We Priced Our Own Entry Zone. 86% of It Has Negative Expectancy.

4 Aug 2026 · RESEARCH NOTE

Every entry plan this site publishes contains four numbers: a zone to buy in, a stop, and two targets. They are computed the same way for all 190+ assets, and they are all anchored to the same two quantities — the 20-day moving average and the 20-day ATR.

We ran the arithmetic on our own plan. At the midpoint of the zone we publish, its historical expectancy is negative 0.19R per trade. This note shows the working, because the number is ours and you should be able to check it.

The plan, stated precisely

From api/calc.js, for every asset scoring 79 or above:

  • Entry zone: MA20 − 0.5 ATR to MA20 + 0.3 ATR
  • Entry midpoint: MA20
  • Stop: MA20 − 2.0 ATR
  • Target 1: +1.5R, Target 2: +2.5R, where 1R is the distance from the midpoint to the stop

Because the targets are defined as multiples of the stop distance, R is the natural unit. One R is 2.0 ATR. Target 1 sits 3.0 ATR above the midpoint; target 2 sits 5.0 ATR above it.

What actually happened, 1,991 times

We walked 23 large, liquid names through five years of daily bars, sampling every sixth trading day, and took every setup that scored 79 or better — 1,991 of them. Then we asked which level the price path touched first, using intraday highs and lows rather than closes, the way a real stop or limit order would fill.

At the live 2.0 ATR stop:

  • Stop first: 67.1%
  • Target 1 first: 32.0%
  • Of those that reached target 1, 77.6% went on to target 2
  • Expired unresolved after 60 trading days: 0.9%

Average time to a stop was 6.0 trading days. To target 1, 13.5. Losses arrive more than twice as fast as gains, which is the ordinary texture of a mean-reversion book and not in itself a problem.

The arithmetic nobody publishes

The usual defence of a 32% hit rate is that winners are larger than losers. It is a real argument. It is also checkable, and here it fails.

Enter at the midpoint and you risk 1R to make 1.5R. Expectancy per trade is therefore:

(0.320 × 1.5) − (0.671 × 1.0) = 0.480 − 0.671 = −0.191R

Hold for target 2 instead of closing at target 1 and it improves, but not enough: −0.100R.

Turn it around and the requirement is stark. To break even at 1.5R with a 1R stop you need to be right 40.0% of the time. The system delivers 32.0%. That is not a small gap, and no amount of position sizing repairs it — sizing scales expectancy, it does not change its sign.

Where the edge actually lives

Here is the part that matters, and it is a point about arithmetic rather than about markets.

The stop and both targets are fixed prices, set at signal time from MA20 and ATR. Whether the subsequent price path touches one of them is a property of that path. It does not depend on what you paid. So the hit rates above — 67.1% and 32.0% — hold for any fill inside the zone, and expectancy can be re-priced at any entry using the same numbers.

Do that across the published zone:

  • Fill at the top (MA20 + 0.3 ATR): risk 2.3 ATR, expectancy −0.295R
  • Fill at the midpoint (MA20): risk 2.0 ATR, expectancy −0.191R
  • Fill at the bottom (MA20 − 0.5 ATR): risk 1.5 ATR, expectancy +0.076R

The sign flips at MA20 − 0.385 ATR.

Our published zone is 0.8 ATR wide. The profitable part of it is 0.115 ATR wide. Fourteen percent of our own entry zone carries positive expectancy. The other 86% does not.

The signal is not the edge. The fill is the edge. Two traders taking the identical signal on the identical day, one filling at the top of the zone and one at the bottom, are not running the same strategy — they are running strategies whose expectancies differ by 0.37R per trade and sit on opposite sides of zero.

Why we are not simply widening the stop

The obvious response is to move the stop out. We swept it, over the same 1,991 setups:

  • 1.0 ATR: stop first 74.7%, target 1 first 25.3%
  • 1.5 ATR: 71.9% / 27.8%
  • 2.0 ATR: 67.1% / 32.0% — live
  • 2.5 ATR: 62.7% / 34.5%
  • 3.0 ATR: 54.8% / 38.3%

Read casually, wider is better: the hit rate climbs 13 points. But the targets are R-multiples, so widening the stop pushes them further away too, and the comparison is only meaningful in R. At 3.0 ATR the midpoint expectancy is +0.026R — positive, and small enough to be indistinguishable from noise at this sample size. It also takes 12.6 days to reach a stop instead of 6, roughly doubling how long capital sits in a losing position, and the target-1-to-target-2 conversion decays from 77.6% to 62.3%.

A number that thin is not a finding. Publishing it as one would be the same error as publishing 32% and calling it an edge.

What we changed

Updated 8 August 2026. The section below originally said we were changing nothing yet. We have now changed the entry zone, and the original text is replaced rather than quietly amended.

Nothing about the score. Layer one measures what it claims to measure, and the exit study never tested the score — it tested the trade plan bolted onto it.

The entry zone has moved. It now runs from MA20 − 1.0 ATR to MA20 − 0.4 ATR, with a midpoint at MA20 − 0.7 ATR. Every point inside it sits at or below the breakeven fill, so the published band no longer contains a region with negative expectancy. At the new midpoint the expectancy is +0.240R and the reward-to-risk is 2.85:1, against −0.191R and 1.50:1 before.

Three things about how that was chosen, because the reasoning matters more than the numbers:

  • The stop and both targets did not move. They keep their MA20-anchored levels, so the 67.1% / 32.0% hit rates still describe them exactly. Re-deriving the targets from the new midpoint would have shifted the levels and silently invalidated the very backtest being used to justify the change.
  • The zone is not at the expectancy maximum. Measured against 586 qualifying asset-days in our archive, expected value per signal peaks near MA20 − 1.25 ATR. We did not go there. That peak comes from a 30-day fill sample and a backtest that never modelled conditional entry; moving to it would be fitting noise. This is a sign correction, not an optimisation.
  • Fill rate is the cost, and it is real. The old midpoint filled on the same day 85.0% of the time; the new one fills 57.2% of the time. You will take meaningfully fewer trades. That is the trade being made, and it should be visible rather than buried.

Separately, a plan whose levels have been overtaken by price now refuses to render as a plan. If price has traded through the stop, the card says so instead of displaying an entry zone sitting above the market. The Trade Desk on 7 August is the worked example: score 89, published zone $17.64–18.58, stop $15.88, market $13.80 — no fill available anywhere, and until this change the card showed it as actionable.

The open question

Taking only fills that reach MA20 − 0.7 ATR selects a subset of signals that dipped deeper than average. The hit rates used above were measured for entries at MA20, and touching a fixed price level does not depend on what you paid — but conditioning on a deeper dip may still change the forward distribution, and our own crowding backtest is a reason to think deeper dips can behave worse rather than better.

That is not tested. A conditional-entry backtest is the follow-up, and until it exists the figures here describe expectancy under an assumption we have stated rather than proved.

The caveats, in full

  • 23 names, not 190. The sweep uses large, liquid US names. Small caps, bonds and crypto are not represented and may behave differently.
  • Same-day ties resolve as losses. When one daily bar touches both stop and target, we count the stop. At the live setting that needs a bar spanning 5.0 ATR, which is rare — but it biases the stop rate upward, and the bias is largest at the narrow stops.
  • Expired trades score as zero. At the live stop that is 0.9% of setups, immaterial; at 3.0 ATR it is 6.9%, which is one more reason not to lean on that row.
  • One fill, assumed. The backtest fills every setup at the midpoint. That assumption is precisely what this note is about.

The point

We publish a 67% stop-out rate on the front page. That number has been on the site for weeks, presented as evidence of honesty, and it was — but it was incomplete, because we never published the expectancy that follows from it. A loss rate is not a result. Expectancy is the result, and ours is negative at the entry price we ourselves highlight.

The correct conclusion is not that mean reversion does not work. It is narrower and more useful: at these targets and this stop, the plan only pays if you are disciplined about where you fill — which is why the zone has been moved rather than annotated. Every number above is reproducible from data/exit-backtest.json and api/calc.js in the public repository. If the arithmetic is wrong, it is checkable, and we would like to know.

Scores in this piece come from the ZonedInOut model. How the score works →

Frequently asked

What is expectancy in trading, and how do you calculate it?
Expectancy is the average profit or loss per trade, expressed in units of risk (R), where 1R is the distance from your entry to your stop. The formula is (win rate x average win in R) minus (loss rate x average loss in R). For example, a system that hits a 1.5R target 32% of the time and a 1R stop 67.1% of the time has an expectancy of (0.32 x 1.5) - (0.671 x 1.0) = -0.191R, meaning it loses an average of 0.19 units of risk per trade. Expectancy, not win rate, determines whether a strategy makes money. Position sizing scales expectancy but cannot change its sign.
What win rate do you need to be profitable at a 1.5:1 reward-to-risk ratio?
You need to win 40% of the time to break even at a 1.5:1 reward-to-risk ratio, before costs. This comes from solving p x 1.5 = (1 - p) x 1, which gives p = 0.4. Any win rate above 40% is profitable and anything below loses money. This is why a headline win rate is meaningless without the reward-to-risk ratio attached: 32% is a losing system at 1.5:1 but a winning one at 2.5:1, where breakeven falls to 28.6%.
Does the price you pay change the probability of hitting your stop or target?
No. If your stop and targets are fixed price levels set at signal time, the probability that the market touches them is a property of the subsequent price path and is independent of your fill. What your entry price does change is your risk-to-reward ratio, and therefore your expectancy. This is why the same signal can be profitable for one trader and unprofitable for another: identical hit rates, different R multiples. It also means backtested hit rates can be legitimately re-priced at any entry within a zone without re-running the simulation.
Is a wider stop loss better?
Not necessarily, and the comparison is often made incorrectly. A wider stop does reduce how often you are stopped out — across 1,991 ZonedInOut setups, widening from 1.0 ATR to 3.0 ATR cut the stop-first rate from 74.7% to 54.8%. But if targets are defined as multiples of the stop distance, widening the stop also moves the targets further away, so each win is worth the same in R while each loss costs the same in R. The gain is real but much smaller than the raw hit rate suggests, and a wider stop keeps capital tied up in losing positions roughly twice as long.
What does ZonedInOut publish as its entry zone, stop and targets?
For any asset scoring 79 or above, ZonedInOut publishes an entry zone running from the 20-day moving average minus 0.5 ATR to the 20-day moving average plus 0.3 ATR, a stop at the 20-day moving average minus 2.0 ATR, a first target at 1.5R and a second at 2.5R, where 1R is the distance from the zone midpoint to the stop. Across 1,991 historical setups the stop was hit first 67.1% of the time and the first target 32.0% of the time, and of the setups that reached the first target, 77.6% went on to reach the second.
Why does ZonedInOut publish backtest results that make its own system look bad?
Because a track record that only contains wins is not a track record. The exit study on this site shows a 67.1% stop-first rate and a negative expectancy of -0.19R at the published entry midpoint, and the crowding study found that the model previously mislabelled falling knives as buying opportunities. Both are published with the underlying data files in a public repository so the arithmetic can be checked independently. A system whose failures are documented can be corrected; one whose failures are hidden cannot.
Where should an entry zone sit relative to the moving average?
It should sit wherever expectancy is non-negative, which is not automatically at the moving average itself. On ZonedInOut the stop is placed 2.0 ATR below the 20-day average and the first target 3.0 ATR above it. Because those are fixed prices, the historical hit rates attached to them (67.1% stop-first, 32.0% target-first) hold for any fill, so expectancy can be computed at each price in the zone. Doing that put breakeven at 0.385 ATR below the 20-day average, meaning a zone centred on the average itself was mostly loss-making. The zone now runs from 1.0 to 0.4 ATR below the average. The cost of moving it down is fill rate: the old midpoint filled on the same day 85% of the time, the new one 57%.

Get the Sunday Brief

One email a week, before the open: what the board did, which calls worked, which didn't, and what changed. Free. No spam, unsubscribe anytime.