What does each exit do to the same entries?
A fixed target closes every winning trade at the same distance and leaves anything beyond it on the table, while a trailing stop follows price at a set distance and closes the trade when price retraces by that amount, capturing more of the long moves and giving back part of every move.
Under a fixed target, the distribution of winning trades has a hard ceiling. Every winner is worth the same number of R, the win rate is whatever fraction of trades reach that distance, and the expectancy is a simple product of the two. The strategy's result is therefore decided by how often price travels the target distance before travelling the stop distance.
Under a trailing stop, the winners have no ceiling and no floor above the trail distance. A few trades capture large moves; many are closed for a small gain or a small loss when the first meaningful pullback arrives. The win rate is typically lower than under the fixed target, the average winner is larger, and the result depends on the tail of the distribution rather than on its centre.
Neither is better in the abstract. They are two ways of sampling the same distribution of post-entry moves, and which samples it more profitably is a property of that distribution.
Why is the trail so often flattered in a backtest?
Trailing stops are flattered because most testers move the stop using the bar's high before checking whether the bar's low would have hit the original stop, which lets a favourable move that happened after the stop-out rescue a trade that had already lost.
The mechanism is an ordering assumption inside the bar. On any bar, the tester knows the high and the low but not which came first. The convenient implementation raises the trailing stop to reflect the high, then tests the new stop against the low. If the low actually came first, the trade was stopped at the old level, and the tester has booked a smaller loss or a gain that never happened. Over a history, that assumption resolves every ambiguous bar in the trail's favour.
The honest ordering is the reverse: test the existing stop against the bar's adverse extreme first, and only if the trade survives that does the favourable extreme move the stop. A bar whose low would have hit the stop ends the trade at the stop, whatever the high did.
The size of the difference between the two implementations is large for tight trails on low timeframes, where ambiguous bars are frequent, and it is the reason two backtests of the same trailing rule can disagree by the whole of the strategy's apparent edge.
How do you compare the two fairly?
Compare them on the same entries over the same history with the same costs, resolve the trail pessimistically, grade both in R, and look at the distribution of each trade's maximum favourable excursion to understand why one exit did better rather than only that it did.
The maximum favourable excursion is the furthest a trade moved in its favour before it ended, and its distribution across all the entries is what both exits are sampling. If most trades reach one and a half R and few go further, a fixed target near that level captures nearly all the available profit and a trail gives much of it back. If the distribution has a long tail of trades that run five or ten R, the fixed target leaves that tail on the table and the trail collects it.
The comparison is therefore not a contest between exits so much as a reading of the distribution. Knowing the shape tells you which exit fits it, and it tells you something about the entries as well: a rule whose winners rarely run far is a reversion rule whatever it was designed to be.
Costs enter differently for each. The trail closes more trades for small amounts, and each of those pays the spread, so a trail on a low timeframe can lose to costs a comparison that it would win before them.
When does each tend to win?
The trail tends to win when the market's moves persist beyond the point at which they began, so that a trade that has gone one R in its favour is more likely to continue than to reverse, and the fixed target tends to win when moves are mean-reverting and the first target-sized move is usually the whole move.
Persistence is a property of the instrument and the timeframe rather than of the exit. Some markets and periods trend cleanly enough that a trail collects a small number of large winners which pay for many scratches; others chop, and every trail is closed by the first pullback for a small loss after the target was already available.
The trail distance is a parameter with the same grid-and-plateau logic as any other. A tight trail locks in small gains and is closed by ordinary noise; a wide trail gives back more of every move and survives more pullbacks. The setting that suits a distribution is again a property of that distribution.
Hybrid exits, such as a fixed target for part of the position and a trail for the rest, or a trail that activates only after a stated gain, are further rules with further parameters. Each is a legitimate strategy and each needs its own grade.
How does QuantParadox grade exits?
QuantParadox grades trailing stops, break-even moves, time stops and partial exits with a pessimistic intrabar rule: on every bar, the existing stop is tested against the adverse extreme before any favourable movement is allowed to move it, so a within-bar favourable move can never rescue a trade the adverse move had already ended.
That ordering is the correction for the flattering implementation described above, and it is applied to every managed exit, not only to the simple trail. Where the archive has finer data for the bar in question, the actual order of events is used instead of the pessimistic assumption, and the trade is labelled by which of those resolved it.
The limitation is that the trail moves once per bar on the entry timeframe unless finer data is being consulted, so a trail that in live trading would ratchet several times inside an hourly bar is graded as moving once, at the pessimistic point. That makes some trailing results on the platform worse than a tick-by-tick simulation would show, and better than the flattering bar-based one.
The per-trade excursion figures are recorded, so the distribution that decides the comparison between exits is available for any run, not only the exits' aggregate results.