Your strategy exits at its profit target, and the market continues for another hour. The obvious reaction is that the target was too close. Sometimes that is true. Sometimes the extra move was available only after a large reversal that would have stopped the trade out. Exit analysis becomes useful when it distinguishes favorable movement, executable profit and information that was available only after the fact.
Start with a limited question: how much of the movement observed during a defined tracking window did the strategy capture? Then ask the harder question: would a different executable exit policy have improved the complete strategy, including losses, costs, overlapping trades and capital constraints? The first is a diagnostic. The second requires a simulation of actual rules.
Define the trade before measuring the excursion
For a long position, favorable excursion is the price movement above entry, and adverse excursion is the movement below entry. For a short position, reverse the signs. Maximum favorable excursion, or MFE, is the greatest favorable open movement during the observation interval. Maximum adverse excursion, or MAE, is the greatest adverse open movement. A common convention reports both as nonnegative magnitudes.
Long MFE = max(0, highest observed price − entry price)
Long MAE = max(0, entry price − lowest observed price)
Short MFE = max(0, entry price − lowest observed price)
Short MAE = max(0, highest observed price − entry price)Those formulas describe price excursions. Converting them to money requires quantity and contract value. Converting them to R requires a valid initial risk definition. If risk is entry-to-stop distance times quantity and value per point, keep that initial risk fixed for the comparison. Replacing it with a later, tighter stop changes the meaning of every R-multiple.
You also need a valuation convention. Last-trade highs and lows are not necessarily executable bid and ask prices. A long position liquidates by selling, so available bids matter. A price-only excursion based on bar extremes can still be informative, but it must not be represented as guaranteed realizable profit.
A trade with two different observation windows
Consider a hypothetical long entry at 100 with an initial stop at 98 and a target at 104. Initial price risk is two points. Before the trade exits at 104, its lowest observed price is 99 and highest is 104. Ignoring costs and assuming the stated fill, realized profit is two R, MFE is two R and MAE is half an R.
Now continue virtual tracking until a preselected session close. After the actual exit, price reaches 107 before ending at 105. The largest movement from entry in this extended window is seven points, or 3.5R. The additional favorable movement beyond the target is three points, or 1.5R. These are post-exit diagnostics, not profits the original trade earned.
| Measure | Price points | Initial R |
|---|---|---|
| Initial risk | 2 | 1.0 |
| Realized target gain | 4 | 2.0 |
| MFE during the actual trade | 4 | 2.0 |
| MAE during the actual trade | 1 | 0.5 |
| Maximum favorable movement through tracking end | 7 | 3.5 |
| Additional favorable movement after target | 3 | 1.5 |
The target captured 4/7, or 57.14%, of the maximum entry-relative favorable movement in the extended window. During the original holding period, it captured 100% of the measured MFE. Both statements are arithmetically correct because their denominators describe different windows. A report that labels both simply “capture ratio” invites confusion.
Choose a tracking end before looking at the path
Possible endpoints include the original stop, session close, the next signal or a fixed number of bars. Each answers a different question. Tracking until session close asks about the remainder of that session. Tracking for twenty bars asks about a fixed strategy-timeframe horizon. Tracking until the next signal relates the opportunity to the strategy's own event process.
If the original stop is an endpoint, stop the virtual path when it is first reached. Do not count a later rebound as available continuation after that event. When multiple endpoints apply, define which occurs first and how simultaneous events are resolved. A precise stopping rule is more valuable than a visually impressive extension curve.
For strategies without an explicit original target or stop, do not invent one and call it historical fact. You can measure entry-to-exit excursions and define a hypothetical risk unit or alternative target for a new experiment. But the report must distinguish a supplied rule from an inferred or user-selected assumption.
Capture ratio is not a standalone optimization objective
A strategy can achieve a high capture ratio by taking very small opportunities and avoiding every difficult trade. It may still have poor net expectancy. A low ratio can accompany a sensible exit that gives up exceptional continuation in exchange for a more repeatable result. Compare capture with realized profit, downside exposure, turnover and the full distribution of outcomes.
Be careful with zero denominators. If the measured favorable excursion is zero, a capture ratio based on dividing by MFE is undefined. Do not replace it with zero or 100% without explanation. Negative realized outcomes require a stated convention as well. Restricting the ratio to target exits may be useful, but then the denominator is the number of eligible target trades, not all trades.
Average ratios and ratios of totals are different. A trade capturing one point out of two has a 50% ratio. Another capturing ten out of ten has 100%. Their average individual ratio is 75%, while total captured movement divided by total available movement is 11/12, or 91.67%. Choose the statistic that matches the question and label it explicitly.
Target reach probability needs event order
To estimate the fraction of entries reaching a proposed target before the stop, follow each price path until the first relevant event. Compare several predeclared target multiples, such as 1R, 2R and 3R. Keep the stop, time limit, eligible entries and price-resolution assumptions consistent so the comparison isolates target distance.
A bar can contain both target and stop. OHLC data reveal the high and low but not their order. If a long bar trades below the stop and above the target, assuming the target always occurred first biases results. Use finer information where available or a documented conservative and optimistic bound. An ambiguous path is not a license to choose the favorable sequence.
CME's order documentation also distinguishes a trigger from execution. A stop-limit order can be activated without being filled at its limit. A touched resting target does not automatically guarantee your queue position receives a fill. The simulator should reflect the order type and the available data rather than treating every plotted price as executable.
From reach probability to a target frontier
In a simplified experiment, suppose all trades resolve at either a target or a one-R stop, there are no time exits and round-trip costs equal 0.05R. If 1R is reached first on 60% of entries, expectancy is 0.60 times 1 minus 0.40 times 1 minus 0.05, or 0.15R. If 2R is reached first on 43%, expectancy is 0.43 times 2 minus 0.57 minus 0.05, or 0.24R.
| Target | Hypothetical reach rate | Net expectancy |
|---|---|---|
| 1R | 60% | 0.15R |
| 2R | 43% | 0.24R |
| 3R | 30% | 0.15R |
These rates are invented inputs to a teaching calculation, not measured market probabilities. The example shows why higher win rate is not automatically better and why a larger target can eventually reduce expectancy. A real frontier must also calculate time exits, variable costs, position sizing and chronological drawdown.
Why replaying isolated trades is not a full strategy rerun
Changing an exit can change the next entry. A longer holding period may overlap a signal the original strategy would have traded. A shorter hold can free capital earlier. Rules such as one open position, daily loss limits and maximum exposure mean that independently modifying each historical trade may create a portfolio that could never have existed.
Use isolated path analysis to screen ideas, then rerun the complete strategy state machine for serious comparisons. The rerun should regenerate entries when they depend on position state and apply costs to the actual resulting fills. Otherwise you are measuring a conditional diagnostic on historical entries, not the performance of a deployable alternative.
Partial exits introduce another layer. Closing half at 1R and half at 3R gives a gross 2R result only if both portions reach those targets. If the remainder stops at the original stop, the combined gross result is zero R. Additional fills can add costs, and whole-contract constraints may make a fifty-percent split impossible for a one-contract position.
Trailing and time exits need information discipline
A trailing stop must be calculated from information available before the execution decision. Updating the trail using a bar's high and then assuming a stop fill earlier within that same bar can introduce hidden look-ahead. Define whether the trail updates intrabar, on completed bars or at another checkpoint, and use a compatible execution model.
The same discipline applies to ATR-based distances. ATR must use the intended completed observations, not future volatility. A time exit needs a precise clock: elapsed minutes, completed bars, trading-session time or a deadline. Twenty one-minute bars may span more than twenty clock minutes if the series contains a session break or missing observations.
Validate the alternative on separate periods
Select the alternative using development data, then lock its rules before evaluating validation. Keep the final holdout separate until the intended final check. If the holdout has already informed the choice, its result remains useful history but is no longer independent evidence for that choice.
Plot each period on its own clearly defined starting capital, or chain nonoverlapping realized returns chronologically under one capital policy. Do not splice independently rebased equity levels together as if they were one continuous account. A jump caused by resetting capital is not trading profit, and overlapping test periods must not be counted twice.
The practical output is a comparison of executable policies, not a promise to capture every historical high. Keep the original exit as a baseline. Report net expectancy, drawdown, holding time, trade count and sensitivity to execution assumptions. An improvement that survives those checks is more meaningful than a dramatic amount of theoretical uncaptured movement.