It is what a positive-EV process looks like from the inside, and it is the reason 'am I winning?' is a much worse question than 'is my probability estimate calibrated?'
Frequently asked
Is a positive-EV bet a good bet?
Positive EV is necessary but not sufficient. A bet also has to be sized so the variance is survivable, and the probability estimate behind it has to be trustworthy. Positive EV computed from a badly calibrated model is just a confidently wrong number.
How do I find the break-even probability of a price?
It is the price's implied probability: 1 / (net decimal payout + 1). For +150 that is 1 / 2.5 = 40%; for -110 it is 1 / 1.909 = 52.38%. Beat that rate and EV is positive, before margin considerations across the whole book.
Does the overround come out of my winnings?
No — it raises the win rate you need. Both sides of a -110 market imply 52.38%, summing to 104.76%. The extra 4.76% is the book's margin, and it means an even-money proposition has to be won 52.38% of the time just to break even.
How long until EV shows up in results?
Longer than most people expect. The signal grows with the number of selections while the noise grows with its square root, so the ratio improves only as the square root of sample size. Small edges need thousands of decisions before the average becomes visible above the noise.
Try it yourself
Everything in this article is something you can run and tweak in the lab — with your own settings and a reproducible seed.
Open the lab →