Risk of Ruin: The Math That Keeps Accounts Alive

Define Ruin Before Calculating It

“Ruin” can mean literal bankruptcy, a personal stop level, a maximum drawdown from peak equity, a broker liquidation threshold, or a prop-program rule breach. Those are different stopping events. A probability calculated for one cannot be relabeled as another.

A trader can choose a conservative boundary as a policy, but mathematics does not declare one drawdown level universally “safe,” “gambling,” or terminal. The risk-management framework helps translate the modeled boundary into an operating rule.

The Classical Gambler's Ruin Formula

The textbook model is useful precisely because its assumptions are narrow. The process starts i equal units above zero. Each independent round moves exactly one unit up with probability p or one unit down with probability q = 1 − p. The stake does not change.

Against an infinitely rich opponent, the probability of eventually hitting zero is:

P(ruin) = (q / p)ⁱ when p > q; when p ≤ q, eventual ruin is 1.

With a second absorbing boundary at N, the probability of reaching zero before N is:

P_i(hit 0 before N) = [(q/p)ⁱ − (q/p)^N] / [1 − (q/p)^N] for p ≠ q. In the fair case, it is 1 − i/N.

These are exact results for that random walk. They are not automatically exact for a trading record with unequal R-multiples, variable size, slippage, gaps, correlated setups, a finite evaluation, or a trailing drawdown.

A Reproducible Toy Example

Assume independent even-payoff rounds, p = 0.55, q = 0.45, a fixed one-unit stake, no costs, no upper target, and an infinite horizon. Starting 10 units from zero gives:

(0.45 / 0.55)¹⁰ = 13.443063%.

Starting 20 units away gives 1.807160%; 40 units gives 0.032658%. The arithmetic is reproducible, but the result belongs only to the declared toy model. Changing to a finite horizon or upper target changes it.

This example also disproves the baseline slogan that doubling risk “roughly quadruples” ruin probability. For a fixed absolute barrier in this toy model, doubling the stake halves the number of units: moving from 20 units to 10 raises the eventual probability from about 1.81% to 13.44%, not by a universal factor of four. A different edge or barrier gives another multiplier.

Why the Popular Trading Shortcut Fails

The expression ((1 − A) / (1 + A))^N can be obtained from the equal-step model when A = p − q, because q/p = (1 − A)/(1 + A). The substitution does not turn it into a general formula for any trading expectancy.

The baseline defined A from win rate and average winner/loser, then inserted it into an equal-step ruin formula. That mixes two models. Two systems can share the same win rate and average R yet have different ruin risk because their payoff dispersion, tail losses, order of returns, and sizing mechanics differ.

Expected net R is still useful:

E[R] = p × average win R − (1 − p) × average loss R − average cost R.

But expected R is a first moment, not a complete path model. It does not say how often losses cluster, whether one loss can exceed the nominal stop, or whether an account rule resets intraday. Use the risk-per-trade guide to define the loss budget before modeling its path.

Fixed Stake and Fixed Fraction Are Different Processes

Fixed Stake

If every result adds or subtracts the same monetary unit, the number of units between equity and a fixed barrier is stable. That is the natural setting for classical gambler's ruin. It may approximate a strategy that keeps nominal risk constant over the modeled interval.

Fixed Fraction of Current Equity

If a fraction f of current equity is risked, the path is multiplicative. For an even-payoff sequence with W wins and L losses, wealth becomes V₀(1 + f)^W(1 − f)^L. The loss amount shrinks after losses and grows after gains.

With 0 < f < 1 and no loss beyond the modeled amount, equity approaches but does not literally reach zero in a finite number of losses. It can still cross a practical drawdown, liquidation, or program-failure boundary. “Zero probability of mathematical bankruptcy” therefore does not mean “safe account.”

Kelly's 1956 result concerns maximizing long-run expected logarithmic growth under a specified repeated-bet model. It is not a promise of tolerable drawdowns. Later risk-constrained Kelly research explicitly adds a drawdown-probability constraint and evaluates it with bounds and Monte Carlo simulation. Growth optimization and survival constraints are related, not interchangeable.

Build a Trading Risk-of-Ruin Model

1. Freeze the population. One strategy version, instrument/route scope, session policy, and cost treatment.
2. Define the equity update. Fixed amount, fixed fraction of current equity, volatility target, or another fully specified sizing rule.
3. Define the stopping event. Static floor, peak-to-trough drawdown, daily limit, trailing threshold, margin event, or exact account-rule breach.
4. Define the horizon. A program phase, a year, or a fixed number of eligible trades. Do not report eventual probability when the decision is finite.
5. Choose a return generator. A justified parametric distribution, resampled empirical trades, or regime-conditioned/block-resampled sequences.
6. Simulate the full path. Update equity and any trailing/daily state after every observation; stop the path when the boundary is crossed.
7. Report uncertainty. State path count, random seed, assumptions, estimate interval, sensitivity grid, and missing tail risks.

A plain shuffle assumes trades are exchangeable and removes sequence dependence. If results cluster by regime, setup, day, or execution route, use regime-conditioned simulation or resample contiguous blocks. Neither approach creates crashes, slippage, or rule behavior absent from the source data; stress them separately.

Minimal Monte Carlo Contract

For each simulated path, begin with normalized equity of 1. Apply the frozen sizing rule to a generated net-R sequence. Track current equity, running peak, daily state where relevant, and the exact barrier. A path is a ruin event only when the declared stopping condition is met. The estimate is ruined paths divided by all paths.

Run the model across a grid of risk fractions and harsher assumptions: lower win frequency, smaller winners, larger costs, worse tail loss, longer loss blocks, and altered regime mix. A single point estimate is fragile; a surface shows where the decision breaks. The strategy stress-testing guide covers parameter perturbation, cost shocks, regime splits, and holdout discipline.

Sample Uncertainty Is Model Risk

There is no universal trade count at which win rate, payoff distribution, or ruin probability becomes “reliable.” One hundred independent, identical binary outcomes convey different information from one hundred trades split across several setups and regimes. Average winners can also hide a small number of outliers.

Keep the uncertainty visible:

• show the count of eligible trades and excluded/missing records;
• use an interval for win probability rather than only the point estimate;
• bootstrap or otherwise quantify uncertainty in payoff and drawdown statistics;
• rerun the model with conservative parameters and explicit tail shocks;
• separate development, holdout, and forward samples;
• never choose the risk fraction on the same simulations used to advertise its safety.

If the result changes from negligible to unacceptable under a small plausible perturbation, the correct conclusion is not the prettier estimate. The sizing decision is sensitive and needs more evidence or a larger safety margin.

Drawdown Recovery Is Arithmetic, Not a Ruin Threshold

A decline of fraction d from peak requires a gain of d / (1 − d) on the remaining equity to return to that peak. A 20% decline requires 25%; a one-third decline requires 50%; a 50% decline requires 100%. Those are arithmetic identities.

They do not prove that a 20% drawdown is a mandatory diagnosis point, one-third is a universal pause point, or 50% is empirically terminal. Intervention levels are governance choices that depend on capital, strategy, obligations, risk tolerance, and account contract. The drawdown-recovery analysis separates recovery arithmetic from claims about the probability or time required.

Prop-Program Ruin Is Contract-Specific

A prop evaluation or funded account can fail far above zero equity. The relevant process may include a daily loss rule, static or trailing maximum loss, intraday versus end-of-day calculation, unrealized P/L, balance-high-water marks, consistency conditions, minimum days, resets, and payout effects.

Model the exact firm, program, account size, phase, and current rule version. A fixed “risk per trade” percentage is not enough when several positions overlap or when a trailing threshold moves. The prop-firm drawdown-rules guide shows which contract details must be resolved before simulation. No generic claim such as “use 0.3–0.7%” can be transferred across programs.

Use TSB to Model the Right History

Ownership disclosure: Trader's Second Brain is our product. It does not certify a future ruin probability or replace a declared stochastic model. It helps establish the clean account, strategy, execution, and rule evidence the model depends on.

TSB has processed 600K+ imported trades across the product's import history. That means imported trades—not users and not trades analyzed by Coach. The canonical registry recognizes 330 broker, exchange, platform, and prop-export profiles. Verify the exact route in the supported-source directory, import a representative sample, and reconcile it before estimating any path distribution.

Use account and setup tags to isolate one strategy version; keep fees, partial exits, skipped trades, rule breaches, and missing records visible. Reports and replay expose the actual sequence and drawdown path. Prop Rule Tracker supplies the account-specific boundary when a program failure—not zero equity—is the event being modeled.

TSB Coach is the high-leverage intelligence layer across that selected evidence. It can connect the exact account, setup tags, net outcomes, drawdown sequence, Reports, replay, prop-rule state, detected leaks, and saved focus into a traceable diagnosis. It can reveal that one setup or clustered period drives the tail and help define the next stress test. If the import lacks a fee, rule version, or complete trade population, Coach leaves the affected conclusion Unknown rather than disguising missing evidence as mathematical certainty.

Common Risk-of-Ruin Mistakes

• No explicit barrier. “Ruin” is meaningless until the stopping event is defined.
• Eventual/finite confusion. Infinite-horizon formulas cannot answer a 30-day or evaluation-phase question unchanged.
• Fixed-stake/fixed-fraction confusion. One has equal additive steps; the other compounds multiplicatively.
• Mean-only modeling. Win rate and average R do not preserve tails or order.
• Independent-trade assumption by accident. Shuffling can erase the loss clusters that matter most.
• Gross rather than net returns. Fees, spread, slippage, and losses beyond the planned stop change the path.
• Point-estimate sizing. Optimizing against uncertain inputs creates false precision.
• Calling Kelly a safety rule. Long-run log-growth optimization can allow drawdowns the trader or account cannot tolerate.

Methodology Note

The classical equal-step formulas were checked on September 9, 2026 against Stanford and Yale probability materials and MIT OpenCourseWare's gambler's ruin notes. Kelly's objective and repeated-bet setting were checked against J. L. Kelly Jr.'s 1956 paper, A New Interpretation of Information Rate. The drawdown constraint and Monte Carlo boundary were checked against Busseti, Ryu, and Boyd's peer-reviewed Risk-Constrained Kelly Gambling.

The toy probabilities are direct evaluations of (0.45/0.55)ⁱ at i = 10, 20, 40. They are illustrations, not trading estimates. The baseline sensitivity tables and trader-tier recommendations were removed because no reproducible model or primary evidence supported them. Product statements were checked against local canonical import truth, supported-source, Reports, replay, Prop Rule Tracker, and Coach evidence contracts. See our editorial methodology.

Final Verdict: Survival Starts With the Right Model

Risk-of-ruin math is powerful when its event is precise. It forces the trader to name the failure boundary, time horizon, sizing rule, payoff process, costs, and dependence assumptions. That is a better foundation than treating one percentage or one table as a property of the trader.

Use the classical formula to understand a classical random walk. Use a fully specified simulation for unequal, path-dependent trading returns and account rules. Stress uncertain inputs, preserve the complete sequence, and choose sizing against the unfavorable plausible region—not the most flattering point estimate. Survival first remains the right principle; credible survival requires the right model.