Risk of ruin: the position-sizing math that decides whether your edge survives
Position size, not setup quality, is what ends most accounts. The drawdown and recovery arithmetic, and how to measure your own risk of ruin from your trades.
Published by TradeFlow Quantum. Filed under risk, position-sizing, drawdown.
Most of the effort in retail trading goes into the entry — which setup, which level, which confirmation. Almost none of it goes into the number that actually determines whether the account is still there in two years, which is how much of it you put behind each of those entries.
This is not a motivational point about discipline. It is arithmetic, and the arithmetic is unusually hostile: losses compound against you at a rate that gains do not compound back. A trader with a genuine edge and the wrong position size loses money. A trader with a mediocre edge and the right position size survives long enough to find a better one.
The recovery asymmetry is the whole problem
To recover from a drawdown of d, you need a gain of d / (1 − d) on the reduced balance. That formula is why drawdown depth matters more than drawdown frequency:
- Down 10% — you need +11.1% to get back to flat.
- Down 20% — you need +25.0%.
- Down 30% — you need +42.9%.
- Down 50% — you need +100.0%. Half the account is gone and you have to double what's left.
Nothing in that table is a trading opinion; it is division. But it reframes what a position size is for. Sizing is not about maximising the good outcome — it is about keeping the bad outcome inside a range you can climb out of with the same edge you already have, rather than needing a better one.
Losing streaks are ordinary, not exceptional
The intuition that breaks accounts is that a long losing run means something has gone wrong — the edge has died, the market has changed, the setup no longer works. Usually it means nothing at all. Five losses in a row at a 50% win rate has a probability of 3.1% in any given window. At a 40% win rate it is 7.8%. Over a few hundred trades, windows are not scarce, and a run like that stops being a possibility and becomes something you should expect to see more than once.
So the useful question is not "how likely is a bad streak?" but "what does a bad streak cost me at my current size?" Risking a fixed fraction r per trade, n consecutive losses leaves you at (1 − r)^n:
- 1% per trade — five losses is a 4.9% drawdown, ten is 9.6%, twenty is 18.2%.
- 2% per trade — five is 9.6%, ten is 18.3%, twenty is 33.2%.
- 5% per trade — five is 22.6%, ten is 40.1%, twenty is 64.2%, which needs a 179% gain to undo.
The 5% row is worth sitting with, because 5% does not feel reckless in the moment. It feels like conviction. It is the number at which an entirely normal run of bad luck — the kind that arrives without anything being wrong — puts the account somewhere it mathematically cannot return from at the same rate it left.
Size to the stop, not to the dollar you're comfortable losing
The mechanical version of all of this is one line: decide the risk per trade as a fraction of the account, find the distance from entry to the price that invalidates your thesis, and let those two numbers determine the share or contract count. Position size is an output, never an input.
The common inversion is to pick a size first — a round number of shares, or the amount that produces a satisfying dollar swing — and then place the stop wherever it happens to leave room. That produces a stop set by comfort rather than by the chart, which is why it gets moved: it was never anchored to anything that could be wrong in the first place.
This is also the practical argument for thinking in R rather than dollars. If every trade risks 1R and your stops are placed where the thesis fails, then a sequence like +2R, −1R, +1R, −1R is legible as a result. The same sequence in dollars, across positions of different sizes, tells you almost nothing about whether the decisions were good. Tracking R-multiples is the prerequisite for every other number in this post.
Check the fraction against your own history, not a rule of thumb
"Risk 1%" and "risk 2%" are conventions, not findings. The version with information in it uses your own trades: take your realised per-trade results, find the worst losing streak you have actually had, and compute what that streak would cost at the size you are using now. If the answer is a drawdown you would not have traded through calmly, the size is too large — and you know that empirically rather than as a maxim.
The more complete version resamples your own trade results many times over to see the range of paths your edge can produce, rather than the single path it happened to produce. That is what a bootstrap simulation does: it treats your closed trades as the distribution and asks what a few hundred more trades drawn from it would look like. The output that matters is not the median — it is the fifth percentile, because that is the outcome you have to be solvent through.
Three ways sizing goes wrong
- Sizing up after a loss. The size increase arrives exactly when the streak is running, which is the worst possible correlation between conviction and exposure. It is also easy to catch after the fact — the timestamps and the size column tell on you.
- Sizing up after a win. Less discussed and nearly as expensive. Euphoria produces the same deviation from plan as tilt, and the trade after a large win is where a lot of accounts give back the week.
- Never recalculating. A fixed fraction of a $10,000 account is a different number after the account is $14,000 or $6,000. Traders who set a dollar risk once and never revisit it are running an accidental Martingale on the way down — a constant dollar risk is a rising percentage of a shrinking balance.
How TradeFlow Quantum handles this
Win rate and expectancy are read from each closed trade's P&L. Average R and the R-multiple rows of the cohort view read only the trades that carry an R, and trades from a broker connection or a CSV arrive with no R: an R reaches a trade only from a restored JSON backup that carries it, or on the sample trades, and this section said until November 10, 2026 that every closed trade carries one. The risk page adds Sharpe and Sortino (both per trade, not annualised), max drawdown in dollars, recovery factor, payoff ratio, and value-at-risk and conditional value-at-risk at 95% over your own realised results. Calmar, ulcer index and pain index are implemented in the codebase but are deliberately not on that page: all three divide by a drawdown percentage measured against peak cumulative P&L, which without a starting balance produced figures like -1101% on real data, so the page was made dollar-only in August 2026.
The simulation described above is the Monte Carlo panel on /analytics/risk. It bootstraps from your closed trades — it needs at least 30 of them before it will run — and reports the median and mean final equity, the probability of ending profitable, the fifth and ninety-fifth percentile paths, and a probability of ruin measured against a starting equity you enter yourself. Horizon, number of simulations and a random seed are all adjustable, so a run is reproducible.
Two limits worth stating plainly rather than leaving you to find them. First, the max-drawdown figure is measured against peak cumulative P&L, not against account equity — the app does not know your account balance, so a 20% drawdown there is 20% of your realised profit peak, not 20% of your capital. That is why the Monte Carlo asks you for a starting equity and the drawdown stat does not. Second, the app does not tell you what to risk. A Kelly-fraction calculator used to live in this codebase and was deliberately removed, because a function whose output is "the fraction of your bankroll you should risk" is a position-sizing recommendation, and this is a journal rather than an advisor. The maths is one line in a spreadsheet if you want it.
Honest disqualifier
If you are trading a funded or prop account, the binding constraint is not risk of ruin in the classical sense — it is the firm's daily loss limit and drawdown rule, which can end the account long before the maths above would. The sizing logic still applies, but the number you are managing to is theirs, not yours, and you should be reading it from the evaluation rules rather than from a percentage-of-account convention. And if you have fewer than about 30 closed trades, none of the measurement in this post will say anything yet — log the trades first; the statistics only become real with a sample behind them.
Related posts
Not financial advice. This post reflects the author’s opinion based on publicly-available information at the time of writing. Mention of third-party products is not an endorsement; product features and prices change over time. Past performance does not guarantee future results.