Where the money actually is
Every conversation about this market is about what to buy. Almost none is about when to leave. The panel is blunt about which of the two decisions carries the money.
Take 204,136 sampled entries across the panel. Every fifth bar of every symbol, 2016-2026, and follow each one for 20 sessions:
| Over the next 20 sessions | Median | Mean |
|---|---|---|
| Best price reached (the peak) | +7.54% | +14.14% |
| Price at the end (the close) | 0.00% | +2.94% |
[MEASURED 2026-07-31]
The gap between them is 7.54 percentage points of median return that existed, was available, and was handed back.
This is not a ZEME finding about ZEME picks. It is a property of the market itself, at every single price, on average. Almost every position you have ever held was, at some point, up more than it was when you sold it.
The trap this sets
The obvious response is: "then use a stop-loss and a target." The obvious response has been tested here repeatedly and it has been expensive:
- A blanket -15% stop on 339 picks over 12 weeks turned a +4.59% median hold into -15.00%, stopping out 171 of 339 positions and cutting the win rate from 57% to 42%. [LEDGER:
project_exit_rules] - The ATR(14) x 1.5 stop that was actually shipped to customers fired on about 70% of positions. On one model it converted +35.25% into -4.01%. [LEDGER:
project_exit_rules] - A stop at the first support level, a sensible-sounding idea, sits about 8.8% away and fires on 73% of positions. [LEDGER:
project_models_adopted]
Every one of those was proposed for exactly the reason you are thinking of right now, and every one of them was killed by measurement. The peak-to-close gap is real, and the naive ways to capture it are worse than doing nothing.
What survives
Only two things survive the sweep: a wide stop that rarely fires, and partial exits taken into strength rather than one all-or-nothing decision. The measured sweep put no stop at +18.30% and a flat -25% at +15.65%, while every tight variant went negative. [LEDGER: project_models_adopted]
The real lesson is not "use stop X". It is: the exit is the highest-value unsolved problem in your process, the gap is 7.54 points wide, and the cheap fixes have already been tried and have already failed.
Evidence · History · Confidence · Counterargument
Evidence, n=204,136 entries, 20-session horizon, whole panel. [MEASURED 2026-07-31]
History, the gap widens with horizon. At longer holds the ledger records peak-vs-close of +30.9% / +6.3% at 12 weeks and +48.8% / +11.3% at 6 months. [LEDGER: project_exit_rules]
Confidence, high on the existence and rough size of the gap; it is enormous, it is measured on a very large sample, and it agrees with independent runs at other horizons.
Counterargument. The strongest case against acting on this: the peak is only knowable afterwards. A 7.54pp median gap is not 7.54pp of available profit; it is the distance to a point you can never identify in advance. Any strategy that assumes you can sell at the peak is backtesting a time machine. Add survivorship bias on top and the honest reading is: the gap is real, its capturable fraction is unknown, and every mechanical attempt so far has captured a negative amount.