Study 07 · risk methodology

Measuring drawdown and probability of ruin

A worked methodology for peak-to-trough drawdown, finite-horizon ruin estimates, and Wilson confidence intervals using 25,000 fixed-seed sessions.

Profit and loss at the final spin hides the route taken to get there. This study defines the risk measures used throughout 0xRoulette, applies them to 25,000 fixed-seed sessions per bankroll, and publishes confidence intervals instead of treating a simulation percentage as exact.

25,000
Sessions per bankroll
500 spins
Session horizon
$10
Fixed wager
Wilson 95%
Confidence interval

Definitions

For equity values E₀…Eₜ, the running peak at time t isMₜ = max(E₀…Eₜ). Drawdown is Dₜ = Mₜ − Eₜ; maximum drawdown is the largest Dₜ in the session. Ruin means the bankroll reached zero before the scheduled 500th spin. We estimate finite-horizon ruin as ruined sessions divided by all sessions, then attach a Wilson score interval for the unknown probability.

Master seed
0x6d2ba4ff
Wheel / bet
European / $10 red
Starting bankrolls
$100, $250, $500, $1,000
Shared schedule
The same 25,000 seeds at every bankroll level

Ruin depends on bankroll relative to the wager

BankrollUnitsRuin estimate95% intervalMedian max DDP90 max DD
$1001081.740%81.256%82.214%$150$250
$2502546.768%46.150%47.387%$270$350
$500508.440%8.102%8.791%$310$510
$1,0001000.028%0.014%0.058%$310$530
$600$420round 500
Worked path 0xd12a0d0d: $500 starting bankroll, $10 flat bet, maximum drawdown $170.

Reading drawdown correctly

Maximum drawdown can exceed the initial bankroll. A session might first rise above its starting value, establish a higher peak, and then fall to zero. That peak-to-trough move is larger than the starting cash. Drawdown also says nothing about recovery time unless that is measured separately. Two paths can share a maximum drawdown and reach it at very different speeds.

The 100-unit bankroll produced only 7 ruins in 25,000 trials, so the estimate is 0.028% but the interval remains wider than the point estimate. Reporting all three numbers prevents false precision.

Limitations

Ruin here is finite-horizon: it asks what happened within 500 spins. Over an unlimited horizon, a negative-drift game answers a harsher question. Wilson intervals quantify Monte Carlo sampling error, not model error. They do not account for wheel bias, changing wagers, correlated outcomes, or a mismatch between simulated rules and a real table.

Audit the numbers.The chart data is checked in as plain CSV; the combined snapshot is also available as JSON.

Sources and implementation references