⚠️ Personal research and trading journal — not investment advice. The author does not provide licensed advisory services.
I discovered by accident that changing a sell rule — not an entry condition, not a filter — turned a walk-forward test from FAIL to PASS.
The entry logic was identical. Only what happens after you're in the trade changed.
Background: Two Backtest Modes
When I test my trading systems, I run two modes:
Same-day (optimistic): entry fills at the exact close price on the signal day. No slippage, no waiting.
Realistic (conservative): entry fills at the next day's open price. This accounts for the fact that signals are generated after the close, and real trades happen the following morning.
The realistic mode is harder to pass. Buying the open after a breakout day means paying more, especially when the stock gaps up on strong momentum. The price difference between "close on day X" and "open on day X+1" sounds small, but across hundreds of trades over years, it significantly degrades edge.
The Test
I ran the first_pullback system in walk-forward mode across 6 years (2020-2025), 6 windows, with two configurations:
Configuration A: Same exit rules — fixed stop at 7% below entry, fixed target at 25% gain. Take-profit all at once.
Configuration B: Partial take-profit at 2R (take half off), trail the remaining half on a 21-day moving average exit.
In same-day mode: both configurations produced positive results in walk-forward.
In realistic (next-open) mode: - Configuration A (fixed exit): FAILED. Negative walk-forward over the test period. 3 of 6 windows produced losing R. - Configuration B (partial-TP 2R + MA21 trail): PASSED. Positive walk-forward. 5 of 6 windows positive.
Same system. Same entries. Same stocks. One sells everything at once. One takes half off at 2R and trails the rest. Difference: one passes a 6-year walk-forward, one fails it.
Why the Partial-TP Fixes the Realistic-Mode Problem
The realistic-mode problem is a timing problem. When you buy the open after a breakout day, you're buying after the initial momentum is already priced in. The gap between prior-day close and next-day open acts like a form of slippage.
The fixed-exit system compounds this: you set a target at +25%, but your effective entry cost is already +1-3% higher than the signal price. Your actual target from open is +22-24%, which sounds close — but when multiplied by the stop at -7%, the reward/risk ratio degrades meaningfully. The edge in the system is thin enough that this degradation pushes the realistic-mode walk-forward from marginal-pass to fail.
The partial-TP system handles this better because:
2R is a lower target: selling half at 2R (entry risk × 2) means you're securing profit on a smaller move than the fixed 25% target. The 2R target at a 7% stop is +14% from entry — achievable even after a +3% overnight gap.
The trail captures the extension: the remaining half isn't capped. If the stock extends to +40%, the MA21 trail lets you ride it. You don't sacrifice the big winners to make the small ones viable.
Variance reduction improves WF stability: the partial-TP creates a more consistent return distribution. Fewer full-size exits at maximum loss, fewer full-size exits at targets that were just barely missed. The realized R profile is smoother, which stabilizes the walk-forward windows.
What This Means Practically
The partial-TP isn't just "a little better." In this system, it's the difference between a strategy with a validated walk-forward and one without.
I tested this before the 2026 live launch decision (which moved to January 2027). The validated configuration uses partial-TP at 2R / MA21 trail. The fixed-exit configuration is not in production.
This finding connects directly to the IBD run-up study I ran separately (see "IBD Said Buy. 91% Moved Higher. Most Gave It All Back."). That study, using a different dataset (IBD video mentions rather than backtest entries), found that: - 91% of IBD actionable-buy mentions showed positive max run-up - But t+120d mean returns were negative
The interpretation is the same in both cases: the upside of momentum trades is real and common. But if you hold all of it until a single exit, mean reversion eats the gains. Taking half off early locks in the part of the distribution that actually exists. The rest you let ride.
The Lesson About Walk-Forward Testing
This also illustrates something about how walk-forward tests can be used correctly: as a diagnostic on whether an edge is real or fragile under realistic conditions.
In same-day mode, both configurations passed. I might have shipped either one if I'd only tested optimistically. The realistic mode revealed that one configuration was riding on favorable entry assumptions, while the other had genuine resilience to those constraints.
When a strategy only works with ideal entry prices, it's not a robust strategy. The partial-TP didn't create edge — it revealed which configuration actually had edge under realistic conditions.
Track. Study. Wait. Strike.
Personal research and trading journal — not investment advice. The author does not provide licensed advisory services. — MOEasymmetry
Draft 2026-06-12. Source: first_pullback system, walk-forward on US data 2020-2025, 6 windows. Configuration A (fixed 25% target): FAIL in realistic mode. Configuration B (partial-TP 2R + MA21 trail): PASS in realistic mode. Same-day mode: both pass. Difference = entry price (same-day close vs next-day open). Configuration B is the locked production configuration.