Risk of Ruin and Money Management in Trading

Risk of ruin is the probability that an account falls below a floor from which it cannot recover — zero, a broker minimum, or a drawdown deep enough that you stop. It is the binding constraint on every strategy, because a positive expectancy is worth nothing if bet size lets a losing streak end the game first. The non-obvious result, which this article derives, is that risk of ruin falls exponentially in bankroll units and is dominated by bet size rather than edge. We cover the gambler's-ruin and continuous-diffusion formulas, the distinction between finite-horizon and eventual ruin, a Monte Carlo that uses your real return distribution, and the estimation-error caveats that make backtested ruin estimates optimistic.

Defining ruin precisely

Risk of ruin (RoR) is a probability between 0 and 1, and the definition of "ruin" materially changes the number:

Absolute ruin: equity reaches zero or a hard floor (broker minimum, margin call).
Practical ruin: equity falls below a threshold where you can no longer trade

meaningfully or would psychologically quit — typically a deep drawdown such as -50%.

Finite-horizon vs. eventual: the probability of ruin within N trades is always

lower than the probability of ruin eventually. Many textbook formulas give the eventual (infinite-horizon) probability, which can be alarmingly higher than the finite-horizon figure that matters for a real trading career.

RoR depends on three quantities acting together: how often you win, how much you win versus lose, and how large you bet relative to capital. The third dominates.

The discrete math: gambler's ruin

For even-money bets (payoff ratio R = 1) with win probability W, define the edge edge = 2W - 1. With a bankroll of N betting units, the classic gambler's-ruin probability of eventual ruin is:

RoR = ((1 - edge) / (1 + edge)) ^ N

The structure is the key lesson: edge enters the base, but bankroll units N enter the exponent. Doubling your edge helps roughly linearly; halving your bet size (doubling N) drives ruin down exponentially.

Bankroll units (N)	Risk of ruin, W = 55% (edge = 0.10)
5	~66%
10	~45%
20	~20%
40	~4%
100	~0.006%

Same edge, wildly different survival, purely from relative bet size. This is the mathematical core of money management.

The continuous approximation and its link to drawdown

For a strategy whose log-wealth follows an approximate diffusion with drift m (the mean of log returns) and volatility s, the probability that equity ever falls to a fraction a of its starting value (with a < 1 and positive drift) is:

P(ever reach fraction a) ≈ a ^ (2*m / s^2)

This single formula unifies risk of ruin with the drawdown distribution and the Kelly results: the exponent 2m/s^2 is governed by the Sharpe-like ratio of the strategy, and increasing leverage raises s faster than m, shrinking the exponent and raising ruin probability. It also shows why betting beyond full Kelly is fatal: there m turns negative, the exponent flips sign, and ruin becomes certain rather than merely possible.

With a payoff ratio

Real trading has asymmetric wins and losses. Per-trade expectancy is:

expectancy = W * avg_win - (1 - W) * avg_loss

A strategy is only worth trading if expectancy is positive — but positive expectancy does not bound risk of ruin away from danger. Over enough trades, a losing streak long enough to exhaust an over-sized account is statistically guaranteed; the only question is whether your bet size survives the streaks that will occur.

Simulating risk of ruin honestly in Python

Closed forms assume constant bet size and fixed payoffs. Real strategies have variable outcomes and fat tails, so Monte Carlo on your actual return distribution is the honest tool. The function below reports both finite-horizon and (approximate) eventual ruin, and accepts an empirical return array so you can feed real trade results including their tails.

import numpy as np

def risk_of_ruin(returns_sample=None, win_rate=0.55, payoff=1.5,
                 bet_fraction=0.02, n_trades=500, n_sims=20_000,
                 ruin_level=0.5, seed=0):
    """Fraction of simulated paths whose equity falls below ruin_level.

    If returns_sample is given (per-trade fractional P&L on risked capital),
    it is bootstrapped to preserve the real distribution and fat tails;
    otherwise a binary win/loss model is used.
    """
    rng = np.random.default_rng(seed)
    ruined = 0
    for _ in range(n_sims):
        equity = 1.0
        for _ in range(n_trades):
            if returns_sample is not None:
                pnl = rng.choice(returns_sample)        # empirical outcome
                equity *= (1 + bet_fraction * pnl)
            else:
                risk = equity * bet_fraction
                if rng.random() < win_rate:
                    equity += risk * payoff
                else:
                    equity -= risk
            if equity <= ruin_level:
                ruined += 1
                break
    return ruined / n_sims

for f in (0.02, 0.05, 0.10, 0.20):
    print(f"bet={f:.0%}  RoR≈{risk_of_ruin(bet_fraction=f):.3f}")

Bet fraction	Approx. risk of ruin (-50% floor)
2%	near 0%
5%	low single digits
10%	meaningful
20%	severe

Same edge, same win rate — only bet size changes, and survival collapses as you size up. Feed in returns_sample from real trades and the fat tails will push every figure higher than the clean binary model suggests.

Why backtested risk of ruin is optimistic

The estimates above are only as good as the inputs, and the inputs are estimated:

Win rate and payoff are noisy. A backtested W = 0.55 has a standard error of

about sqrt(0.55*0.45/n); over 200 trades that is ~3.5%, so the true W could be 0.51, which changes ruin dramatically. Stress the inputs to a pessimistic case before sizing.

The worst streak is not yet sampled. Your historical maximum losing streak is one

draw; the live worst streak will eventually exceed it. Ruin is driven by exactly these unsampled tails.

Correlation hides exposure. Ten "independent" positions that crash together are

one big bet; their joint ruin probability is far higher than per-trade math implies. See correlation in trading.

Costs erode expectancy. Slippage and fees

reduce W and payoff and quietly raise ruin.

Relationship to Kelly

The Kelly criterion gives the growth-optimal fraction for a payoff ratio R:

f* = W - (1 - W) / R

Full Kelly maximizes growth but produces stomach-churning drawdowns and meaningful practical-ruin risk; betting beyond full Kelly guarantees eventual ruin even with a real edge, because log-drift goes negative. Fractional Kelly (half or quarter) trades a little growth for a superlinear reduction in deep-drawdown and ruin probability, which is why nearly all serious practitioners use it. Risk of ruin and drawdown are two views of one danger: drawdown describes how deep the holes are; risk of ruin is the probability one hole is deep enough to end the game.

Finite-horizon ruin and the danger of streaks

The eventual-ruin formulas above assume infinite play, but a real trader operates over a finite number of trades, and the finite-horizon probability behaves differently in an instructive way. Over a fixed N trades, ruin is driven almost entirely by the worst run of consecutive losses the horizon is likely to contain. The expected length of the longest losing streak in N independent trades with loss probability q grows logarithmically, roughly log(N) / log(1/q). For a 45% loss rate (W = 0.55) over 500 trades, that is a longest expected streak of about 8 consecutive losses — and the tail of the streak distribution routinely delivers 12 or more. If 12 consecutive losses at your bet fraction breach your floor, your finite-horizon risk of ruin is high regardless of what the per-trade expectancy looks like. The design rule that falls out of this: choose a bet fraction such that your floor survives a losing streak two to three times longer than the expected longest streak for your horizon. This streak-based sizing is more intuitive and more robust to fat tails than plugging point estimates into a closed-form ruin formula.

Reducing risk of ruin

Ruin is almost entirely within your control, in rough order of leverage:

Bet smaller. The dominant lever — exponential, not linear. Risking 1-2% instead

of 10% per trade moves ruin from severe to negligible at the same edge.

Cap aggregate correlated risk. Limit total exposure across positions, not just

per-trade size.

Size to volatility. Scale notional down as volatility rises so a fixed risk,

not a fixed notional, is taken per trade — see volatility targeting.

Use a pre-committed kill switch. Decide the drawdown at which you cut size or

stop, and automate it.

Improve the edge. Higher W or payoff helps, but usually far less than reducing

size.

Separate trading capital from life capital so account ruin is never personal

ruin, preserving your ability to return.

A concrete example

A trader with a genuine edge — 52% win rate, payoff ratio 1.3, positive expectancy — starts with \$50,000.

Aggressive (15% per trade): a plausible six-loss streak halves equity, and Monte

Carlo shows double-digit practical-ruin probability. The edge is real; the sizing is suicidal.

Conservative (1.5% per trade): the same streak costs under 10%, easily

survivable, and simulated ruin is effectively zero.

Identical strategy and edge; the only difference is bet size, and it decides survival. This is why professionals obsess over position sizing far more than entry signals.

Conclusion

Risk of ruin is the probability you do not survive to collect your edge, and it is governed less by strategy quality than by bet size — falling exponentially in bankroll units and rising fast with leverage. The continuous formula ties it directly to drawdown and Kelly: ruin probability is set by the exponent 2m/s^2, and over-betting flips that exponent toward certain ruin. Simulate it with your real, fat-tailed return distribution rather than a clean formula, stress the inputs to a pessimistic case because backtested estimates are optimistic, trade fractional Kelly, control drawdown, and respect correlation. The market rewards the traders still in the game, and staying in the game is exactly what managing risk of ruin is about.