What does a round turn cost, in the unit an evaluation is scored in?
An evaluation is scored in units of risk. The target is a multiple of the money you put behind one trade, the allowance is another multiple of it, and every rule in the document is denominated the same way. Execution, on the other hand, arrives in currency: a commission, an exchange fee, a clearing fee, the spread you crossed, the tick you did not get.
Converting one into the other is a single division, and it is the only piece of arithmetic in this article that a reader can check against a document they already have. A round turn costing $12.50 on a trade risking $250 is 0.050R. Nothing about the account size, the firm, the instrument or the strategy enters it.
Reproducible by hand; arithmetic, not simulation.
Every profile on this page is a stated hypothetical, not a measurement of anyone. The figures come from figures.data.js, which ships beside this page and prints what it emits when run directly. No seed is disclosed because nothing here is drawn at random: the evaluation is a walk on a finite set of states and its outcome probabilities are enumerated exactly.
That number then does exactly one thing, and it does it independently of every other parameter. Over any run of trades, gross profit accumulates at the gross edge and execution is charged per trade, so the share of gross profit handed to execution is the cost divided by the gross edge — and the target, the allowance, the account size and the number of trades all cancel.
Take the middling profile below. Reaching a target of eight units of risk at a gross edge of 0.150R against a cost of 0.050R takes 80 trades. Those trades earn 12R gross and pay 4R to execution. That is 33.3% of the gross, and it is the ratio of the two numbers and nothing else.
The share of gross profit that execution consumes is the cost of a round turn divided by the gross edge of the trade it buys. No other quantity enters.
How much cost does it take to turn a passing profile into a failing one?
To answer that, the evaluation has to be written down. This one starts at zero, passes on reaching eight units of risk in profit, and fails on giving back six of them measured from where it started. Cost is charged on every round turn, winner or loser, which is the convention this site's own engine already uses. There is no time limit and no daily rule; both are deliberately absent and both are discussed at the end.
Two things follow that are worth separating. The first is obvious: raise the cost far enough and the net edge goes negative, after which the evaluation is lost slowly rather than quickly. The second is not obvious, and it is where the useful answer lives.
Target eight units of risk. Allowance six, measured from the starting balance and not from a peak. Cost on every round turn. Probabilities enumerated exactly over the state space, with the recursion truncated at 2000 trades and anything still running counted as a failure.
A trade wins some multiple of the risk unit and loses one of it, and that multiple follows from the gross edge and the win rate by rearranging the expectancy identity. The identity itself is derived in full on why win rate is the wrong number, and is borrowed here rather than restated.
| 0.032R | gross edge below which the evaluation is a coin flip with free execution |
| 0.084R | the edge the assumed cost puts on the line |
| 0.137R | the edge double that cost puts on the line |
The frontier does not sit at the cost that erases the edge. It sits well below it. With no net edge at all, this evaluation is not a coin flip — it passes 42.5% of the time, because the allowance is nearer than the target and a driftless walk hits the nearer barrier more often. The ratio of the two distances is 42.9%, and the exact enumeration lands on it.
So a trader needs net edge left over merely to be even money on the evaluation, quite apart from being even money on a trade. That gap is what pulls the frontier down: the thin profile crosses at 0.027R against a gross edge of 0.060R, which is under half of it.
The cost that flips an evaluation is strictly less than the cost that erases the edge, because a target further away than the allowance demands surplus edge before it demands anything else.
Read off the frontier and three numbers fall out. Below a gross edge of 0.032R the evaluation is already a coin flip with execution that costs nothing at all, so no broker can rescue it. At the cost this site assumes, the line moves to 0.084R. At twice that cost it moves to 0.137R — the required edge rises by 63.1% on the strength of an assumption about commissions.
Why does identical execution cost one trader sixteen times what it costs another?
Cost in units of risk is a fraction, and almost all of the variation between traders is in its denominator rather than its numerator.
The numerator is what a round turn takes out of the market, and it is naturally expressed in ticks: commissions and exchange fees divided by that instrument's tick value, plus the spread crossed, plus whatever the fill gave up. The denominator is the stop, also in ticks. Cost in R is one divided by the other, and that is the whole of it.
The two ways a trader meets execution — a futures round turn billed per contract, and a spread paid on the way in and out — look like different animals and arrive at the same place. A futures commission is a fixed sum per contract, and the number of contracts a fixed dollar risk buys is inversely proportional to the stop, so the bill per trade falls as the stop widens. A spread is a fixed number of ticks, divided by a stop measured in the same ticks. Different numerators, identical denominator.
No commission, fee schedule or spread is asserted anywhere on this page. The round turn drawn above is a stated quantity in ticks, chosen so the arithmetic is legible, and the identity holds for any other value a reader wants to substitute.
| 2 | ticks per round turn, the same for every bar |
| 40 | ticks of stop that reproduce this site's assumed cost |
| 16× | between the tightest stop drawn and the widest |
Which makes the ordinary assumption legible in a way it usually is not. Charge 2 ticks for a round turn and the frontier of this site's assumed 0.050R is a stop of 40 ticks. It is not a claim about brokers. It is a claim about how far away people put their stops, applied to everybody.
Move to a four-tick stop and the same execution costs 0.500R — 333% of the middling profile's entire gross edge, which is to say the trader is paying more than three times what the trade is worth. Move to 64 ticks and it costs 0.031R, which barely registers.
Execution quality is not what separates a cheap trader from an expensive one. The stop is. The same fill, the same broker and the same instrument span a sixteen-fold range of cost across the stop distances drawn above.
Does trading more often make execution cost matter more?
The intuitive answer is that it must: four trades a day pays four times what one trade a day pays, so cost is a frequency problem and frequency is a choice. Half of that is exactly right and the other half does not survive the arithmetic, and the instrument below is the quickest way to see which half is which.
The instrument marches the surviving distribution forward one trade at a time. Every build drives its real controls in a real browser and requires the readouts to equal model.js, which runs the same process backwards from every state instead. The two never meet in code, and they parted company once — over a single state whose profit was exactly the target, worth four parts in ten thousand.
Your round turn, priced in the unit that decides
Set the edge you believe you have, the money you put at risk on one trade, and what a round turn costs you. The instrument converts the cost into units of risk and finds, by search, the cost at which the same profile becomes a coin flip.
What one trade is worth on average before execution cost, in units of the money you risk.
RSets the shape of the trade. The reward per winner follows from this and the edge.
%Commissions, exchange and clearing fees, the spread crossed, and slippage. Both sides, one trade.
$Move the frequency control and watch the pass probability. It does not move. Move it and watch what is paid per day, and it moves in proportion: at 0.050R a round turn, four trades a day costs 0.200R of risk every day and one trade a day costs 0.050R.
Both readings are correct, and they disagree because they are answers to different questions. An evaluation is not scored per day. It is scored in units of risk, and the number of trades required to accumulate them is set by the net edge of a trade, not by how quickly the trades arrive. Doubling the frequency at an unchanged edge halves the calendar and leaves the odds exactly where they were.
Frequency sets the bill and the calendar. It does not set the odds. What sets the odds is the ratio, and frequency enters that ratio only through what travels with it.
What travels with it is the stop. A trader taking many trades a day is generally reaching for a smaller move against a tighter stop, and the previous section showed that a tighter stop is the entire mechanism by which cost becomes expensive. That is a regularity in how people trade, not a law of arithmetic, and this page does not measure it. The honest statement is that frequency is a symptom of the thing that matters rather than the thing itself, and a reader who trades often with a wide stop is not exposed and a reader who trades rarely with a four-tick stop is.
Does this site's own cost assumption survive being doubled?
It has to be said plainly, and the plain answer is worse than the question. This site does not have one execution-cost assumption. It has two, and neither of them is derived anywhere.
Four of its published pages state a flat cost of 0.050R per trade as part of the profile their figures are run on — why win rate is the wrong number, how much the floor mechanic alone moves pass probability, why a profitable trader still blows up and which rule actually ends evaluations — and the same figure is the shipped default in the application's own parameter registry. Two further pages in the same corpus, both still in draft and so without an address to link to, publish simulated pass probabilities from generators that contain no cost term at all — not a stated zero, an absent one. So the question is not only what happens if the stated cost is wrong by a factor of two. It is also what happens between charging 0.050R and charging nothing, which is a gap this site currently straddles without saying so.
Both counts are counts of files, made on the date at the head of this article, and the file list is in references.md. The stated cost was read from the application's own parameter registry, where it is the shipped default, and from the profile printed alongside the figures on the pages that carry it. The pages that charge nothing were established by reading their generators, which contain no cost term of any kind.
| 10.7 | points of pass probability lost to the assumed cost, at worst |
| 10.5 | further points lost to doubling it, at worst |
| 4.4 | further points lost to doubling it, at best |
The answer depends entirely on whose edge is being modelled, which is the same finding as the frontier arriving by a different road. For the strong profile, doubling the assumption costs 4.4 points of pass probability — real, but it moves nothing that a reader would decide differently. For the thin profile it costs 10.5 points, and more to the point that profile is already under the coin-flip line at the assumed cost, at 45.0%.
The published reference profile on this site is a strong one. Its results are, on this evidence, not sensitive to the stated cost at the doubling scale, and the pages that carry them are not wrong. What is missing is the boundary. Any conclusion this site publishes about a trader whose gross edge is under roughly 0.137R is sensitive to a cost assumption that has never been examined, because that is where doubling it moves a trader across the line rather than along it.
The pages that charge nothing are in a different position, and it is not a better one. Charging nothing puts the frontier at a gross edge of 0.032R; charging what this site says it charges puts it at 0.084R. Those two live assumptions are 0.052R apart in required edge, which is very nearly the 0.053R that the doubling costs. Omitting execution is not a smaller error than getting its size wrong; it is the same size of error, pointed in the direction that flatters the trader.
The stated cost survives for the profiles this site models with it, and it should not be carried to thin ones. The greater exposure is on the pages that state no cost at all, where the omission moves the frontier as far as doubling would.
Underneath both of those sits a problem that a better constant would not fix. A single flat number in units of risk cannot represent execution at all, because the same broker and the same fill produce wildly different numbers depending on the stop. A trader with a four-tick stop and this site's assumed cost is being modelled at a fraction of what execution actually takes from them. The correction is not a better constant; it is to stop treating cost as a constant.
Which account was this cost measured against?
The allowance here is measured from the starting balance and never moves. A floor that follows a high-water mark is a different and harsher object, and it is set out in full on the line that closes a trailing-drawdown account; it is cited here rather than reworked. The direction of the error is worth stating even so, as an inference from that mechanic rather than a measurement made here: a floor that rises with the high-water mark spends survival room on the way up, so a per-trade drag has more to bite on and not less. Modelling a static allowance understates the case this article is making.
There is no daily loss limit here, no minimum trading day, no time limit and no consistency rule. Any of them can end an evaluation on its own, and which one usually does is a question this site answers elsewhere; a per-day cost interacts with a per-day rule in a way this model cannot see. There is one trade shape, no clustering of losses, no scaling in or out, and no variation in position size.
Trailing floors. Daily loss limits. Minimum and maximum trading days. Consistency caps. Position-size variation. Correlated losses. Slippage as a distribution rather than a constant. Any named firm's rules.
Cost itself is treated as a constant, which is the weakest assumption in the article after the ones it is criticising. Commissions and fees genuinely are constant. A fill is not, and the asymmetry is definitional rather than empirical: a stop order becomes a market order the moment the price touches it, so it is filled while the price is moving against the position. That is what a stop is, not a claim about any particular market — but it means the cost of the trades that lose is drawn from a different part of the distribution than the cost of the trades that win, and a single constant cannot hold both.
No commission, fee or spread is asserted here for any broker or instrument, and none was looked up. Everything is expressed in ticks and in units of risk precisely so that no such claim is needed. What that costs the reader is that the article cannot tell anyone what their own number is. It can only tell them that the number is a division they can perform from a statement they already hold, and roughly where on the frontier the answer puts them.
How much of this depends on the cost assumption?
Is 0.050R a reasonable cost assumption?
It is a reasonable assumption for one kind of trader and a bad one for another, and the thing that decides which is the stop rather than the broker. Cost in units of risk is the execution bill in ticks divided by the stop in ticks, so at a round turn of 2 ticks that figure describes a stop of 40 ticks. A trader whose stop is a quarter of that pays four times as much for identical execution. The division takes a few seconds against a real statement and is worth more than any general answer. Free path: /app?src=article-trading-costs-that-decide.
Why is the break-even cost lower than the gross edge?
Because an evaluation is not symmetric. The allowance sits nearer to the starting balance than the target does, so a walk with no drift at all reaches the allowance more often than the target — 42.5% of the time on the evaluation modelled here. A trader therefore needs surplus edge to be even money on the evaluation, and cost eats the surplus before it touches the edge itself.
Does trading less often reduce the cost of execution?
It reduces what is paid per day and lengthens the calendar, and by itself it does not change the probability of passing. An evaluation is scored in units of risk rather than in days, and the trades required to accumulate them depend on the net edge of a trade rather than on how fast the trades arrive. Trading less often usually does help, but through a different channel: fewer, more selective trades tend to be taken with wider stops, and the stop is the denominator that decides what execution costs.
Does this contradict anything else published on this site?
It qualifies it, and it corrects the premise this article started from. The premise was that every simulated page here bakes in a flat cost of 0.050R. Counting the files says otherwise: four pages state that cost and two simulate with no cost term at all. On the evidence here the stated figure is sound for the strong reference profile those pages model, where doubling it costs 4.4 points of pass probability, and it is not sound as a universal default, because for a thin edge the same doubling costs 10.5 points and moves the trader across the coin-flip line rather than along it. The pages that charge nothing carry the larger unstated error of the two.
Every number in the prose and in the figures is emitted by figures.data.js, which ships beside this page and prints what it emits when run directly. Nothing is simulated and no seed is disclosed: the evaluation is a walk over a finite state space and its probabilities are enumerated exactly. Every number the instrument shows is computed in the page and checked on each build against model.js, an independent implementation that runs the recurrence in the opposite direction. Profiles are illustrative and describe no real trader. No firm is named, no commission is asserted, and this page ranks nothing.
- The execution cost examined in the fifth section was read from this product's own parameter registry, where it is the shipped default, and from the profile printed beside the figures on why win rate is the wrong number, how much the floor mechanic alone moves pass probability, why a profitable trader still blows up and which rule actually ends evaluations. Read on the date at the head of this article; the exact files and lines are in references.md.
- Charging the cost on every round turn, winner and loser alike, is the convention the calculator on this site already uses. This article adopts it unchanged, so that its results are comparable with the site's own rather than a different accounting of the same word.
- The expectancy identity that turns a gross edge and a win rate into a reward per winner is derived on why win rate is the wrong number. The trailing floor this article deliberately does not model is set out on the line that closes a trailing-drawdown account. Both are cited rather than reworked.
- No external source is cited, because no external fact is claimed. Commissions, fee schedules, tick values and spreads are properties of particular brokers and instruments; this page states none of them and expresses everything in ticks and in units of risk instead.
Free path: measure commissions and slippage under your own numbers
Frontier at 0.084R gross edge — below it, round-turn cost decides the evaluation; above it, almost nothing. Round turn in R = fee ÷ stop. Open the free simulator with this page's attribution before the first paid document.
Open free path — frontier 0.084R · fee ÷ stop →