A trade copier that handles entries correctly can still handle exits badly. An entry is one decision, rounded once. A scale-out is a sequence of decisions, each rounded against a position the previous rounding already distorted. Those errors do not cancel. They accumulate with a direction, and that direction is almost always toward the follower holding more risk than the master during the exact window the master is taking risk off.
When a master exits in stages, a copier can either tell each follower how much to close or tell it how much it should still be holding. Telling it the target holding is strictly better: the follower's size is recomputed from the master's absolute remaining position at every stage, so rounding errors and rejected orders self-correct instead of compounding. Under close-this-much mirroring with floor rounding, a follower stays over-exposed while the master de-risks, and any follower too small for the master's smallest exit stage to round up to one contract skips that stage entirely.
Why a staged exit breaks a copier that handles entries fine
The copier sees less than the master's platform does. On the master side a staged exit is one plan expressed as several orders, and the platform knows they are coupled. NinjaTrader's ATM engine is explicit about that coupling: the NinjaTrader 8 ATM Strategy help guide states that decreasing one profit target's contract count automatically increases another, and that when scaling into a position all stop loss and profit target orders are automatically updated to reflect the new position size. What reaches the copier is a stream of independent partial fills. The intent behind them, that these legs are one exit plan, is not on the wire. A copier that treats each fill as a standalone close instruction is reconstructing a plan from its shadow.
This is a different question from how much the follower buys in the first place, which is covered in copy trading position sizing methods. It is also different from the ticket-level bookkeeping of which underlying lot a partial close offsets, which depends on account mode and is covered in copying MT5 hedging and netting accounts. Assume size is chosen and tickets are tracked. The only question here is what quantity to send at each exit stage.
The master's exit, expressed exactly
The master opens 10 contracts and exits in three stages: closes 3, then 4, then the final 3. As fractions of the original position that is 30%, 40% and 30%. Remaining after stage 1 is 7 (70% of original), after stage 2 is 3 (30%), after stage 3 is zero. The stages sum to 10.
Every example below uses this master. Two followers copy it: Follower A entered 5 contracts (a 0.5x ratio) and Follower B entered 2 contracts (0.2x).
Is percentage mirroring different from ratio scaled counts?
No, not when the follower's entry equals the ratio times the master's entry exactly. Vendors present percentage mirroring and ratio-scaled counts as two distinct features. They are the same arithmetic. For Follower A, percentage mirroring computes 0.30 x 5 = 1.5 at stage 1. Ratio-scaled counting computes 3 x 0.5 = 1.5. Same number, every stage.
The two modes diverge only once the follower's entry was itself rounded or risk-capped, which is precisely the situation where the copier stops being trustworthy. Take a follower configured at 0.5x whose own risk cap allowed only 3 contracts, giving an effective ratio of 0.3x. Under percentage-of-own-position, stage 1 computes 0.30 x 3 = 0.9, floors to 0, and the follower holds all 3 contracts, 100% of its original, against the master's 70%. Under configured-ratio-scaled counting, stage 1 computes 3 x 0.5 = 1.5, floors to 1, and the follower holds 2, which is 66.7% of its original against the master's 70%. Same master event, same account, errors of opposite sign, decided by an implementation detail most products never document.
A third mode exists: percentage of current remaining rather than percentage of original. The master's stage 2 closes 4 of the 7 it still holds, which is 57.142857% of current, not 40%. A copier reading each fill as a fraction of what is still open emits a different follower sequence than one reading it as a fraction of the original entry. Which of the three your copier implements is usually undocumented, so test it on a demo or evaluation account before trusting it with a staged exit.
What happens to a 5-contract follower under naive mirroring?
It ends up holding a constant half-contract more than it should, and its relative over-exposure more than doubles as the master's base shrinks. Follower A holds 5 contracts against the master's 10, and the copier uses percentage mirroring with floor rounding.
Stage 1. 0.30 x 5 = 1.5, floors to 1. The follower closes 1 and holds 4. The master holds 7. The proportionally correct holding is 0.5 x 7 = 3.5, so the follower carries an excess of 0.5 contracts. Relative over-exposure is (4/5) divided by (7/10) = 8/7 = 1.142857, or +14.3%. One stage in, the positions have separated.
Trace where the half contract goes. The master's fill event enters the copier, is multiplied by the 0.5 ratio to give 1.5, hits the rounding gate, and leaves as a single order for 1 contract. The gate discards the remaining 0.5. Nothing downstream absorbs it. It stays on the follower's book as live futures exposure the master no longer carries.
Stage 2. 0.40 x 5 = 2.0 exactly, so 2 close and the follower holds 2. The master holds 3. The proportional holding is 0.5 x 3 = 1.5, so the excess is again 0.5 contracts. But relative over-exposure is now (2/5) divided by (3/10) = 4/3 = 1.333333, or +33.3%. The absolute gap held steady at half a contract while the relative gap more than doubled, because the master's remaining base shrank from 7 to 3.
Stage 3. The master flattens its last 3 and the follower closes its remaining 2. Total closed: 1 + 2 + 2 = 5. The books balance at the end. They were wrong the entire way through.
The rounding error never cancels, because floor rounding on a close order always closes less than intended.
The drift is not distributed around zero. Flooring a close closes less than the exact figure, every time, on every account, so the follower is systematically more exposed than the master during the precise window in which the master decided the trade needed less risk.
Why a small follower skips entire stages
Below a certain size the failure stops being gradual and becomes categorical: the follower emits no order. Follower B holds 2 contracts. Stage 1 computes 0.30 x 2 = 0.6, which floors to 0. Stage 2 computes 0.40 x 2 = 0.8, which also floors to 0. Nothing is sent either time.
After stage 1 the follower holds 2 (100% of its original) while the master holds 7 (70%). The proportional holding is 0.2 x 7 = 1.4, so relative over-exposure is 2/1.4 = 1.428571, or +42.9%. After stage 2 the follower still holds 2 while the master holds 3. The proportional holding is 0.2 x 3 = 0.6, so the follower carries 2/0.6 = 3.33 times its intended exposure on a trade the originating strategy considers 70% closed. It participated in zero of the master's risk reductions, then dumped everything at stage 3 alongside the master's final leg.
Platforms do not paper over this. The MetaQuotes-published book MQL5 Programming for Traders, in its section on permitted volumes of trading operations, states that attempts to trade a volume less than the minimum, more than the maximum, or not a multiple of the step result in an error, and treats normalising volumes before calling the trading functions as the calling code's job. Those minimum, maximum and step values are broker-defined per symbol, not fixed by the platform or the exchange, so read them from the symbol rather than assuming. An unnormalised scale-out leg is rejected whole. The copier owns the rounding decision because nothing downstream will make it.
Does carrying the fractional remainder fix the drift?
Largely, yes, and it rescues the small-account case that floor rounding cannot. Compute the exact owed quantity, round to nearest, send that, and carry the signed difference into the next stage.
Run it on Follower B. Stage 1: owed = 0.30 x 2 = 0.6, rounds to 1, close 1, residual carried = 0.6 - 1 = -0.4, holds 1. Stage 2: owed = 0.40 x 2 = 0.8, plus the carried -0.4 gives 0.4, rounds to 0, close nothing, residual = 0.4, holds 1. Stage 3: owed = 0.6 plus the carried 0.4 = 1.0, close 1, holds 0. Total closed 1 + 0 + 1 = 2. The exposure path is 1, 1, 0 against an intended path of 1.4, 0.6, 0. Compare the naive path of 2, 2, 0. Carry-forward turned a total miss into a close approximation.
On Follower A with round-half-up, stage 1 owes 1.5 which rounds to 2, close 2, residual -0.5, holds 3. Stage 2 owes 2.0 plus -0.5 = 1.5, rounds to 2, close 2, residual -0.5, holds 1. Stage 3 owes 1.5 plus -0.5 = 1.0, close 1, holds 0. Total 2 + 2 + 1 = 5. The exposure path is 3 then 1 against an intended 3.5 then 1.5, so this errs consistently under, which is the safer direction, whereas the naive floor path of 4 then 2 errs consistently over. With a 0.5 ratio and odd master remainders the half contract is unavoidable. The only real choice is which way the error points.
Why target-remaining beats close-this-much
Because it reproduces residual carry with no accumulator, and because it absorbs failed orders. Instead of translating each master fill into a close quantity, compute the follower's target holding from the master's absolute remaining position and issue the difference: target = round(ratio x master_remaining), then order = currently_held - target.
On Follower B, rounding half up: stage 1 target is round(0.2 x 7) = round(1.4) = 1, holds 2, close 1. Stage 2 target is round(0.2 x 3) = round(0.6) = 1, holds 1, close 0. Stage 3 target is 0, close 1. Path 1, 1, 0, identical to the residual-carry result, with no carry state maintained anywhere. That is not a coincidence. The target is recomputed from the master's absolute state at every stage instead of being integrated from a stream of deltas, so nothing is left to accumulate.
The decisive advantage is not rounding at all. It is failure absorption. Suppose Follower A's stage-1 close of 1 contract is rejected for margin, or lost to a disconnect, or blocked by a halt. A delta copier still believes stage 1 happened. Its stage-2 instruction is close 2, so the follower closes 2 from its untouched 5 and holds 3 while the master holds 3. Intended holding is 0.5 x 3 = 1.5, so the follower sits at exactly twice its intended exposure and the stage-1 miss is never recovered. A target copier computes stage 2 as target = round(0.5 x 3) = round(1.5) = 2, sees the follower still holding 5, and issues a close for 3 in one shot, landing exactly on 2. The missed stage self-heals with no reconciliation pass.
Rejections are not rare across a fleet of prop accounts. Different brokers, margin states and connection paths mean one account failing a leg while the others fill is routine, the same class of problem discussed in multi-account copy trading sync and slippage risk. A delta copier turns a transient failure into a permanent size error. A target copier turns it into a one-stage delay.
Which direction should the copier round?
There is no standard, so the choice is yours and its consequences are asymmetric. Floor, ceiling, round-half-up, round-half-down and banker's rounding are all defensible, and neither exchange rules nor platform trading APIs prescribe a rounding rule for a scaled partial close. They only reject volumes that violate the symbol's own minimum, maximum and step. Two copiers reading the identical master event legitimately produce different follower orders.
Rounding direction and mirroring mode are undocumented in most copy trading products. Run a deliberate three-stage exit on a demo or evaluation account and record the exact contract counts each follower closes before you trust the tool with a scale-out strategy.
The table prices the 2-contract follower under three designs. The master's three stages fill at +10, +20 and +40 points in the winning sequence, and at +10, +20 and -5 points in the losing sequence. The E-mini S&P 500 (ES) contract is $50 per index point. These levels are chosen to make the arithmetic legible, not to model typical price behaviour.
| Copier design | Fills at stages 1 / 2 / 3 | Winning sequence (target $2,300) | Losing sequence (target $950) |
|---|---|---|---|
| Naive percentage of own original, floor | 0, 0, 2 | $4,000 (+$1,700) | -$500 (-$1,450) |
| Target-remaining, round to nearest | 1, 0, 1 | $2,500 (+$200) | $250 (-$700) |
| Target-remaining, round down | 1, 1, 0 | $1,500 (-$800) | $1,500 (+$550) |
Rounding the target down guarantees the follower is never more exposed than the master, and pays for that guarantee by cutting runners early. In the third row the follower is flat before the master's final leg: a three-stage master exit became a two-stage follower exit. Round-to-nearest tracks the proportional result most closely in both directions. Naive delta mirroring is worst in both.
One platform behaviour works in your favour on the final stage. The MQL5 Standard Library reference for CTrade::PositionClosePartial states that if the requested value exceeds the volume of the position being partially closed, the position is closed in full and no position in the opposite direction is opened. On MetaTrader 5 a copier that over-computes its last close therefore fails safe rather than reversing into a fresh short. Do not assume every platform in your stack behaves that way; verify each one.
What exit drift costs on a funded account
It consumes drawdown against exposure the strategy no longer wants. Follower B holds 2 ES contracts after stage 2 while its proportional target is 0.6 contracts. A 20-point retrace costs 2 x 20 x $50 = $2,000 of open equity, against a proportional expectation of 0.6 x 20 x $50 = $600. That is $1,400 of extra drawdown burned on a position the originating strategy already regards as 70% closed, and on an account with a trailing threshold it is charged against the survival budget. Whether your firm computes drawdown on unrealised or realised equity, and whether it trails intraday or end of day, varies by firm, so confirm the current rule with yours.
The upside is no compensation. Using the same exit levels, the master's 3/4/3 winning sequence produces 3 x 10 + 4 x 20 + 3 x 40 = 230 points, or $11,500. Follower A under naive mirroring closes 1 at +10, 2 at +20 and 2 at +40 for 130 points, or $6,500, against a proportional target of $5,750. It beat target by $750, purely from being over-exposed into the best fill. Now stop the final leg at -5 points instead: the master books 95 points, or $4,750, and Follower A books 40 points, or $2,000, against a target of $2,375, missing by $375. The same rounding rule produced +$750 and -$375. That is dispersion the follower is paid nothing to carry.
For Follower B the losing sequence is worse than dispersion. It skipped both profitable stages and closed all 2 contracts at -5 points: 2 x (-5) x $50 = -$500, against a proportional target of +$950. The master booked a $4,750 winner and the follower booked a $500 loser on the identical trade.
An established fairness benchmark exists for splitting one execution across differently sized accounts. In its 2003 rulemaking on bunched orders under CFTC Regulation 1.35, the Commission required that allocations be fair and equitable, that no account or group of accounts receive consistently favorable or unfavorable treatment, and that the allocation method be sufficiently objective and specific to permit independent verification of fairness by regulators, self-regulatory authorities and outside auditors. That rule reaches account managers with written investment discretion over customer accounts, not a trader mirroring between accounts they own, and paragraph numbering within Regulation 1.35 has been reorganised since 2003, so read the current text before citing it. The arithmetic standard is still the right one to hold yourself to: a rounding rule that systematically favours large followers over small ones fails it on the numbers, whatever its regulatory status.
When faithful scale-out replication is impossible
Below a threshold, no copier design fixes this, because futures contracts are indivisible whole units. The rule: the master's smallest stage fraction times the follower's contract count must floor to at least 1. For a smallest stage of 30%, that means at least 4 contracts, since 0.30 x 4 = 1.2 which floors to 1, while 0.30 x 3 = 0.9 which floors to 0. Any follower under 4 contracts silently skips that stage no matter what software you point at it. Recompute the threshold for your own master's plan.
Micro contracts dissolve the problem where they are available. A 0.5x follower of a 10-contract ES master can hold 50 Micro E-mini S&P 500 (MES) contracts instead of 5 ES. Notional check: 50 x $5 per point = $250 per point against the master's 10 x $50 = $500 per point, exactly 0.5x. Scale-out check: 30% of 50 = 15, 40% of 50 = 20, 30% of 50 = 15, all whole numbers summing to exactly 50. Zero rounding error at every stage. The cost is real: 50 micro round turns instead of 5 mini round turns is ten times the per-contract commission events, and micro book depth is thinner, so granularity is bought with fees and fill quality. Confirm your current per-side commission and how your firm counts micros against any maximum-contract limit before restructuring an account this way.
The honest tradeoff: for a sub-4-contract follower on a three-stage percentage exit, a trade copier is the wrong tool. Not a weak tool, the wrong one. The options are to size the account so partial exits round sensibly, move it to micros, trade it manually, or deliberately give it a coarser two-stage exit plan and size its entry knowing that is what it will get. What you should not do is run a three-stage master into a 2-contract follower and assume the software is handling it.
Frequently asked questions
What is a scale-out in copy trading?
A scale-out is an exit taken in stages rather than all at once, for example a 10-contract position closed as 3, then 4, then 3. In copy trading it matters because each stage produces a separate quantity for the copier to scale and round for every follower. An entry rounds once; a three-stage exit rounds three times, each against a position the previous rounding already distorted.
Why does my follower account close fewer contracts than the master?
Because the scaled quantity was fractional and the copier rounded it down. A 0.5x follower of a 10-contract master closing 3 contracts computes 0.30 x 5 = 1.5, and floor rounding sends an order for 1, leaving half a contract of unintended exposure on the follower's book. Floor rounding on a close always closes less than intended, so the bias is systematic rather than random.
Should a copier close a percentage or a fixed number of contracts?
Neither, if you have the choice: it should send a target remaining position. Percentage-of-original and ratio-scaled-count mirroring are algebraically identical whenever the follower's entry equals the ratio times the master's entry exactly, so they are not really two options. They diverge only once the follower's entry was itself rounded or risk-capped, and at that point they produce errors of opposite sign on the same account.
What is target-remaining copying?
Target-remaining computes what the follower should still be holding rather than how much it should close: target = round(ratio x master_remaining), then order = currently_held - target. Because the target is recomputed from the master's absolute position at every stage instead of integrated from a stream of deltas, it tracks the proportional holding without maintaining any fractional carry state. Its bigger advantage is that a rejected or missed stage self-corrects at the next stage.
Can a 2-contract account copy a 10-contract master's scale-out?
Not faithfully, and not because of the software. Under naive floor rounding a 2-contract follower computes 0.30 x 2 = 0.6 and 0.40 x 2 = 0.8, both of which floor to zero, so it sends no order for the first two stages and carries 100% of its size while the master de-risks to 30%. Target-remaining with round-to-nearest improves the holding path to 1, 1, 0 contracts, but a three-stage exit still cannot be represented exactly on two indivisible units.
Do micro contracts solve partial close rounding?
Yes, where they are available, because they multiply the granularity by ten. A 0.5x follower holding 50 MES instead of 5 ES has the same notional exposure (50 x $5 = $250 per point versus 10 x $50 = $500 per point) but its 30/40/30 exit becomes 15, 20 and 15 contracts, all whole numbers summing exactly to 50. The cost is ten times the per-contract commission events and thinner book depth, and you should confirm how your firm counts micros against any maximum-contract limit first.
How does MetaTrader 5 handle a partial close larger than the position?
The MQL5 Standard Library documents that when the requested value exceeds the volume of the position being partially closed, CTrade::PositionClosePartial closes the position in full and opens no position in the opposite direction. A copier that over-computes its final-stage close therefore fails safe on MT5 rather than reversing into an unintended trade. It does not mean MT5 rounds for you: volumes below the symbol minimum, above the maximum, or not a multiple of the step return an error, and those values are broker-defined per symbol.
What is the minimum follower size for a scale-out to replicate correctly?
The follower needs enough contracts that the master's smallest stage fraction times that count floors to at least 1. For a smallest stage of 30%, that is 4 contracts, since 0.30 x 4 = 1.2 which floors to 1 while 0.30 x 3 = 0.9 which floors to 0. Recompute the threshold for your own master's plan, because a master whose smallest stage is 20% needs a follower of at least 5 contracts.
How much drawdown does exit drift cost a funded account?
It costs the full unrealised loss on exposure the strategy no longer wants to hold. A follower carrying 2 ES contracts when its proportional target is 0.6 loses 2 x 20 x $50 = $2,000 on a 20-point retrace against a proportional expectation of 0.6 x 20 x $50 = $600, so $1,400 of drawdown is consumed on a trade the master considers 70% finished. Whether your firm computes drawdown on unrealised or realised equity and whether it trails intraday varies by firm, so verify the current rule.
When should I not use a trade copier for a staged exit strategy?
When the follower account cannot hold enough contracts for the master's smallest exit stage to round to at least one contract. Below that threshold the failure is arithmetic, not engineering: futures contracts are indivisible whole units, so no copier design recovers a stage that computes to 0.6 contracts. The honest responses are to size the account up, switch it to micros, trade it manually, or deliberately plan a coarser two-stage exit and size the entry accordingly.