Closing line value in football betting: what CLV proves and what it doesn't
Published
Almost every way of judging a bet requires waiting for the result, and the result is mostly noise. Closing line value — CLV — is the exception. It scores the price you got against the price the market settled on at kick-off, and it can be computed the moment the match starts, whether the bet won or lost. That makes closing line value the fastest honest signal available about whether a betting record reflects skill or luck, which is why it belongs on any track record that claims to be auditable, including our own. It is also the metric most commonly computed wrong — and wrong in a direction that flatters everyone who computes it.
What closing line value actually measures
A bookmaker opens a market with an estimate, then lets money and information move the price. Bets from people who know something, injury news, lineup confirmations, weather — all of it gets absorbed until the market closes at kick-off. That closing price is the market’s final, best-informed opinion, and for major football markets it is the most accurate single forecast publicly available for that match.
Closing line value asks one question of your bet: did you buy at a price better than that final opinion? If you did, you were holding a view the market had not yet fully priced, and you were holding it early enough to act. If you consistently do it across hundreds of bets, you found something the market took time to find.
Note what the question does not involve. It does not involve the score. A bet at 2.10 on a side that closes at 1.90 has positive CLV whether the team wins 3–0 or loses to a stoppage-time own goal. That is the entire point: CLV is a leading indicator that measures your decision, while profit is a lagging indicator that measures your decision plus several hundred coin flips.
How to calculate closing line value in football
With decimal odds the arithmetic is one line:
CLV % = (odds you took ÷ closing odds − 1) × 100
Back the home side at 2.10 and watch the market close at 1.95:
(2.10 ÷ 1.95 − 1) × 100 = +7.69%
Take 1.80 on a side that closes at 1.90 and you get the mirror image:
(1.80 ÷ 1.90 − 1) × 100 = −5.26%
Handicap and goal-line markets need one extra step, because there the line can move as well as the price. A bet struck at −0.5 that closes at −0.75 has gained value even if the odds are unchanged at 1.90 in both cases. The clean way to handle this is to stop comparing prices and start comparing probabilities: convert both your price and the closing price to implied probabilities, then compare those. Our guide to converting odds to implied probability covers the conversion itself.
The step most CLV guides skip: strip the margin first
Here is where nearly every closing line value calculation goes wrong. The closing odds you are comparing against are not the market’s probability estimate — they are that estimate with the bookmaker’s margin baked in. Compare a price you took against a margined closing price and you credit yourself with the bookmaker’s cut.
Take a three-way market closing at home 1.95, draw 3.60, away 4.20. The implied probabilities are:
- Home: 1 ÷ 1.95 = 0.5128
- Draw: 1 ÷ 3.60 = 0.2778
- Away: 1 ÷ 4.20 = 0.2381
Those sum to 1.0287 — an overround of 2.87%. Scaling each back so they sum to one gives a fair home probability of 0.5128 ÷ 1.0287 = 0.4985, which is fair odds of 2.006.
Now redo the earlier example against that fair closing price rather than the posted one:
(2.10 ÷ 2.006 − 1) × 100 = +4.69%
The apparent +7.69% was really about +4.7%. Roughly three points of “edge” were the bookmaker’s margin handed to you by the arithmetic. And because proportional de-vigging scales every outcome by the same factor, this is not a quirk of the example: skip the de-vig step and every bettor’s CLV is inflated by approximately the closing book’s overround. On a market with a 5% overround, a bettor who is exactly break-even against the true closing line will still report about +5% CLV and conclude they are sharp.
A closing line value figure that does not say whether the margin was removed is therefore not interpretable at all.
There is a reason to care about this beyond bookkeeping. Once the margin is gone, de-vigged CLV stops being a proxy for edge and becomes the edge. The fair home probability above was 0.4985, and the bet was struck at 2.10, so the expected value of that bet is 0.4985 × 2.10 − 1 = +4.69% per unit staked — the same number the de-vigged CLV calculation returned, because it is the same arithmetic seen from the other side. On an efficient close, margin-adjusted CLV is expected value. Raw CLV, by contrast, is not an estimate of anything.
One caveat on the method itself. Scaling every outcome by the same factor, as above, is the simplest way to remove the margin and not the most accurate. Bookmakers do not spread their cut evenly: they load more of it onto longshots, so proportional de-vigging understates the fair price of favourites and overstates it on long prices. On a heavy favourite or a 6.00 shot, methods that model that skew give a materially different fair value. A CLV figure is only ever as good as the de-vigging behind it.
Why closing line value converges faster than profit
The practical case for CLV is statistical. Consider flat-staking one unit at odds of 2.10 on a genuinely fair coin-flip market. Each settled bet returns either +1.10 or −1.00, which gives a standard deviation of about 1.05 units per bet — enormous relative to any realistic edge.
Suppose your true edge is 5% per bet, or +0.05 units. To separate that from zero at two standard errors you need n bets where:
0.05 × n ÷ (1.05 × √n) = 2, so √n = 42, so n ≈ 1,760 bets
Halve the edge to 2% per bet and the requirement rises to roughly 11,000 bets. At a handful of bets a week, that is a lifetime. This is why a profit curve over a few hundred selections tells you almost nothing, and why hot streaks are so easy to mistake for evidence.
CLV changes the arithmetic because every bet yields a measurement, and the measurement does not depend on the outcome. If your average CLV is +2% with a per-bet spread of around 5%, the same two-standard-error test needs only a few dozen observations in the idealised case. Treat that number with suspicion — your bets are not independent draws. You are likely betting similar markets, reacting to the same team news, and clustering around the same kick-off times, so the effective sample is considerably smaller than the raw count. Even after discounting heavily for that, the honest comparison is a few hundred bets against many thousands. It is an order of magnitude, and it is the difference between a record that can be evaluated and one that cannot.
That gap is what makes CLV the right first question to ask of any published betting record, and it gives you a reading rule. A record showing strong profit alongside negative average CLV is describing good fortune: it bought worse prices than the market’s and won anyway, which is a thing that happens and does not repeat. A record showing modest profit alongside consistent positive CLV is describing a process, and processes are the things that continue.
How a CLV claim gets inflated
Because the metric is powerful, it is worth knowing how it gets bent:
- Unavailable advised prices. Quoting a best price that existed briefly at one book, at limited stakes, against a closing price from elsewhere. The CLV is real on paper and unreachable in practice.
- Shopping the comparison. Closing prices differ between books. Comparing a soft book’s generous opening price against a sharp book’s close is a different — and much more flattering — claim than comparing like with like.
- No de-vig. As shown above, worth roughly the overround to everyone who omits it.
- Selective reporting. CLV is only meaningful over every bet placed, not the ones remembered.
The fix is disclosure, not trust: a CLV figure is auditable only if it names the closing price source, the timestamp used as “close”, whether the margin was removed, and how many bets it covers.
What closing line value cannot tell you
CLV is a strong signal, not a complete one, and three limits matter.
It is not profit. Consistently beating the close and consistently making money are different achievements, separated by stake limits, availability and execution. Records can be positive on one and negative on the other.
It is not proof of a model. Consistently beating the close can come from forecasting better than the market, or simply from acting faster on public information before the price adjusts. Both make money while they last. Only one of them is a prediction model, and only one keeps working once the speed advantage is gone — which is why a CLV figure means much more when it sits next to a probability record that can be scored independently.
It assumes the closing line is efficient. In deep markets — major-league 1X2, main goal lines — that assumption holds up well, which is precisely why beating those closes is so difficult and so rarely done. In thin markets, obscure divisions and exotic props, the close may be barely more informed than the open, and beating it proves correspondingly less.
Finally, CLV evaluates prices, not probabilities, and it only exists where a bet exists. It grades selections — the subset of matches where somebody thought the price was wrong and acted — so it is silent on every fixture you passed over, and silent on whether the probabilities behind the decision were honest. A published prediction model has the opposite obligation: it emits a number for every fixture, including the many where it simply agrees with the market, and it has to be right across all of them. That is a job for a proper scoring rule. The Brier score and its relatives grade the full set of forecasts against what actually happened, and they punish overconfidence and vagueness in ways a selection-level metric cannot.
Did I get a good price? is closing line value. Does 70% from this model mean 70% in reality? is calibration. The two are complements, and a record showing both — decision quality against the market and probability quality against reality — is far harder to fake than one showing either alone.