Risk of ruin is the probability that a trading account crosses a specified failure threshold under a stated model. Without the threshold, horizon and return assumptions, the percentage is not a property of the strategy; it is a property of an unstated question.
Define ruin before calculating it
In ruin theory, the question is how much starting capital is needed to keep the probability of eventual ruin below a chosen level. That is a modelled threshold, not necessarily literal zero. Researchers also treat the maximum deficit of a ruin process as a risk measure (Trufin, Albrecher and Denuit, Properties of a risk measure derived from ruin theory).
For a trader, “ruin” could mean hitting a prop firm's loss boundary, falling below a minimum operating balance, or losing all capital. Those are different events. A prop-rule breach is not the same as zero wealth, and the rule itself may change; the estimate must name which one it models.
Position size changes the path
Position sizing changes the distribution of account outcomes even when the entry and exit signals stay fixed. A study of technical timing strategies finds that relative position sizing materially affects risk and return, and that smaller fractions often had higher risk-adjusted returns in its simulated scenarios (Scholz, Size matters!). That is a result under the paper's setup, not a universal sizing rule.
Kelly's framework chooses a bet fraction to maximise expected long-run logarithmic wealth under a specified outcome distribution. A separate risk-constrained formulation explicitly limits the probability of a wealth drawdown and studies the trade-off with growth (Busseti, Ryu and Boyd, Risk-Constrained Kelly Gambling). Neither turns an estimated edge into certainty: if the assumed win probability, payoff distribution or dependence is wrong, the calculated risk changes too.
Why one percentage can mislead
A useful estimate must make its modelling choices visible:
- Failure boundary: initial-capital loss, running-peak drawdown, or a firm's specific rule.
- Horizon: a finite evaluation period or an unbounded “eventual” probability.
- Trade outcomes: fixed wins and losses, or a distribution that includes variable exits and costs.
- Dependence: whether consecutive trades are treated as independent, despite clustering or shared exposure.
- Sizing rule: fixed units, a fixed fraction of current equity, or a rule tied to estimated Kelly growth.
These choices matter because a strategy can avoid zero while crossing a practical loss limit. A model that assumes independent identical trades can also understate risk when losses cluster. Ruin-theory research makes the threshold and probability explicit; it does not certify that a trading return model describes a particular account.
How to test a risk-of-ruin estimate
Treat the estimate as a falsifiable model, not a badge. First write down the exact failure boundary and observation horizon. Then estimate trade outcomes using only information available at each point, including fees and the strategy's actual stop or exit behaviour. Compare at least two sizing rules on the same signal history so sizing, rather than a changed entry, explains any difference.
Stress the inputs that are least certain: lower the estimated edge, increase costs, and resample outcomes in blocks so losing clusters can survive the resampling. Report how often each path crosses the chosen boundary, the range of outcomes across resamples, and the assumptions that move the result. These are proposed test procedures, not empirical findings from the cited papers.
For a related prop-account framing, see risk of ruin for prop traders. realbacktesting is a trading-software studio for cTrader; its published cBot results are intended to be reproducible in the reader's own cTrader, not treated as a live guarantee (how the results are verified).
The point to keep
“Risk of ruin” does not answer whether a strategy is safe. It answers how often a specified model crosses a specified line. If either part is hidden, the decimal places are decoration.