Restoring Position Size After a Drawdown

Restore position size after a drawdown through a pre-defined sequence of size steps, not a single jump back to normal. Before the drawdown, fix how many steps exist, what fraction of normal size each step permits, and what sample-based evidence unlocks the next one: a minimum number of qualifying trades, a profit-factor floor with a predefined rule for edge cases, an expectancy floor measured in R rather than dollars, and a check that the sample’s MFE/MAE profile still matches the plan. Every gate must clear on the same qualifying sample before size advances one step. A single winning trade, a return to breakeven equity, or a feeling of being “back on track” is not evidence that the process has stabilized.

This is not a general drawdown framework. Trading risk management sets the account, session, and trade-level limits this schedule operates inside, and how to adjust risk during and after a trading drawdown covers the full ladder: measuring the drawdown, setting triggers, and reducing size as it deepens. This article owns the restoration side of that ladder: once a trader is ready to size back up, what threshold gates each step, and what has to be decided in advance for that threshold to hold up under pressure.

How do you restore position size after a drawdown?

Restore it one pre-defined step at a time, where each step is a fixed fraction of the normal risk unit. Advancing requires a qualifying sample, collected under a sample-window design fixed before the drawdown, that clears four gates together: a minimum trade count, a profit-factor floor, an expectancy floor in R, and MFE/MAE consistency with the pre-drawdown baseline. If a sample fails any one gate, size stays at the current step and evaluation continues under the same predefined rule — the definition does not change because a different window or a shorter stretch would look more favorable.

What “returning to normal size” means—and does not mean

“Back to normal” is often treated as a single event: equity crosses a line, and full size resumes. That conflates two questions a restoration threshold needs to keep separate.

Restoration question	What it governs
Has the account recovered financially?	Whether equity or the drawdown measure has moved back through a stated boundary
Has the process been reviewed?	Whether trades since the drawdown began reflect the intended method, not drift
Is there enough evidence to trust a size increase?	Whether the completed step’s sample is large enough, and good enough, to justify moving up
What size is permitted right now?	The fraction of the normal risk unit active until the next gate clears

A step threshold answers the third and fourth questions. It does not diagnose why the drawdown happened, and it does not replace the review that decides whether the strategy itself needs a change.

Equity recovery vs. process recovery

None of the following answers the process question above, though each is routinely treated as if it does:

Breakeven equity shows the account recovered its dollars, not that the trades producing that recovery cleared the sample-floor, profit-factor, expectancy, and MFE/MAE gates.
One winning week is calendar time, not a sample; it may hold few qualifying trades and none collected under the predefined sample-window design.
Several consecutive winners is exactly the short, favorable-looking stretch a predefined sample window (below) exists to keep from being mistaken for a completed evaluation.
Recovering a percentage of the drawdown is a financial threshold, identical in kind to breakeven — it measures how far equity moved, not what produced the move.

Financial recovery can occur alongside process recovery: equity recovery does not substitute for the predefined restoration gates, and none of the four items above, alone or combined, is itself the evidence a restoration gate evaluates.

What counts as a qualifying trade?

Define this before restoration begins, or it becomes a decision made under pressure at the exact moment a borderline trade would change a gate’s outcome. The treatment of rule violations, sizing violations, scratches, breakevens, strategy-version changes, and other borderline observations must be predefined, outcome-independent, and applied consistently — not chosen case by case because a particular treatment happens to help the current sample.

Rule-aligned trades. Decide whether only trades that followed the active method count, or whether every trade counts with deviations flagged for review instead of excluded.
Same strategy version. If the method changes mid-sample, decide whether prior-version trades still count or the change restarts the sample — the section on strategy changes below covers this directly.
Execution or sizing violations. Decide whether a deviating trade counts as-is, counts flagged, or is excluded.
Scratch and breakeven trades. Decide whether they count toward the trade floor and how a near-zero R result enters the expectancy total.
Setup or instrument heterogeneity. Decide whether several setups share one pooled sample or need separate ones — the next section covers the trade-off directly.

A trade should not be included or excluded because its realized outcome improves or worsens profit factor, expectancy, MFE/MAE, or any other restoration gate. The definition cannot be changed retroactively to include or exclude a specific trade because of its effect on the gate math. There is no single universally correct qualifying-trade policy; the requirement is that whichever policy is chosen is fixed in advance and applied the same way regardless of outcome.

Pooled versus stratified samples

The setup-or-instrument heterogeneity question above deserves its own treatment, because a pooled result can look acceptable while concealing a real problem:

Setup A: 10 trades, +4R total = +0.40R/trade
Setup B: 10 trades, -2R total = -0.20R/trade
Combined: 20 trades, +2R total = +0.10R/trade

Against a hypothetical expectancy floor of +0.05R/trade, the combined sample clears while Setup B alone does not — the pooled result can pass an expectancy gate while one subgroup is deteriorating underneath it. A pooled sample answers a combined, process-level question: is the trader’s overall approach, across everything they trade, meeting the restoration gates? A stratified sample — a separate qualifying sample per setup, instrument, or regime — answers a narrower, setup-specific question instead.

Pooling reaches a usable sample size faster, since every trade counts toward one floor, but subgroup deterioration can hide inside an acceptable aggregate, exactly as above. Stratification gives more diagnostic specificity, since each setup’s own gates clear on their own evidence, but each stratified floor takes longer to reach with a smaller sample per subgroup.

Neither approach is universally superior. The sample structure has to match the question the restoration gate is meant to answer: whether the overall process is ready to size up, or whether one specific setup is.

What if the strategy changes during drawdown recovery?

A material change to the process generating the trades can make the original restoration sample no longer directly comparable, so the treatment of strategy changes should be predefined — the same principle as the qualifying-trade definition above, applied to changes in the method itself rather than to individual trades.

Changes worth defining in advance include entry logic, exit logic, stop methodology, target methodology, setup definition, the instrument universe, execution rules, and material trade-management changes. Not every change needs the same response: a predefined rule can keep the current sample valid, partition observations before and after the change, or restart the sample entirely, depending on how directly the change affects the behavior the gates measure. This is a boundary question for the sample, not a strategy-validation exercise — the goal is only to decide whether a change keeps the sample comparable, not whether the changed strategy is sound.

How many trades should you complete before increasing size?

There is no universal number of trades. The right floor depends on the sample’s own characteristics, not a borrowed rule:

Trade frequency — several trades a day reaches a given floor in days; a few a month can take a season.
Payoff distribution — frequent small wins with rare large winners need a larger sample before expectancy and profit factor stop depending on whether the rare winner has appeared.
Variance — higher-variance methods need more observations before an average is signal rather than noise.
Setup heterogeneity — several distinct setups may need a larger pooled sample or separate ones, per the qualifying-trade definition above.
Instrument and market regime — a floor calibrated in one instrument or regime may not transfer to another untested.
The trader’s own historical data — a longer pre-drawdown track record gives a basis for calibrating the floor a method needs. Without sufficient historical data, the floor is harder to calibrate and the uncertainty around whichever threshold is chosen is higher.

Increasing the sample floor by itself does not solve every calibration problem. A larger N does not correct for poor regime matching between the sample and current conditions, setup or instrument heterogeneity pooled into one number, non-stationarity in the method’s underlying edge, or a structural strategy change mid-sample. A larger sample is more observations of whatever process actually generated it — not a substitute for making sure that process is the one the trader intends to evaluate.

Bailey and López de Prado support the narrower statistical point that the reliability of a performance estimate depends partly on the amount and characteristics of the track record behind it, at a chosen confidence level.¹ Their minimum-track-record-length framework was built for evaluating fund and strategy track records broadly. It does not establish that 15 trades is sufficient, that any specific profit-factor or expectancy floor is correct, that any particular MFE/MAE tolerance is correct, that restoration must happen in three steps, or that restoration rules must be fixed before a drawdown. The requirement to predefine the sample floor, window design, evaluation cadence, and restoration gates before the drawdown is a rule of this framework, intended to prevent retrospective threshold selection — not a conclusion the cited paper reaches.

Why a step schedule instead of a jump

Reducing risk on the way down is usually a single crossing: the drawdown measure passes a line, and lower size applies immediately. Restoring risk should not run that logic in reverse, because a single crossing — one winning trade, one green day — is a sample of one. It says little about whether later trades will resemble the intended method any more reliably than the trades that produced the drawdown did.

A step schedule turns that asymmetry into a mechanism instead of an argument: each step permits a fraction of normal size, and advancing requires a sample large and consistent enough on every gate to have been decided as sufficient before the drawdown began — not a single result that feels convincing now.

Define the sample window before you need it

A step’s qualifying sample is collected under one of a few designs, each with a different failure mode.

Fixed-batch. Exactly N qualifying trades are evaluated once as a batch; a failed batch is replaced by a new batch of N, not extended. Simple to compare batch to batch, but a narrow miss restarts from zero, tempting a trader to redefine the batch size after the fact.

Expanding. The sample grows from a fixed start point and is re-evaluated after each new trade; nothing is discarded. More cumulative evidence accumulates over time, but because earlier observations never leave the sample, the aggregate can become slow to reflect a genuine change in the current process — an early loss, or a since-changed market condition, continues influencing the result long after it stops being representative, unless a maximum window is also predefined.

Rolling. The most recent N qualifying trades are evaluated; when a new trade enters, the oldest leaves. It reflects only recent behavior, but its composition changes after every qualifying trade, which raises a decision-rule question distinct from the window definition itself: evaluating after every trade and advancing at the first observation that clears every gate is not the same decision rule as evaluating the identical sample only at fixed, predefined checkpoints. The first creates many separate opportunities for a threshold to be crossed by chance; the next section addresses this directly.

No design is universally correct. The rule that matters is this: the sample-window definition must be fixed before restoration begins and cannot be changed because another window produces a more favorable result.

How often should restoration gates be evaluated?

The evaluation cadence should be predefined separately from the sample-window design. These are two different variables: what observations are in the sample, and when that sample is allowed to be evaluated against the gates.

Sample design	Possible evaluation cadence
Fixed batch	Once, after the batch reaches its full N qualifying trades
Expanding	After every new qualifying trade, or at predefined checkpoints
Rolling	After every new qualifying trade, or at predefined checkpoints

There is no universal correct cadence. The point that matters: evaluating after every trade and advancing at the first threshold crossing is not equivalent to evaluating the same sample only at predefined checkpoints. Checking a rolling or expanding sample after every trade creates many separate opportunities for the gates to appear to clear by chance, in a way a single predefined checkpoint does not — fix the checkpoint schedule before restoration begins, and do not add an extra check because the current sample looks close to passing.

Sample design, sample length, and evaluation cadence are three separate decisions. All three need to be predefined before restoration begins, not settled individually as each becomes relevant.

The gates that unlock the next step

Gate	What it measures	Why it exists
Sample floor	Qualifying trades completed under the predefined window design	A single trade or short run cannot show the method has stabilized
Profit factor floor	Sum of positive realized R ÷ absolute sum of negative realized R, with a predefined rule for undefined or unstable results	Keeps the gate consistent with this framework’s R-normalization, since planned dollar risk changes across steps
Expectancy floor (in R)	Total realized R over the sample ÷ qualifying trades	Stays comparable across steps at different position sizes, unlike a fixed dollar figure
MFE/MAE profile	Median MFE and median MAE, in R, versus the trader’s pre-drawdown baseline	Flags execution drift without mechanically growing just because size did

Expectancy and the MFE/MAE profile are expressed in R — result ÷ that trade’s own planned risk — because planned risk itself changes at every step; a raw dollar figure scales with size even when execution is identical, and R removes that artifact. A dollar figure can still be reported afterward as illustrative, as the worked example below does, but it is not the gate itself. This scoping is specific to comparing across steps within a restoration schedule; it is not a claim that expectancy analysis must always be performed in R in every trading context.

Win rate is not a gate on its own: two samples with an identical win rate can have opposite expectancy if their average win and loss differ, which is why win rate belongs inside the profit-factor and expectancy gates rather than standing alone.

Why all gates must clear together

The four gates are conjunctive, not a weighted score: every predefined gate must clear together, and no gate’s strength substitutes for another’s shortfall unless the trader explicitly designed a weighted framework beforehand.

A profit factor of 2.0 does not compensate for a sample of five trades against a floor of fifteen.
Strong expectancy does not compensate for a batch showing execution-profile drift — that is evidence the review step has not finished its job.
A large sample does not compensate for sub-floor expectancy; more trades of an unprofitable pattern is not progress.
A passing MFE/MAE profile does not compensate for an invalid qualifying sample; a clean execution profile built on the wrong trades is not evidence of anything the schedule is meant to measure.

Handling profit-factor edge cases

Within this restoration framework, profit factor is the sum of positive realized R divided by the absolute sum of negative realized R over the qualifying sample — R, not dollars, for the same cross-step reason expectancy is normalized in R. A trader who uses a different predefined profit-factor convention must apply it consistently across every restoration step. Three cases still need a predefined answer, not an improvised one:

The sum of negative realized R is zero. An all-win batch makes profit factor undefined — division by zero, not proof that the PF gate has been satisfied.
The sum of negative realized R is extremely small. One small loss against several large wins can produce a very large but unstable ratio.
The sample is too small. A ratio from a handful of trades is more outlier-sensitive than the same ratio from a full predefined batch.

Decide in advance how an undefined or unstable observation is handled — for example, an additional full batch evaluated on expectancy and MFE/MAE before profit factor is used again — rather than reading an unusual number as an automatic pass or fail.

Build the schedule before the drawdown

step N permitted size = normal risk unit × step N size fraction

R per trade = realized result ÷ planned risk for that trade
expectancy in R = total realized R over the sample ÷ qualifying trades
profit factor (in R) = sum of positive realized R ÷ absolute sum of negative realized R

advance to step N+1 only if, over step N's sample (under the predefined window design
and evaluation cadence):
  trade count ≥ sample floor
  AND profit factor clears its floor under the predefined edge-case rule
  AND expectancy in R ≥ expectancy floor (R)
  AND MFE/MAE profile (in R) is within the predefined tolerance of baseline

Step	Permitted size	Advance condition
1 (minimum)	Normal risk unit × smallest fraction	All four gates clear on step 1’s sample
2 (partial)	Normal risk unit × intermediate fraction	All four gates clear on step 2’s sample
3 (near-normal)	Normal risk unit × largest sub-normal fraction	All four gates clear on step 3’s sample
Normal	Full normal risk unit	Destination state

The fractions, sample floor, window design, and each floor are the trader’s own decisions, tested outside live trading. There is no universal schedule. Its only job is to make the transition auditable: at any point, the trader can name the current step, the sample collected under it, and which gate is still open.

Worked example: three fixed-batch steps back to normal size

Assume a normal risk unit of $200, reduced to $50 (step 1) during a drawdown. Before the drawdown, the trader fixed a fixed-batch design of 15 trades per batch, a profit-factor floor of 1.3, an expectancy floor of +0.15R, and an MFE/MAE tolerance of ±25% of baseline medians, in R.

Step 1 ($50 = 1R). The first batch of 15 totals +2.85R (expectancy ≈ +0.19R, above floor). Positive realized R totals 5.2R; negative realized R totals 2.35R (absolute); profit factor ≈ 2.21 (above 1.3). Median MFE 0.85R, median MAE 0.50R, within tolerance. All four gates clear, so the next session begins at step 2 ($100) — not full size. As an illustrative dollar figure, +0.19R at $50 is about $9.50 per trade.

Step 2, first batch ($100 = 1R). A new batch of 15 begins — fixed-batch does not extend step 1’s trades. It totals +0.95R (expectancy ≈ +0.06R, below the +0.15R floor). Positive realized R totals 3.35R; negative realized R totals 2.40R (absolute); profit factor ≈ 1.40 (clears). Profit factor passes; expectancy does not, and the gates are conjunctive, so the step does not advance. A second, entirely new batch of 15 begins at the same $100 unit.

Step 2, second batch. Totals +2.40R (expectancy = +0.16R, clears). Positive realized R totals 4.5R; negative realized R totals 2.10R (absolute); profit factor ≈ 2.14. MFE/MAE within tolerance. All gates clear, so the next session begins at step 3. As a comparison, +0.16R at $100 is about $16.00 per trade — a larger dollar figure than step 1’s $9.50, though both batches cleared the identical +0.15R floor. A fixed $10 dollar floor would have been +0.20R at step 1 but only +0.10R at step 2 for the same nominal number; R is what keeps the floor constant across steps.

This example uses a fixed-batch design throughout: a batch that fails is replaced by an entirely new batch, never rolled or expanded forward. It is a mechanics example, not a recommended fraction, floor, tolerance, window design, evaluation cadence, or sample size for another trader’s method.

Restoration at a glance

Check	If passed	If failed
Is the sample eligible for evaluation (predefined checkpoint reached, trades qualify)?	Continue to the size check	Wait for the next predefined checkpoint, or review the qualifying-trade definition
Has the sample reached the required size (sample floor)?	Continue to the numeric gates	Stay at the current step; keep collecting the qualifying sample
Do the numeric gates clear (profit factor and expectancy in R, including edge-case handling)?	Continue to execution-profile validation	Stay at the current step
Does execution-profile validation clear (MFE/MAE profile, in R, within tolerance)?	Advance one step	Stay at the current step
Does the trader advance?	Size moves to the next predefined step	Size remains at the current step
Has an existing drawdown-reduction trigger fired?	Follow the drawdown ladder immediately; reduction overrides restoration	Continue the restoration evaluation as normal

Failure modes that defeat the schedule

Failure mode	What changes under pressure	Repair outside the live session
One win skips a step	A single outcome is treated as proof the process is fixed	Require the full sample floor before checking any other gate
Win rate substitutes for expectancy	Small wins look like recovery while payoff deteriorates	Check profit factor and expectancy together
The sample window is cherry-picked	A favorable stretch replaces the full predefined window	Fix the window design before the drawdown; never switch it mid-stream
Floors are loosened mid-drawdown	A floor is quietly lowered because the trader wants to move up	Treat floors as fixed inputs; a missed gate means more sample, not a new gate
A profit-factor edge case reads as a pass	An undefined or unstable PF (near-zero loss, tiny sample) looks unusually strong	Predefine edge-case handling before it happens
MFE/MAE drift is ignored	The excursion profile shifts while P&L still looks fine	Compare the profile explicitly, in R
Steps are skipped entirely	One strong batch justifies jumping straight to normal size	Advance one step at a time regardless of batch strength
A reduction trigger is ignored after stepping up	The higher step is treated as the new normal	The drawdown ladder always takes precedence over the current step

A schedule that is too conservative is also a failure mode: floors set so high they are rarely reached leave a trader under-sized indefinitely with no review point. That should trigger a design review outside live trading, not an improvised exception.

What happens if performance deteriorates after increasing size?

Restoration gates and the drawdown-reduction ladder govern two different directions, and only one applies once a reduction trigger is active. Restoration only ever authorizes moving up; it says nothing about moving down.

If a predefined reduction trigger fires after a trader has already advanced a step, the reduction rule takes precedence immediately — a step’s gates having cleared in the past does not argue the trigger away in the present. Once the reduction state is resolved, restoration resumes from the state specified by the trader’s predefined drawdown ladder and re-entry rules, not automatically from the step reached before deterioration. This article does not duplicate that re-entry framework; it only establishes that reduction takes precedence over restoration whenever both could apply.

Connect restoration to review

The gates here are inputs to post-trade review, not a replacement for it. Review still asks whether each trade actually followed the stated method, not only whether the aggregate numbers cleared their floors. A numeric pass does not override rule deviations: any deviation contained in the sample must still be handled under the trader’s predefined qualifying-trade and review policy — whether that policy excludes the trade, includes it flagged for review, or applies another outcome-independent treatment — before the restoration decision is finalized.

Restoring full size before a step’s gates clear is a form of risk escalation: exposure moving above what the currently active state permits, regardless of the justification.

Where Costante fits

Costante supports the behavioral side of running a restoration schedule: making the active step and its permitted size visible before a decision, giving the trader low-friction logging to record each qualifying trade as it happens, and structuring review around whether the stated gates were actually checked before size changed.

Costante does not choose the step fractions, sample floor, window design, or metric floors; does not calculate profit factor, R-normalized expectancy, or an MFE/MAE profile from a live price or broker feed; does not connect to a broker or verify a prop-firm account; and does not decide when a trader is ready to return to normal size. Those figures come from the trader’s own records and calculation. The trader remains responsible for the schedule’s design and every sizing decision made against it.

Frequently asked questions

Should you return to full position size when the account reaches breakeven?

Not automatically. Breakeven answers only the financial question, not whether the sample since the drawdown has cleared the sample-floor, profit-factor, expectancy, and MFE/MAE gates.

How many winning trades should you wait before increasing size?

There is no universal consecutive-win threshold. A win streak is not the qualifying sample; use the predefined sample and its gates instead.

Should you increase position size after one profitable week?

Not on the basis of the profitable week alone. Calendar time is not sample quality: a profitable week could coincide with completion of a valid, predefined sample whose gates all clear, or it could contain almost no qualifying trades. The week itself is not the evidence; whether the predefined sample and its gates cleared is.

Should expectancy be measured in dollars or R?

For cross-step comparisons in this restoration framework, R is the more coherent unit, because planned dollar risk changes across steps. Dollar expectancy can still be useful as a reporting or account-level measure — it just should not be the cross-step gate itself.

What if profit factor passes but expectancy fails?

The step does not advance. The gates are conjunctive; a pass on one does not offset a failure on another unless a different scoring framework was deliberately predefined.

Can you skip a restoration step after a very strong run?

Not under a schedule requiring sequential advancement. A strong result at one step only qualifies the next step, not a further one — evidence collected at one size cannot authorize a size it never measured.

Sources

Costante provides educational workflow tools, not financial advice. Trading involves risk.

For the broader performance framework around risk restoration, see trading performance.

Bailey, D. H., & López de Prado, M. (2012). The Sharpe Ratio Efficient Frontier. Journal of Risk, 15(2), 3–44. ↩