How long a losing run can honestly last
Most people's sense of a normal bad run is far too short. Here are the actual numbers, and what they mean for reading your own record.
On this page
Ask somebody with a 50% method how many losses in a row would worry them and many say five or six. Over 500 independent bets, a run of eight losses is more likely than not. A run of ten is plausible but not typical.
The numbers
At a 50% strike rate, the probability of any given run of ten starting at a particular bet is roughly one in a thousand. That sounds reassuring until you count how many starting points a year contains. Over 500 bets there are hundreds of overlapping places such a run could begin. Exact recursion for the longest run puts the probability of at least one run of eight at about 62.5%, of nine at about 38.5%, and of ten at about 21.5%.
Drop the strike rate to 40%, which is entirely normal for somebody backing
prices around 2.60, and the loss probability becomes 60%. Across 500
independent bets, a run of eleven or more then has about a 51.6% chance. A run
of fifteen or more is still possible, but at about 8.8% it is not
unremarkable.
Why this is the most useful thing in bankroll management
A staking plan calibrated only to a short drawdown can fail when a longer run arrives. The run may be compatible with the assumed strike rate even when the stake size was not compatible with surviving it.
A practical stress test is "what happens to this bankroll after fourteen consecutive losses". If the plan is meant to survive that scenario and the answer is that there is nothing left, its stake is too large for its own constraint. That is the argument of staking without going broke, reduced to one number.
What it means for reading a record
A run is evidence about a method, but it throws away most of the record's information. Judge it alongside prices, total wins and losses, calibration and whether the assumptions changed. It also tests whether the staking plan can survive a repeat.
What this does not tell you
These numbers assume independent bets with one constant win probability. Real portfolios can violate both assumptions. Correlation can increase clustering, but its effect depends on the direction and structure of that dependence, so the figures above are a modelled scenario rather than a universal floor.
Common questions
How long can a losing run last with a winning method?
Longer than intuition suggests. With independent bets and a 50% win rate, 500 bets have about a 63% chance of including eight straight losses and about a 21% chance of including ten.
Does a long losing run mean the edge is gone?
Not by itself. A run is evidence, but its meaning depends on the strike rate, odds, sample size, dependence between bets and the rest of the record.
How should a losing run change what I do?
Do not raise stakes to recover losses. Compare the run with the range expected under the model, review whether its assumptions still hold and follow a precommitted loss limit.