Risk of ruin: how a positive-expectancy trader still blows up
With a positive edge, risk per trade still decides how likely a deep drawdown is. Define ruin as a drawdown you choose, then simulate: the same trades at 0.5% and 3% risk end in very different places.
Start in plain English
Risk of ruin is the probability that a trading account falls to a level at which you stop: a drawdown you cannot or do not want to continue from. This lesson shows why a process with a positive expected value can still reach that level, and why the main lever is something you control: how much you risk on each trade.
The core idea: an edge says what happens on average; risk per trade decides whether you survive the path to the average. Losing streaks are a normal part of any distribution with losses in it. At small risk per trade, a long streak is a bad month. At large risk per trade, the same streak ends the account before the average has time to show up.
"Ruin" here does not need to mean zero. Define it as a drawdown from the account's peak that you choose in advance, for example 30%, because that is where many traders stop, lose confidence in the process, or hit an account rule.
Definition and inputs
Risk of ruin depends on four inputs:
- The outcome distribution per trade, in R after costs (how often you win, lose or scratch, and by how much).
- Risk per trade, as a fraction of the account (fixed-fractional) or a fixed dollar amount.
- The ruin threshold, such as a 30% drawdown from peak.
- The horizon, such as the next 200 trades.
This lesson uses Setup X from Thinking in probabilities: 40% of trades reach +2R, 45% hit −1R, 15% scratch at 0R, with costs of −0.1R per trade. Expectancy after costs: +0.25R per trade. The distribution is Illustrative: chosen to teach.
Two paths from the same start
We simulated 1,000 runs of 200 Setup X trades, risking a fixed fraction of current equity per trade, with ruin defined as a 30% drawdown from the account's peak (computed from the assumed distribution, Illustrative; seed 20261005). Below are two of those runs, picked to show the contrast. Picking which runs to show is itself selection, the bias Thinking in probabilities warned about.
Both runs end with the same total: +51R over 200 trades, which is also the median total across all 1,000 runs.
Path ALarge risk, and it holds
Run 1, at 3% risk per trade. Its longest losing streak is 7 trades. The deepest drawdown, 27.7% from peak, comes at trade 66. The run does not cross the 30% threshold and finishes at +51R.
Path BSame edge, same risk, a longer streak
Run 2, also at 3%. After a peak at trade 48, it loses 10 trades in a row, wins one and loses three more: 14 trades, −12.4R. At trade 62 the account is 31.7% below its peak, past the threshold. By the rule written before trade 1, the run stops here, even though its later trades (had they been taken) would also have finished at +51R.
Path CSame streak, smaller risk
Run 2 again, trade for trade, at 0.5% risk. The same 14-trade streak costs about 6% from peak. The run continues and finishes at +51R; its deepest drawdown across all 200 trades is 6.0%.
Paths B and C contain identical trades. Only risk per trade differs, and it decided whether the streak was an ordinary bad stretch or the end of the account. The rest of the lesson shows how often this happens across all 1,000 runs, what happens when the edge is smaller than you think, and how deep drawdowns are to recover from.
How often: the full simulation
Across 1,000 runs of 200 trades each (computed from the assumed distributions, Illustrative; ruin = 30% drawdown from peak):
| Risk per trade | Setup X (+0.25R): runs ruined | Median max drawdown | Edge +0.10R: runs ruined | Edge 0R: runs ruined |
|---|---|---|---|---|
| 0.5% | 0.0% | 5.1% | 0.0% | 0.1% |
| 1% | 0.1% | 9.9% | 1.9% | 11.4% |
| 2% | 8.0% | 19.2% | 34.4% | 69.5% |
| 3% | 39.7% | 27.7% | 76.8% | 94.4% |
The +0.10R column uses 35% wins at +2R, 50% losses at −1R and 15% scratches; the 0R column uses 33.3% wins, 56.7% losses and 10% scratches; both with −0.1R costs per trade.
Two things stand out. At 3% risk, about four runs in ten hit the threshold even with a +0.25R edge. And as the edge shrinks, ruin rises fast at every risk level above 0.5%.
Analytical intuition: risk enters as an exponent
There is a textbook formula for the simplest case, a bet that wins or loses one unit at fixed size, with win probability p. The probability of ever being U units below the start is ((1 − edge) ÷ (1 + edge))^U, where edge = 2p − 1. At p = 55%:
- With a 10-unit cushion, the probability is about 13.4%.
- With a 20-unit cushion (half the risk per trade), about 1.8%.
- With 40 units, about 0.03%.
Trading is not that simple bet, but the shape carries over. Halving risk per trade doubles the number of losses it takes to reach the threshold, and the probability falls roughly with the square, not by half.
Losing streaks are normal
At a 45% loss rate, the probability of at least one losing streak of a given length somewhere in 200 trades is (computed, Illustrative):
| Streak length | Probability in 200 trades |
|---|---|
| 5 or more | 88% |
| 7 or more | 33% |
| 8 or more | 17% |
| 10 or more | 3.5% |
At 3% risk, 10 straight losses of 1.1R each (−1R plus costs) take 1 − 0.967¹⁰ ≈ 28.5% off the account. At 0.5% they take about 5.4%. A streak of 10 is unusual in any one 200-trade block, but over years of trading it becomes likely to appear at some point.
Drawdown recovery
A drawdown needs a larger percentage gain to recover than the percentage it lost:
| Drawdown | Gain needed to return to the peak |
|---|---|
| 10% | 11.1% |
| 20% | 25.0% |
| 30% | 42.9% |
| 50% | 100.0% |
Deep drawdowns are expensive twice: once in the loss, and again in the number of trades needed to climb back at the same edge.
Observe vs. infer
After 60 trades of a real or practice log (figures Illustrative):
| What the log shows (observed) | What a trader might infer | Why it is not established |
|---|---|---|
| Average +0.20R per trade over 60 trades | "My edge is +0.20R" | A rough 95% range for the true average is wide, about ±0.35R at this spread (How many trades before your numbers mean anything) |
| Largest drawdown so far: 9% at 2% risk | "My drawdowns stay under 10%" | 60 trades rarely contain the longest streak a process can produce |
| Seven wins in the last ten | "I can raise risk now" | A hot stretch is weak evidence about the distribution |
| No losing streak longer than 4 | "Long streaks don't happen to my setup" | At a 45% loss rate, a 7-loss streak in 200 trades has about a one-in-three chance |
Odds box
These are properties of assumed distributions, not measurements of any trader or setup. The table's message does not depend on the exact numbers: the uncertainty in your edge is a reason to size as if the edge might be smaller.
What shifts the odds
These conditions change risk of ruin in the simulation's logic; your own numbers decide by how much:
- Risk per trade. The largest lever in the table.
- The true edge, which you only know as an estimate with a wide interval.
- Fatter loss tail. Halts and gaps that produce −3R or −5R losses (Slippage) push drawdowns deeper than a clean −1R model.
- Correlated trades. Several trades in the same stock on the same morning behave more like one large trade.
- Size changes after losses or wins. Raising size after a hot streak raises risk at exactly the moment the estimate is most flattered.
- The threshold itself. A 20% threshold is reached far more often than a 50% one.
Cost of being wrong
Two mistakes, both measured in the same table:
- Overestimating the edge. Sizing at 2% for an estimated +0.25R edge gives an 8.0% ruin rate in this simulation. If the true edge is 0R, the same sizing gives 69.5% (computed, Illustrative).
- Risk drift. A trader who starts at 1% and drifts to 3% after a good month moves from a 0.1% column to a 39.7% column with the same setup.
At a $20,000 account, 3% risk is $600 per trade and the 30% threshold is a $6,000 drawdown from peak. At 0.5% risk it is $100 per trade, and the median run's deepest drawdown was about 5%, roughly $1,000 (Illustrative).
How many trades before this means anything
Risk of ruin is a calculation on an estimated edge, so it inherits that estimate's uncertainty. With a per-trade spread of about 1.4R, a 60-trade log pins the edge down only to about ±0.35R; at 200 trades, about ±0.19R (normal approximation, Illustrative). Until the interval clearly excludes zero, the 0R column above is a plausible description of your process. How many trades before your numbers mean anything explains the method.
What this lesson does not claim
- It does not recommend a risk percentage. Every distribution and result above is Illustrative or computed from one.
- Simulated runs assume independent trades, fixed rules and no change in behaviour after losses. Real trading often violates all three; in the simulation's logic, correlated trades and size increases after losses make drawdowns deeper.
- A low risk of ruin is not a profitable process. At 0R edge and 0.5% risk, ruin is rare and the account goes nowhere on average after costs.
- Setup X is fictional.
Go deeper
- FennEdge first.
- The Trading Journal records net P&L from real broker-synced trades, which is where your real drawdowns live; viewing it needs Plus or above (as of 2026-10-05). Simulated runs and practice trades belong in your practice log.
- Level Up (Plus and above, as of 2026-10-05) is a private challenge scored on your Journal net P&L, with an optional daily max loss per level. It does not track tags or entry timing, so keep process goals such as "risk per trade fixed for a month" in your own log.
- Fixed-fractional vs. fixed-dollar risk. Fixed-fractional risk shrinks size after losses, which slows a drawdown; fixed-dollar risk keeps size constant, which makes drawdowns deeper in percentage terms as the account falls.
- Growth-optimal sizing. The Kelly criterion (Kelly, 1956) maximises long-run growth for a known edge. With an estimated edge, full Kelly sizing magnifies any overestimate and produces deep drawdowns; using a fraction of it is a common response.
- Monte Carlo on your own log. Resampling your own trades in R, many times, shows the drawdowns your actual distribution can produce. It still cannot show losses that did not appear in the log.
- Related lessons: Expectancy, Position sizing, Daily loss limits, Reviewing scalps.
Exercises
Six exercises follow this lesson: three concept checks, two calculations (the gain needed to recover a drawdown, and the drawdown from a losing streak at fixed-fractional risk), and one scenario about choosing risk per trade when the edge estimate is uncertain. Exercises, answers and progress are available after signing in.
Primary sources
- Feller, W. (1968). An Introduction to Probability Theory and Its Applications, Vol. 1, 3rd edition. Wiley. Chapter XIV, random walk and ruin problems.
- Kelly, J. L. (1956). A new interpretation of information rate. Bell System Technical Journal 35(4).
- FINRA. Frequent Intraday Trading: Understanding the Basics.
Sources checked 2026-10-05. Setup X, the lower-edge distributions and all simulated runs are fictional and Illustrative; probabilities are computed from the assumed distributions (1,000 runs of 200 trades, seed 20261005). This lesson is education, not a recommendation to buy or sell any security. Past patterns do not indicate future results.