Exit Strategy Design: 50 EMA Trail, 5R Breakeven, No Partial

A systematic comparison of 78 trade-management configurations, evaluated on a 100,000-trade stratified sample and validated on the full 2,803,855-trade backtest. This report explains the three choices that govern our trade exits — the trailing stop, the breakeven trigger, and whether to take partial profits — and the evidence behind the configuration we selected.

Project · Savell Momentum Playbook Dataset · 2020-03 → 2025-03 Tickers · 3,133 Entries · Buy stop above pivot
Backtest disclosure. All figures in this document are hypothetical results from simulated trades on historical price data. They do not represent actual trading and no client money was managed. Past performance — simulated or actual — does not guarantee future results. This report is methodology documentation, not investment advice or an offer to manage funds.
Mean R · Backtest
+0.385
vs +0.070 under prior rules
Configurations Tested
78
2 trails × 13 triggers × 3 partials
Winner Mean R
+4.15
24.4% of simulated trades
Selected Configuration
50 EMA · 5R · No partial
Best in every paired cell

1The question

Every trade-management decision in a breakout system reduces to three axes: which moving average to trail, when to move the stop to breakeven, and whether to take a partial profit. Our prior rules chose the tightest valid EMA (8→13→21→50 cascade), moved to breakeven at 1R, and scaled half at 1R. In backtest they produced a mean R of +0.070 per trade. This report asks whether those three choices are individually and jointly correct, and quantifies the change from the alternatives we tested.

·Sweep design 100,000-trade stratified sample · 2.8M-trade rebuild
Trail axis (2)

tightest — cascade 8→13→21→50, exit on close below the tightest EMA the stock is still holding.

ema50 — fixed 50 EMA; exit only on close below it.

Trigger axis (13)

0.5 → 5.0R in 0.25R / 0.5R steps, plus never (trail only, no BE). Breakeven only engages after the trigger is tagged.

Partial axis (3)

none — full position held until trail fires.  half_1r — sell 50% at +1R, trail remainder.  half_2r — sell 50% at +2R, trail remainder.

2The 50 EMA trail outperforms the tightest-EMA trail in every configuration

Pair every trigger/partial cell across the two trail choices. The 50 EMA row has a higher mean R than the tightest-EMA row in all 39 matchups. The gap widens with the trigger: at a 5R breakeven with no partial, the 50 EMA trail produces +0.4145R vs +0.3766R for the tightest-EMA cascade — a 10% relative edge on the best cell.

·Mean R vs breakeven trigger — both trail types, partial = none
Why a wider trail helps. A tightest-EMA cascade exits quickly when momentum cools. In a system with a capped left tail (losers stop at −1R) and an uncapped right tail, every R of compression on the right tail is net cost to expectancy. The 50 EMA sits far enough below price that normal consolidation doesn't trigger it. Winners exit at a mean of +1.6R under the tightest-EMA trail vs +2.8R under the 50 EMA trail.

3Holding the full position outperforms partial profit-taking in every configuration

Hold trail and trigger constant. Compare no partial vs half at 1R vs half at 2R across all 26 cells. No partial produces a higher mean R in every single pairing. Taking half the position off at 1R or 2R caps upside in a distribution where roughly 10% of trades produce the majority of the total R.

·Mean R by partial strategy — 50 EMA trail, across BE triggers
Why partial profit-taking reduces expectancy. The return distribution in this backtest is heavy-tailed. The 95th-percentile winner produces +5.4R; the 99th percentile produces +9.9R. Selling half the position at 1R monetizes only 0.5R from those shares. On a +10R winner, that converts the trade to +5.25R — a 47.5% reduction. Because a small number of outlier winners drive the majority of positive expectancy, clipping them is a direct cost to the edge.

45R is the strongest breakeven trigger tested — and removing the trigger is worse

Hold the trail (50 EMA) and partial (none) constant. Sweep the breakeven trigger from 0.5R up to 5.0R, plus a never option where no breakeven stop is applied. Mean R rises monotonically from +0.091 at 0.5R to +0.414 at 5R, then drops to +0.111 when no breakeven trigger is used. That drop is why a breakeven rule matters at all: without one, the 50 EMA trail catches weak trades that never developed and gives back the edge through the left tail.

·Mean R as a function of BE trigger — 50 EMA trail, no partial
The mechanism. At low triggers (0.5–1R), the breakeven stop fires on a large share of trades that pulled back briefly before advancing — it exits winners mid-development. With no trigger at all, the trail fires on trades that grazed a bit of profit and then reverted — losing money on positions that never earned the right to be held. 5R sits between these two failure modes: the trade has proven itself, initial risk has been earned back several times over, and the trail can finish the job.
Note on 5R as the selected level. 5R is the highest trigger in our tested range; the true maximum may sit somewhere between 5R and no-trigger. Additional testing at 6R, 7R, 8R would sharpen that boundary. The decision to use 5R reflects diminishing marginal gains above 3.5R (the improvement per 1R step collapses from +0.063 at 3R→4R to +0.050 at 4R→5R) and the practical cost to the trader of watching large gains drift back.

5The full sweep matrix

Every cell tested. Green row is the selected configuration; red row is the prior configuration used by the system before this review. Every step from red to green — widening the trail, delaying the breakeven, removing the partial — moves mean R in the same direction.

·All 78 configurations · mean R on 100K-trade sample
Trail BE Trigger None Half @ 1R Half @ 2R

6Full-scale validation — 2.8 million simulated trades

The sweep used a 100,000-trade stratified sample to make 78 configurations tractable. The selected configuration was then rebuilt against the full backtest: 2,803,855 simulated trades. The change raised mean R from +0.070 (prior rules) to +0.385 (selected rules). Almost all of the improvement comes from the right tail of the winner distribution — trades exited by the 50 EMA trail after reaching 5R average +5.25R and a median hold of 150 days.

·Prior configuration vs selected configuration
Metric Prior Selected Δ
Mean R per trade+0.070+0.385+0.315
Median R−0.16−1.00heavier tails
Trades reaching 2R12.8%21.7%+8.9pp
Winner mean R+1.66+4.15+2.49R
Winner median hold~50d169d+119d
Loser median hold~14d12d
Trail-exit frequency40.2%15.4%fewer, larger
·Exit-reason distribution
Outcome % Mean R Med Days
Initial stop hit51.7%−1.0528
Failed breakout23.1%−0.371
50 EMA trail exit15.4%+5.25150
Position still open9.8%+2.09234
Breakeven stop hit0.1%−0.2630

7The return distribution — heavy-tailed by design

This is not a comfortable equity curve. In the backtest, 75.6% of trades end in a loss, median R is −1.00, and most trades cluster at the full-stop outcome. Positive expectancy comes from the right tail: roughly the top 10% of trades produce about 90% of total R. That concentration is the mathematical reason partial profit-taking is costly — cutting the winners down to size directly reduces the tail that produces the edge.

·R-multiple distribution · 2.8M simulated trades
R-multiple percentiles
p05−1.06
p25−1.00
p50 (median)−1.00
p75−0.06
p90+4.14
p95+5.42
p99+9.92
Winners vs Losers
Winners (R>0)24.4% · N=683K
  Mean R+4.15
  Median hold169 days
Losers (R≤0)75.6% · N=2.12M
  Mean R−0.83
  Median hold12 days

8Model grades on out-of-sample data

We train a gradient-boosted model to predict which simulated trades will reach 5R based on setup quality, fundamentals, and market context. The model produces a conviction score from 0 to 1, which we bucket into letter grades. Apply these grades to out-of-sample trades only (N = 466,895, dates on or after 2025-03-05). Grade performance is monotonic: every higher conviction bucket has a higher mean R, higher positive-R rate, and higher 5R-reach rate than the bucket below. A+ isolates a tier with +3.19R mean and 54% positive-R on roughly 0.24% of the out-of-sample population.

·Backtest performance by model grade · out-of-sample
Bucket N Mean R R>0 Hit 5R Winner Med Days
A+ (≥0.40)1,111+3.1954.0%38.3%130
A (0.25–0.40)23,060+1.6141.5%23.3%110
B (0.20–0.25)44,572+0.9535.1%18.7%104
C (0.15–0.20)98,946+0.7133.3%16.3%102
D (<0.15)299,206+0.2831.3%9.8%103
Flagged (≥0.20)68,743+1.2137.6%20.6%107
All466,895+0.5132.6%12.8%104
How grade and exit rules compound. The exit rules produce a mean of +0.385R across all unfiltered simulated trades. Restricting to the Flagged bucket (conviction ≥ 0.20) raises that to +1.21R on roughly 2.5% of the trade population. Restricting to A+ raises it to +3.19R on roughly 0.24% of the population. Selectivity and trade management stack — they solve different problems and their benefits don't overlap.

9What to expect from this approach

These exit rules are designed around infrequent, large winners rather than frequent, small ones. That shape has specific characteristics anyone using or following the system should understand before drawing conclusions from short-run results.

1 · Most trades are small losers

Roughly three out of every four simulated trades end at or near the initial stop. The feedback loop is sparse: statistical confirmation that the system is working only arrives when the occasional fat-tail winner catches a long move. Short-term samples look discouraging even when the long-run math is sound.

2 · Winners are held for months

The median winner holds 169 days. The 75th-percentile winner holds 295 days. During that hold, price will regularly pull back 10–15% without triggering the 50 EMA stop. The system requires tolerance for substantial unrealized-gain volatility between entry and eventual exit.

3 · Some open gains are given back

Without a partial, open gains are given back on the exit day when the 50 EMA is breached — a trade that ran to +7R can close at +2.5R. The prior half-at-1R rule locked in a more comfortable median exit; the current rule gives up that comfort for higher mean R per trade in backtest.

A note on execution. The benefit quantified in Section 3 depends on keeping the full position through the trailing stop. Any discretionary decision to take profits at 2R or 3R "because it feels safer" reintroduces the cost shown there. In a heavy-tailed distribution, the discipline of the exit rule is the primary source of the edge — the entry signal determines which trades are taken, but the exit rule determines how much each one is worth.

10Summary

Three statistical findings, each independently verified on the 78-configuration sweep and jointly verified on the 2,803,855-trade backtest rebuild:

  • The 50 EMA trail produces a higher mean R than the tightest-EMA trail in all 39 paired configurations. A wider trail preserves outlier upside; a tighter trail compresses the right tail.
  • Holding the full position produces a higher mean R than taking half profits at 1R or at 2R in all 26 paired configurations. In a heavy-tailed distribution, partial profit-taking reduces the contribution of the outlier winners that drive most of the expectancy.
  • Mean R rises monotonically from +0.09 at a 0.5R breakeven trigger to +0.41 at 5R, then drops to +0.11 when no breakeven trigger is used. 5R sits between the two failure modes: aggressive triggers exit developing trades, and no trigger lets the trail fire on trades that never earned the right to be held.

The selected configuration — 50 EMA trail, 5R breakeven trigger, no partial — is therefore the strongest in our testing along each axis independently and jointly. It is not a single local optimum but the combination of three individually dominant choices.

In backtest, the change raises mean R per trade from +0.070 under the prior configuration to +0.385 under the selected one. Restricting to trades the model flags (conviction ≥ 0.20) raises the out-of-sample mean R further, to +1.21. These are backtest results on simulated trades; live performance has not been established and actual results will differ.

Published 2026-04-19 · Savell Trading Playbook · savell.io Methodology document · v1.0

Important disclosures. The figures in this document are hypothetical results derived from a backtest simulation applied to historical price data. They do not represent the performance of any real trading account, and no client funds were managed in producing them. The simulation uses assumptions about entry price, slippage, and execution that may not match real-market conditions.

Past performance, whether hypothetical or actual, is not a reliable indicator of future results. Backtest performance in particular is subject to overfitting, look-ahead bias, selection of the historical period, and the specific design choices documented here; forward performance will differ, often materially.

This document is methodology documentation for a personal trading system operated by the author. It is not investment advice, a recommendation to buy or sell any security, an offer to manage money, or a solicitation of any kind. Nothing here should be relied upon to make investment decisions. Readers should consult a qualified financial professional regarding their own circumstances.

Data sources: U.S. equity OHLCV and fundamentals via Financial Modeling Prep (FMP). Backtest and model training code maintained by the author.