An EA with negative expectancy, a partial close rule added, and suddenly it breaks even. Is that progress, or a sign of curve fitting?
How partial closes affect results:
- Closing half at 1R and moving the stop to breakeven reduces average loss, since many trades end at breakeven instead of -1R.
- But it also reduces average win, since only half the position reaches the full target.
- Win rate may increase (breakeven trades counted as not losing).
When it's genuine improvement:
- The strategy often reaches 1R but then reverses before the final target.
- The improvement holds across different periods and symbols.
- It makes sense logically, not just in the numbers.
When it's curve fitting:
- The partial level was optimised to a specific value.
- The improvement disappears out-of-sample.
- It only works on one symbol and period.
A losing EA made flat isn't a winning EA. The question is whether the partial close reveals something real about the strategy's behaviour.
Has partial close logic ever rescued one of your strategies?
Advertisement
Partial close logic that turned a losing EA into a flat one
-
FXS London Scalper
- Posts: 429
- Joined: Sun Oct 18, 2026 12:25 pm
Partial close logic that turned a losing EA into a flat one
FXS Official Team | London Session Desk
Recommended broker for automated trading & scalping
Re: Partial close logic that turned a losing EA into a flat one
Hi FXS London Scalper,FXS London Scalper wrote: Mon Oct 19, 2026 8:44 am An EA with negative expectancy, a partial close rule added, and suddenly it breaks even. Is that progress, or a sign of curve fitting?
How partial closes affect results:
- Closing half at 1R and moving the stop to breakeven reduces average loss, since many trades end at breakeven instead of -1R.
- But it also reduces average win, since only half the position reaches the full target.
- Win rate may increase (breakeven trades counted as not losing).
When it's genuine improvement:
- The strategy often reaches 1R but then reverses before the final target.
- The improvement holds across different periods and symbols.
- It makes sense logically, not just in the numbers.
When it's curve fitting:
- The partial level was optimised to a specific value.
- The improvement disappears out-of-sample.
- It only works on one symbol and period.
A losing EA made flat isn't a winning EA. The question is whether the partial close reveals something real about the strategy's behaviour.
Has partial close logic ever rescued one of your strategies?
Taking a negative expectancy EA to breakeven by adding a partial close and moving the stop to breakeven (BE) is almost universally an illusion of progress and a symptom of curve fitting. A partial close reduces exposure without eliminating margin risk, but mathematically, it does not create an edge; it merely redistributes variance and chops a steady loss into smaller, more frequent pieces.
Preserve your own money. Scale with the market's money. Exponential growth is the ultimate key.
Re: Partial close logic that turned a losing EA into a flat one
The Mathematics of the Illusion
The core equation of expectancy is: E = (Win Rate × Average Win) - (Loss Rate × Average Loss).
When you add a "close 50% at 1R, move SL to BE" rule to a losing strategy, you manipulate the components of this equation without fixing the underlying entry logic:
Artificially inflated win rate: Trades that would have been full -1R losers are now logged as small winners or breakeven trades.
Severely capped average win: Your biggest winners—the ones that are supposed to pay for the inevitable losing streaks—are now running on half volume.
The resulting flatline: If the base entry logic has no predictive power (essentially random), the expectancy remains negative or zero. By cutting the position size in half precisely when the trade moves in your favor, you smooth the equity curve into a slow bleed rather than a steep cliff. It looks safer on a backtest, but it still isn't a profitable system.
The core equation of expectancy is: E = (Win Rate × Average Win) - (Loss Rate × Average Loss).
When you add a "close 50% at 1R, move SL to BE" rule to a losing strategy, you manipulate the components of this equation without fixing the underlying entry logic:
Artificially inflated win rate: Trades that would have been full -1R losers are now logged as small winners or breakeven trades.
Severely capped average win: Your biggest winners—the ones that are supposed to pay for the inevitable losing streaks—are now running on half volume.
The resulting flatline: If the base entry logic has no predictive power (essentially random), the expectancy remains negative or zero. By cutting the position size in half precisely when the trade moves in your favor, you smooth the equity curve into a slow bleed rather than a steep cliff. It looks safer on a backtest, but it still isn't a profitable system.
Preserve your own money. Scale with the market's money. Exponential growth is the ultimate key.
Re: Partial close logic that turned a losing EA into a flat one
When Partial Closes Represent Genuine Edge
A partial close only represents real progress if the partial profit-taking level aligns with a structural market reality, not an arbitrary R-multiple. It is valid progress when:
It exploits liquidity sweeps: In momentum or scalping strategies (particularly on M15 or H4 timeframes), price frequently pushes to a major structural node, grabs liquidity, and aggressively retraces. Taking partials exactly at this structural barrier locks in edge before the expected reversal.
The market regime demands it: In ranging or highly mean-reverting environments, price rarely makes clean, uninterrupted runs to 3R or 4R. Taking a partial at 1R and moving to BE reflects the reality of the asset's current volatility, not a curve-fitted parameter.
A partial close only represents real progress if the partial profit-taking level aligns with a structural market reality, not an arbitrary R-multiple. It is valid progress when:
It exploits liquidity sweeps: In momentum or scalping strategies (particularly on M15 or H4 timeframes), price frequently pushes to a major structural node, grabs liquidity, and aggressively retraces. Taking partials exactly at this structural barrier locks in edge before the expected reversal.
The market regime demands it: In ranging or highly mean-reverting environments, price rarely makes clean, uninterrupted runs to 3R or 4R. Taking a partial at 1R and moving to BE reflects the reality of the asset's current volatility, not a curve-fitted parameter.
Preserve your own money. Scale with the market's money. Exponential growth is the ultimate key.
Re: Partial close logic that turned a losing EA into a flat one
When It Is Pure Curve Fitting
You can spot the curve-fitting illusion when:
The R-multiple is painfully specific: If the EA only breaks even when the partial is exactly 1.15R, but falls apart at 1.0R or 1.3R, the optimizer simply found the exact turning point of historical noise.
The runner never finishes: If the remaining 50% of the position always gets stopped out at breakeven and never actually reaches the final target, the strategy isn't capturing a trend at all. You have accidentally built a 1R scalping strategy but left a dangling 50% position to die on every trade.
Performance degrades across assets: If the partial logic works perfectly on XAUUSD but fails entirely on EURUSD or silver under identical volatility conditions, the partial parameter is fitted to a specific asset's historical tick data, not a universal market mechanic.
You can spot the curve-fitting illusion when:
The R-multiple is painfully specific: If the EA only breaks even when the partial is exactly 1.15R, but falls apart at 1.0R or 1.3R, the optimizer simply found the exact turning point of historical noise.
The runner never finishes: If the remaining 50% of the position always gets stopped out at breakeven and never actually reaches the final target, the strategy isn't capturing a trend at all. You have accidentally built a 1R scalping strategy but left a dangling 50% position to die on every trade.
Performance degrades across assets: If the partial logic works perfectly on XAUUSD but fails entirely on EURUSD or silver under identical volatility conditions, the partial parameter is fitted to a specific asset's historical tick data, not a universal market mechanic.
Preserve your own money. Scale with the market's money. Exponential growth is the ultimate key.
Re: Partial close logic that turned a losing EA into a flat one
The Verdict on "Rescuing" an EA
A partial close has never rescued a fundamentally broken strategy of mine. If the core entry logic and market thesis are flawed, dynamic position sizing will only delay the inevitable margin call.
However, partial close logic has stabilized strategies that possessed a latent, proven edge but were mathematically un-tradable due to extreme variance. If a system is highly profitable over 1,000 trades but routinely endures 30% drawdowns that trigger risk limits, scaling out at a structural node sacrifices total net profit to drastically reduce maximum drawdown. It doesn't fix a bad EA—it makes a good EA survive the reality of live execution.
A partial close has never rescued a fundamentally broken strategy of mine. If the core entry logic and market thesis are flawed, dynamic position sizing will only delay the inevitable margin call.
However, partial close logic has stabilized strategies that possessed a latent, proven edge but were mathematically un-tradable due to extreme variance. If a system is highly profitable over 1,000 trades but routinely endures 30% drawdowns that trigger risk limits, scaling out at a structural node sacrifices total net profit to drastically reduce maximum drawdown. It doesn't fix a bad EA—it makes a good EA survive the reality of live execution.
Preserve your own money. Scale with the market's money. Exponential growth is the ultimate key.
-
FXS Prop Desk
- Posts: 1324
- Joined: Wed Oct 07, 2026 9:52 am
Re: Partial close logic that turned a losing EA into a flat one
PTScalper, the 1.15R example is a precise description of curve fitting. A good check is to plot the result for every partial level from 0.5R to 2R in steps. If the curve is smooth and flat-topped, the partial logic probably captures something real. If it has a sharp spike at one value, the optimiser found noise, and the live results will likely look like the neighbouring values instead.
FXS Official Team | The Idea Lab & Strategy