Position sizing is the least interesting decision in trading and the one that decides whether you are still trading next year. It is a single division. The difficulty is not the arithmetic — it is that the arithmetic needs two numbers most traders have never written down, and that the answer it returns is frequently smaller than the position they wanted to take.
This guide covers the formula and its inputs, the common sizing rules and what each is robust to, why the stop has to be decided first, what Kelly actually says, how volatility and correlation change the answer, sizing inside a prop-firm evaluation, and the specific ways sizing failures show up in a journal. It is descriptive. Nothing here is a recommendation to risk any particular amount, and no number below is a claim about what works.
What position sizing actually decides
Sizing does not decide whether a trade wins. It decides what a loss costs and what a win pays, and because losses and wins arrive in an order nobody controls, it decides the shape of your equity curve far more than trade selection does. Two traders taking identical trades in identical order can finish a year one up and one broke; the difference is the size.
The useful way to hold this is that a strategy has an expectancy per unit of risk, and sizing sets how many units you have to spend before the expectancy has time to show up. A positive-expectancy strategy sized so that a normal losing streak takes the account below the point where you can keep trading it never gets to be right. That is the whole content of risk of ruin, and it is an arithmetic property of the size, not a psychological weakness of the trader.
It follows that sizing is the one decision you can get right without knowing anything about the market. You cannot verify your edge today. You can verify today that a loss will cost what you intended it to cost.
The formula, and the three numbers it needs
Position size = (dollars you are willing to lose on this trade) ÷ (dollars you lose per unit if the idea is wrong).
The numerator is a decision about your account. The denominator is a decision about the chart — the distance from your entry to the price at which the trade is no longer the trade you meant to take. Divide, round down, and you have a quantity. There is no third step, and no version of this that works without both numbers existing before the position does.
Written out with the three inputs it actually consumes: account equity, the fraction of it you will risk, and the per-unit distance to your stop. A $50,000 account risking 1% is risking $500. If the entry is 100.00 and the stop is 97.00, the per-share risk is $3.00, and $500 ÷ $3.00 = 166.67, so you buy 166 shares. Rounding down rather than up is not fussiness: 167 shares risks $501, and a rule you round up on is a rule you have already started negotiating with.
Notice what the formula does not contain: your conviction, the quality of the setup, how the last trade went, or how much of the account the position ends up occupying. Those belong to other decisions. The moment conviction enters this division, the number it produces stops being comparable across trades, and every aggregate statistic your journal computes afterwards is measuring two things at once.
A worked example you can redo
The following numbers are invented to show the arithmetic. They are not a recommendation, a real trade, or a claim about any strategy.
Setup. Account $25,000. Risk per trade 1%, so $250. Instrument ABC, entry 40.00, invalidation 38.50, target 44.00.
Size. Per-share risk is 40.00 − 38.50 = $1.50. Position size is 250 ÷ 1.50 = 166.67, rounded down to 166 shares. Actual dollar risk is 166 × 1.50 = $249. Notional is 166 × 40.00 = $6,640, which is 26.6% of the account — a number worth looking at even though it played no part in the sizing, because a 27% notional position on a gapping instrument can lose more than the stop says.
Reward, for the same trade. Per-share reward is 44.00 − 40.00 = $4.00, so the planned risk-reward ratio is 4.00 ÷ 1.50 = 2.67:1, or a 2.67R target.
Now move the stop and watch the size move with it. Same account, same $250, but the invalidation is at 39.25 instead: per-share risk is $0.75, size is 333 shares, notional is $13,320 — 53% of the account for the same $250 of intended risk. The dollar risk did not change. The exposure to anything that jumps the stop doubled. This is the single most useful thing the formula teaches, and it is invisible if you size by picking a share count that feels right.
And the direction people usually get backwards. A wider stop does not mean more risk. It means a smaller position at the same risk. Traders who size first and place the stop afterwards experience it the other way round, which is why their worst losses cluster on the trades where they were most sure.
Fixed fractional, fixed dollar, and Kelly
Three sizing rules cover nearly everything retail traders use, and they differ mainly in what they are robust to.
Fixed fractional risks a constant percentage of current equity per trade. It compounds when you are winning and de-sizes automatically when you are losing, which is the behaviour most people want without having to be disciplined about it in the moment. Its cost is that it never quite lets you get back to even at the old rate after a drawdown — risking 1% of a shrunken account is a smaller dollar bet, by design.
Fixed dollar risks the same dollar amount per trade regardless of equity. It is easier to execute, it makes your R-multiples directly comparable in dollars, and it is what most traders actually do in their first year. Its cost is that it does not scale in either direction: it under-risks a growing account and over-risks a shrinking one, and the second half of that sentence is the dangerous one.
Kelly sizes according to your measured edge. The Kelly fraction for a strategy with win rate w and average winner b (in R) is (w × b − (1 − w)) ÷ b. For a 60% win rate with 1.5R winners that is (0.6 × 1.5 − 0.4) ÷ 1.5 ≈ 33% of the account per trade, which is the correct output of the formula and an obviously unusable position size. Kelly assumes the edge is known exactly. Yours is an estimate from a small sample, and Kelly is brutally sensitive to that estimate being high.
Which is why fractional Kelly exists: half-Kelly cuts drawdown by roughly half while giving up about a quarter of the expected growth, and quarter-Kelly is more conservative still. Even so, the practical reason most journals use a flat 1% or 2% instead is not that Kelly is wrong. It is that a fixed fraction is robust to your edge estimate being wrong, and Kelly is not.
Why the stop has to be decided first
Every sizing rule above needs a denominator, and the denominator is the stop. This has an order-of-operations consequence that is easy to state and hard to keep: the invalidation level is chosen from the chart, the risk is chosen from the account, and the size is what falls out. Any other order produces a stop placed where the size you already wanted happens to be survivable, which is a stop that has nothing to do with the trade being wrong.
The tell is a stop that moves when the position is on. If the level was chosen because it is where the idea fails, a losing trade reaching it is information. If it was chosen because it was as much as you could afford at that size, reaching it is a cash-flow problem, and traders solve cash-flow problems by widening the stop. That is the mechanism, and it is why undisciplined sizing and un-honoured stops show up in journals as the same finding rather than two.
A related failure that the arithmetic makes obvious: sizing off a stop you have no intention of using. A mental stop does not bound the denominator, so the position size computed from it is a number describing a trade you are not actually in.
Your inputs are estimates, and the sample is smaller than you think
Fixed-fractional sizing needs one input you control. Kelly needs two you have to measure — your win rate and your average winner — and measuring them well takes more history than most traders have.
The confidence interval on a 60% win rate computed from 20 trades runs roughly 40% to 78%. Those same 20 trades are consistent with a losing strategy and with an excellent one, and Kelly evaluated at the top of that range prescribes a size that would ruin an account operating at the bottom of it. A win-rate estimate off by five percentage points, or an average-R estimate off by 0.2, moves the prescribed Kelly size by a factor of two or more.
The practical reading is not "use Kelly once you have enough trades." It is that sizing rules requiring an estimate of your edge inherit the uncertainty of that estimate, and the direction of the error that hurts is always the optimistic one. Rules that require no estimate — a flat fraction, a flat dollar — are wrong in a boring way that does not compound.
Volatility: the same 1% is a different trade in different instruments
The formula normalises dollar risk, not exposure. Applied across instruments that move differently, an identical 1% rule produces positions with very different odds of hitting the stop on noise alone.
A stop placed 1.5% away on an instrument whose average daily range is 1% is a stop inside the ordinary daily noise: the position is small, the dollar risk is exactly what you intended, and the probability of being stopped out by nothing in particular is high. The same 1.5% stop on an instrument whose average range is 0.3% is a wide stop, a small position, and a very different trade. Neither is wrong; they are simply not the same bet, and a journal that groups them together is averaging two distributions.
This is what volatility-based stop placement — a multiple of average true range, say — is for. It makes the denominator scale with the instrument so that a fixed fraction produces comparable trades rather than merely comparable dollar losses. If you do not do that, the honest fallback is to group by instrument in your review rather than to assume the R-multiples are directly comparable.
There is one more multiplier to be explicit about: contract size. Options are priced per share and traded per hundred, futures have per-point values that differ by contract, and a sizing formula fed a per-unit risk in the wrong unit produces a position that is off by that factor. The division does not check your units.
Sizing more than one position at a time
Everything above sizes a trade in isolation. Accounts do not hold trades in isolation, and the moment two open positions can lose together, per-trade risk stops being total risk.
Three highly-correlated positions each risking 1% are, on a day when the thing they have in common moves against you, closer to a single 3% position than to three independent bets. The formula has no opinion about this because it never sees the other two. Traders who cap only per-trade risk and never total open risk discover the gap during exactly the sessions that matter.
The common structural answers are all forms of a second ceiling on top of the first: a cap on total risk open at once, a cap on risk within a sector or a theme, or a rule that correlated positions share a single risk budget rather than getting one each. Which ceiling to use is a strategy decision. Having one is what stops a rule that is careful trade by trade from being careless in aggregate. Value-at-Risk is the institutional version of this same question, computed across the whole book instead of per position.
Sizing inside a prop-firm evaluation
A funded-account evaluation replaces "how much am I willing to lose" with a set of numbers somebody else chose, and it changes the sizing problem in a way worth stating plainly: the binding constraint is usually not the profit target, it is the daily loss limit and the drawdown.
Take a published preset — a Topstep 50K Combine carries a $3,000 profit target, a $1,000 daily loss limit and a $2,000 trailing drawdown. Sizing at 1% of the $50,000 nominal account is $500 a trade, which means two full losses in a session ends the day at the limit, and four ends the evaluation. The account size in the name is not the number the sizing rule should be reading. The daily loss limit divided by your intended number of trades is much closer to it.
The trailing part compounds this. A trailing drawdown measured from the account's high-water mark means the room you have shrinks as you profit, so a size that was survivable at the start of the evaluation is a larger fraction of the remaining cushion later. Sizing that is fixed against the nominal balance gets progressively more aggressive against the constraint that actually fails you, without the trader changing anything.
The useful unit here is not percentages but losses: how many average losing trades fit inside each remaining cushion. That number is directly comparable across the daily limit, the drawdown, and the target, and it answers the only question the rules ask.
The daily loss limit is a sizing rule
A daily loss limit is usually filed under psychology, and it belongs here as well: it is a ceiling on cumulative risk within a session, which makes it the second half of the per-trade rule. Per-trade sizing bounds one loss. The daily limit bounds the sequence.
The two have to be set together or one of them is decoration. If the per-trade risk is $500 and the daily limit is $2,000, the limit permits four full losses, and you have implicitly decided that four consecutive losing trades is a day you are willing to have. If the limit were $600 instead, the rule stops you after the first loss and a bit, which is a very different trading day than the sizing rule assumed. Neither is wrong; the failure is not knowing which one you chose.
The reason the limit does work as a sizing control is that the alternative is sizing under conditions where sizing is least reliable. Position size is the input that traders adjust after losses — the tilt escalation — and a limit that ends the session removes the opportunity rather than relying on the judgement that is already impaired.
How sizing failures look in a journal
Sizing is one of the few disciplines that is fully visible in a well-kept journal, because the intended size is a number recorded before the outcome was known and the actual size is a number imported afterwards. Four patterns show up repeatedly.
The loss that was bigger than the plan allowed. Any trade closing beyond its committed maximum risk means either the size was larger than the plan or the stop was not honoured. Both are recorded, and separating them is a matter of looking at the fill.
Size that rises after losses. Plot position size against the outcome of the previous trade. If the sizes after losing trades are systematically larger, that is tilt with a number attached, and it is invisible from inside any individual trade.
Size that rises after wins. The same plot in the other direction is overconfidence. It is more expensive than it looks, because the largest positions land at the end of winning streaks — which is precisely where the strategy is most likely to be at the top of its ordinary variance.
Size that varies with conviction. If your A-grade setups are also your largest positions, then your per-setup statistics are measuring the setup and the sizing together and cannot be untangled afterwards. This one is not necessarily an error — deliberate conviction-weighted sizing is a real strategy choice — but it has to be a choice, because if it is unconscious it invalidates the comparison you were trying to make.
How TradeFlow Quantum handles it
Stated plainly rather than as a pitch, for completeness about the tool this page is published by.
There is a Position Sizer in the app's tools section: enter account size, risk percent, entry and stop, and it returns the share count, the actual dollar risk, the notional and what percentage of the account that notional represents — and it flags a position whose notional exceeds a quarter of the account, since that concentration is worth being deliberate about even when the dollar risk is correct. A risk-reward calculator and a structural plan checker sit beside it. All three are pure arithmetic on numbers you type: they do not read your positions, do not connect to a broker, and do not tell you what to trade.
The part that is wired into the journal is the commitment. A pre-trade plan requires a side, a thesis of at least thirty characters, an invalidation price, and a maximum risk in R — the last of which is capped at 10R, on the grounds that a plan needing more than that is a sizing decision rather than a risk parameter. After the trade closes, the drift detector compares the realized R-multiple against that commitment and raises a maxriskexceeded finding when the loss ran past it, with the overage stated in R and a severity that scales with it. The post-trade review then asks, as one of its five process questions, whether you sized correctly — a self-reported yes/no/n-a that aggregates over time.
On the account side there is a soft daily-loss-limit setting: crossing it raises a banner, and it is deliberately not a block, because the journal does not sit between you and your broker. Prop-firm evaluations are modelled with the published presets for Topstep, Apex, FTMO and FundedNext, and the remaining room under each rule is reported in units of your own average losing trade. That last number carries a caveat that belongs in this guide too: it is computed from closed trades, so open positions are not counted.
Two limits worth naming. The sizer is a calculator, not an enforcement layer — nothing in the product prevents an oversized position, it only records that one happened. And forex pip and lot arithmetic is explicitly out of scope, so FX trades import and analyse in account currency rather than in pips. TradeFlow Quantum is $15/month or $150/year with a 7-day free trial that takes a card and charges nothing until day 7.
What position sizing will not do for you
It will not create an edge. Correct sizing applied to a negative-expectancy strategy produces a slower, more orderly loss — which is genuinely better than a fast disorderly one, and is not the same as a profit. Sizing changes the variance and the survival odds; the sign of the expectancy comes from somewhere else entirely.
It will not protect the number you calculated. The formula bounds your loss at the stop if the stop fills at the stop. Gaps, halts, and thin books do not respect it, which is why the notional matters even though it plays no part in the division, and why sizing off an instrument's ordinary range says nothing about its extraordinary days.
And it will not survive being applied selectively. A 1% rule followed on nineteen trades and abandoned on the twentieth is not a 1% rule; the twentieth is where the damage is, and it is always the trade that felt different. The rule's entire value is that it was decided when nothing was at stake, which is the one condition that cannot be recreated once the position is open.