Risk of ruin calculator

Define the starting capital, upper target and round limit. Compare reaching zero, reaching the target and remaining between them in a simple fixed-unit model.

Two stopping boundaries. One stated horizon.

Fictional model: each independent round adds or subtracts exactly one unit. Stop permanently at 0 or the target. No fees, leverage or deposits.

A unit is the fixed amount won or lost, not a percentage of remaining capital. “Ruin” here means reaching zero before the upper target.

By round 20

Three mutually exclusive states, including paths that have not stopped.

Reached zero first
15.0209%
Reached target first
40.9682%
Neither boundary yet
44.0109%

Eventually reaching zero before the target

26.8283%

No round limit for this result. The upper target still stops the process. It is not the probability of losing everything while trading indefinitely without an upper stop.

What this calculator measures

Starting with k whole units and an upper target N, every independent round gains one unit with probability p or loses one with probability 1 −p. Reaching 0 or N ends that path permanently. This educational model has no costs, deposits, changing stakes or simultaneous positions.

Finite horizon vs eventual ruin

The first result asks whether zero is reached before the target by the entered round limit. The three states sum to 100% before display rounding. The eventual result removes the round limit, while retaining the upper stopping target. Changing the horizon does not change the eventual formula.

Gambler’s ruin formula

For p ≠0.5, set r =(1 −p)/p. The eventual zero-before-target probability is (r^k −r^N)/(1 −r^N). For p =0.5 it is 1 −k/N. At p =0, eventual ruin is 100%; at p =1 it is zero. The finite-horizon distribution is calculated round by round, keeping both boundaries absorbing.

MIT’s stochastic-processes lecture presents the fair fixed-unit, two-boundary model. Our risk of ruin guide derives the recurrence and explains the assumptions using original examples.

A small example you can verify by hand

Set starting units to 1, target to 3, win probability to 50% and horizon to 2. The model reaches zero with probability 50%, reaches the target with probability 25%, and remains inside with probability 25%. Eventual ruin before target is 66.6667%. The unresolved quarter is not automatically a success or a failure.

Using this for prediction-market research

A 0.80 outcome share paying 1.00 on a win produces a 0.20 gain versus a 0.80 loss before costs. That is not the symmetric payoff used here. Several positions tied to the same event can also violate independence. Do not treat the calculator’s output as the actual probability that a Kalshi, Polymarket or other account will fail.

Use the expected-value calculator to separate a supplied probability from cost and payout. Use the joint-probability calculator when thinking about dependence. Neither tool validates a forecasting edge.

Input and precision limits

Enter 1 to 199 starting units below a target of 2 to 200 units, a probability from 0% to 100%, and 0 to 1,000 rounds. Whole-unit boundaries keep the model explicit. Results use floating-point arithmetic and four-decimal percentage display; extremely small representable probabilities are labeled below 0.0001%. Very extreme probabilities can underflow to zero numerically, which is not evidence of zero real-world risk.

Frequently asked questions

What does this risk of ruin calculator assume?

Independent rounds with fixed win probability, a gain or loss of one unit, no costs, and permanent stops at zero or an upper target. It does not model a live trading account.

Why are finite-horizon and eventual ruin different?

Some paths reach neither boundary by the chosen round limit. Eventual ruin allows those paths to continue, while retaining both stopping boundaries.

Can I use an ordinary win rate for prediction-market trades?

Not without checking the payoff model. Event shares can have unequal gains and losses, changing prices, costs and correlated outcomes; those violate this calculator’s fixed plus-or-minus-one model.