Variance and drawdowns
This lesson is about the part of betting nobody advertises: losing, often and in clusters, even when you are doing everything right. Knowing what normal looks like is what keeps a good process from being abandoned at the worst moment.
What a good bettor's swings look like
Take a bettor who wins 55% of their bets at −110. That is a strong long-run record, and most bettors never reach it. Each bet earns about 5% on average, so 500 bets at 1 unit each should come out about 25 units ahead. On the way there, the same bettor will typically:
- hit a losing run of about 7 in a row somewhere in those 500 bets, and a run of 9 or more in about 1 season in 5;
- see their bankroll drop roughly 10 to 25 units from a high before recovering, in our simulations;
- still finish all 500 bets behind about 1 time in 9.
None of that means anything is broken. It is what 55% looks like. Variance is the cost of betting at all, and a real edge only shows through it over a long run.
How many bets before a record means anything
A 55% bettor and a break-even bettor at 52.4% look almost identical over 100 bets. One standard error on a 100-bet win rate is about 5 points, twice the gap between them. At 500 bets it is still about 2.2 points. It takes roughly 1,400 bets before a true 55% record sits two standard errors clear of break-even.
Luck fools people in the other direction too. Give 35 people a coin and have each pick 100 games with it: on average, one of them posts a 60% record. A hot month proves as little as a cold one. That is why closing line value matters: it tells you whether your prices are good long before results can.
Losing runs are normal. Chasing is the danger.
What breaks bankrolls is not the losing run but the reaction to it: raising stakes to win it back, adding bets you wouldn't normally make, or abandoning a sound process after two bad weeks. The next bet doesn't know about the last one.
- Size for the bad run, not the good one. At 1u a bet, a 25-unit drawdown is a quarter of your starting bankroll. Consensus Edge keeps its suggested stakes small for exactly this reason.
- Judge your process on hundreds of bets and on CLV, not on a week.
- Decide your limits before you start, and keep betting money separate from everything else.
- Take a break when it stops being fun. If you find yourself chasing losses or betting money you need, stop. The responsible gaming resources linked at the foot of this page are there for that.
Example. 500 bets at 55%, all at −110, 1u each (computed or simulated, approximate where marked).
| Figure | Working | Result |
|---|---|---|
| Average result per 1u bet | 0.55 × 0.909 − 0.45 × 1 | +0.05u (5%) |
| Expected profit | 500 × 0.05u | +25u |
| Typical longest losing run | ln(500 × 0.55) ÷ ln(1 ÷ 0.45) = 5.62 ÷ 0.80 | about 7 |
| Longest run of 9 or more | computed | about 1 in 5 |
| Biggest drop from a high | simulated | roughly 10u to 25u |
| Finishing behind | 261 wins or fewer, since 1.909 × 261 = 498.3 is under 500 | about 1 in 9 |
Profit after 500 bets is 1.909 × wins − 500 units, so a bettor who wins 261 times ends down about 1.7u despite a real edge. At $10 a unit, a 20-unit drawdown is $200 gone before the edge shows. Expect it, size for it, and don't chase it.
Terms in this lesson
VarianceUnitBankrollWin RateROIClosing Line ValueKelly Criterion