Positive Expected Value (+EV) in Ice Hockey Betting

Calculating +EV and Fair Hockey Betting Odds

Three years ago I took a Carolina road moneyline at +145 against a Florida team my model rated as a marginal favourite. Florida won 4-2. I lost the bet. By every traditional measure, the bet was a bad one. By expected value, it was one of the best plays I made that month. The market closed at +120, the line had moved aggressively toward my side after I bet it, and the underlying probability I had estimated turned out — across a season-long sample — to be accurate. The single result was a loss. The bet itself was a win. Learning to distinguish those two things is the entire job.

“+EV betting is the only mathematically proven long-term winning strategy in sports betting,” as the Betstamp methodology puts it. The phrase rolls off the tongue easily; the discipline behind it is what most bettors never actually internalise.

The formula and what it tells you

Expected value on a bet is calculated as (probability of winning × profit if winning) minus (probability of losing × stake lost). For a £100 bet at decimal odds 2.10 where you estimate the win probability at 52 %, the EV is (0.52 × £110) – (0.48 × £100) = £57.20 – £48 = +£9.20.

That positive £9.20 is your expected return per bet, on average, across infinitely many repetitions of the same bet at the same price. Your actual return on any single bet is either +£110 (if you win) or -£100 (if you lose). The expected value sits between those two extremes and represents the long-run average if you could place this exact bet many times.

The implication is crucial. A +£9.20 EV bet that loses on the night was still a +£9.20 EV bet. The result does not retroactively change whether the bet was good. The market moved your way after, the probability estimate held up on the season, and the negative outcome was variance.

Finding the edge

Positive expected value comes from probability estimates that diverge from the market’s implied probability. The market implied probability comes from the price, after stripping out the vig. The bettor’s estimate has to come from somewhere — a model, a structural analysis, an informational edge.

For NHL hockey, the most common sources of edge are: confirmed line-up news the market has not yet absorbed, fatigue spots where the public underweights back-to-back effects, specific team home/road splits that the market prices to league average, and special-teams matchups where strong PP meets weak PK or vice versa. Each of these has been profitable across the long-run samples I have tracked.

The mistake to avoid is treating a hunch as an edge. “I think Tampa wins” is not an edge. “I estimate Tampa wins 60 % of the time in this matchup because their PP-to-PK matchup advantage is 8 percentage points and the goalie matchup favours them by another 3” is an edge. The first is a guess. The second is a defensible probability estimate.

EV versus CLV

Expected value and closing line value are related but distinct concepts. EV is forward-looking — your estimate of the bet’s value at the time you place it, based on your own probability estimate. CLV is backward-looking — a measurement of how the price moved after you bet, used as a proxy for whether the market eventually agreed with you.

Positive EV and positive CLV usually go together but not always. A bet with high EV by your model can have negative CLV if the market moves against your side because of news you did not have. A bet with low EV by your model can have positive CLV if the market panics on bad news that does not actually change the underlying probabilities much.

Both measurements matter. EV tells you whether your process is generating value at the time of placement. CLV tells you whether your timing is putting you on the right side of subsequent market movement. Across a large sample, the two converge if your estimates are accurate.

Sample size and variance

Hockey is high-variance. A single goal swings games, save percentages bounce around even for elite goalies, and individual matchups are noisy. The implication for EV bettors is that the sample size required to know whether your process is working is much larger than you think.

For an estimated 3 % edge on flat-staked bets, you need roughly 1,000 bets to be 95 % confident your edge is real rather than noise. A bettor placing five hockey bets a week reaches that sample in four years. That is not a typo. The variance is enormous, and the time horizon to verify your own process is genuinely measured in years.

This is why CLV is useful as a faster signal — it converges within hundreds of bets rather than thousands, because it does not depend on the binary outcome of each individual wager. Positive CLV with negative ROI over 200 bets is reassuring. Negative CLV with positive ROI over 200 bets is a warning.

Common EV mistakes

Three errors recur. First, treating every winning bet as a confirmation of process and every losing bet as bad luck. Outcomes are noise. Process is signal. The bet that loses but had positive EV at placement was still a good bet.

Second, scaling up after a hot streak. A run of winners is statistically consistent with both excellent process and pure luck. Increasing stakes after the hot streak without a process improvement is gambling, not investing. Stake sizes should respond to changes in your assessed edge, not to recent results.

Third, ignoring the vig. Many bettors estimate probabilities, find a price that beats their estimate by a margin, and bet without checking the no-vig fair line. A bet at 2.10 with a 50 % estimated probability and 10 % vig has a fair line closer to 2.00 — the apparent edge is 5 % rather than 10 %. Always strip out the vig before claiming the edge.

Putting it all together

The bettor’s job is to build defensible probability estimates, calculate the implied probabilities of the available prices, identify the gap, and bet only when the gap is large enough to clear the vig and provide a margin of safety. Fractional Kelly sizes the stake. CLV measures the process across hundreds of bets. ROI measures the results across thousands.

None of this is glamorous. The work is repetitive — line-up news, goalie confirmations, pace projections, comparison shopping across UK books. The reward is a positive expected value distribution that, given enough time, produces a positive bankroll trajectory. The catch is that “enough time” is years, not weeks.

The bettors who succeed are the ones who can hold the line during the variance — staking discipline intact, process unchanged, ego in check — through the inevitable drawdowns that are statistically certain to happen even on a positive-EV strategy. That mental discipline is harder than the maths. Most of the bettors I know who eventually go broke do so because of stake-sizing decisions made during a downswing, not because their underlying process was wrong.

For the measurement side of this process — the closing-line-value framework that tells you whether your EV estimates are translating into market-confirmed edges — the closing line value in hockey betting piece is the operational companion.

Published by the Ice Hockey Betting Tips team.