Variance is the statistical spread of possible outcomes around an expected value (EV). Two bets can carry the exact same EV — the same theoretical profit per dollar staked over the long run — and still behave completely differently in practice, because one clusters tightly around that average while the other lurches between big wins and steady losses. Edge tells you the direction you should be heading. Variance tells you how rough the road will be on the way there.

The size of the swings comes directly from the shape of the bet. A bet with a high probability of winning a small amount (a heavy favourite at decimal odds of 1.20, say) pays out often but rarely by much, so results bunch close to the mean. A bet with a low probability of winning a large amount (a 25.00 outsider, a same-game parlay, an each-way bet on a 40-horse handicap) pays out rarely but massively when it lands, so the outcomes are spread far and wide even if the long-run EV is identical. Stack multiple legs together, as in accumulators, and variance compounds fast — each additional leg multiplies the odds and shrinks the win probability, stretching the distribution further from the average without necessarily improving the edge.

This matters because a bettor’s actual experience is never the EV — it’s one draw from a wide distribution of possible results. A model or price that’s genuinely +3% against the market can still produce a losing month, a losing quarter, sometimes a losing year, purely from variance. Confusing a bad run with a bad model (or a hot streak with real skill) is one of the most expensive mistakes in betting, and it’s variance, not edge, that’s usually to blame.

Example

A bettor finds a market they believe is mispriced: a two-way line priced at decimal odds 1.91 (roughly -110), where their model puts the true win probability at 54% instead of the 52.4% the price implies. They stake $100 per bet, flat, across 100 of these bets.

Per bet: a win returns $91 profit (100 × 0.91), a loss costs $100.

  • EV per bet = (0.54 × $91) + (0.46 × −$100) = $49.14 − $46.00 = $3.14
  • Expected profit over 100 bets = 100 × $3.14 = $314

Now the variance. For a single bet, E[X²] = (0.54 × 91²) + (0.46 × 100²) = 4,471.74 + 4,600 = 9,071.74. Subtracting EV² (3.14² ≈ 9.86) gives a per-bet variance of about 9,062, so the standard deviation per bet is √9,062 ≈ $95.20 — over thirty times the size of the expected profit on that single bet.

Because the bets are independent, variance adds across them: total variance = 100 × 9,062 = 906,200, so the standard deviation of the total result after 100 bets is √906,200 ≈ $952.

That means a bettor with a genuine, provable 3%+ edge should expect to land, one standard deviation either side of the mean, somewhere between roughly −$638 and +$1,266 after 100 bets — a real chance of being underwater despite doing everything right. Only after several hundred or a few thousand bets does the SD of the total (which grows with √n) become small relative to the accumulating EV (which grows linearly with n), and the edge starts to show through the noise.

Key Points

  • Edge and variance are separate axes: edge tells you if a bet is worth making; variance tells you how unpredictable the ride will be. A +EV bet can still be high-variance, and a break-even bet can be low-variance — don’t judge a strategy by how smooth its results feel.
  • Longer odds mean fatter swings: longshots, each-way bets, and especially multi-leg parlays carry far more variance per dollar of EV than short-priced favourites, because the win probability is low and the payout is concentrated in rare events.
  • Small samples don’t prove anything: as the worked example shows, a real 3% edge can still show a loss after 100 bets. Don’t discard a well-tested model after a bad month, and don’t trust a hot streak as evidence of skill until the sample is large enough for variance to have settled down.
  • Size your stakes for the variance you’re taking on: flat staking or fractional Kelly staking exist specifically to keep a high-variance strategy from wiping out a bankroll during a normal, expected downswing rather than an actual breakdown in edge.
  • Track process, not just results: because short-term outcomes are so variance-dominated, serious bettors judge themselves on whether their prices beat the closing line (closing line value) rather than on whether last week’s bets won, since CLV converges to the truth far faster than raw profit and loss does.
  • Correlated bets don’t reduce variance the way independent ones do: stacking multiple bets on the same match or event (backing a team and their top scorer, for instance) can look like diversification but actually keeps the outcomes tied together, so the swings don’t shrink the way they would with truly independent action.