SMB Capital  •  Trading Research  •  Executive Summary

Pullback & Reclaim — Findings

Prepared for management  •  18 August 2026  •  Two-minute read  •  plain-English intro included

New To This? Read This First

What the trade is. A stock has been rising for a while. It pauses and drifts back down to one of the average-price lines traders watch. It goes quiet there for a few days. Then it closes back above the fastest of those lines — and that is the signal to buy. We expect it to travel back up to the high it made before the pause.

How we checked it. We wrote the rules down precisely and had a computer apply them to 150 large US stocks over four years — about 4,100 trades. Crucially, we chose the rules using only the first two years, then scored them on the last two years the rules had never seen. A method that only works on the data used to invent it fails that test immediately. This one did not.

What we found. The trade itself works. But the money is made or lost after you are already in, and the exit rules the desk had written down were giving back about half the available profit. Different exit rules — same stocks bought, same money risked — roughly double the result. Nothing about the buying changes.

What the numbers mean. Results are quoted in “R”, which simply means “multiples of what you risked”. Risk any amount and make that same amount back and that is +1R. Our result is +0.39R — so the average trade returns about 39 cents for every dollar put at risk, after commission and spread. Real, but not enormous.

The catch worth knowing. More than half of these trades lose money. The profit comes from a minority that run a long way. That has a practical consequence: traders cannot pick and choose, because skipping the one trade that becomes the big winner removes the month's profit. A run of six or eight losses in a row is normal here, not a warning sign.

The one-sentence version The trade is real and worth doing, our written exit rules were leaving roughly half the money on the table, and the evidence is strong enough to change those rules but not strong enough to put more capital behind the strategy.

At A Glance

Six things to know. Click any line for the detail.
+0.39R The edge is real, and modestAverage net profit per trade, out of sample, after costs. That is 39 cents back for every dollar risked.The numbers → 44% Costs consume nearly half the gross result+0.69R gross becomes +0.39R net. The strategy stops working entirely if round-trip costs reach 0.34%.Gross vs net → +80% Fixing the written exit rules is the prizeNo new capital, technology or risk — only a change to how existing positions are managed.The numbers → >50% Most individual trades loseProfit comes from a minority of large winners. Every qualifying signal must be taken or the edge is not captured.How much to trust it → 9 : 2 The assumptions lean optimisticNine flatter the result against two that understate it. No stop slippage and no bear market are the two that matter most.Assumptions → 3 Three things needed to finishFour desk decisions, the scanner's failed setups, and three remaining tests.What we need →

Bottom Line

A trading pattern the desk already runs was formalised into written rules and tested across 4,528 trades over four years. The pattern shows a real, statistically durable edge — and the desk is currently capturing roughly half of it, because the exit rules were never written down.

Fixing the exit rules is a documentation and discipline change, not a new strategy. It requires no new technology, no new capital, and no new risk.

The Numbers

Per trade, netPer trade, netWhat it isThe average profit per trade after subtracting commission and spread, expressed in R.What it tells youWhether the strategy makes money once real-world trading costs are paid. Gross figures flatter every strategy; this is the one that counts.How it is usedThe single headline number. Multiply by expected trade count to project a period.Our number+0.39R. The average trade returns 39% of what it risked. Costs consume 44% of the gross result, which is normal for a strategy with three exit points.Good — solid, not spectacular
+0.39R
out of sample, after costs
95% range, net95% rangeWhat it isThe band the true long-run average is very likely to fall inside — not the range of individual trades, which is far wider.What it tells youHow much is still unknown. A narrow band means the size of the edge is well pinned down; a wide one means it could be much better or much worse than the headline.How it is usedPlan around the bottom of the range, never the middle. If the bottom is negative, the strategy could in principle lose.Our number−0.10 to +0.79 after costs. Before costs the range is entirely positive (+0.21 to +1.10). After costs the pessimistic end dips just below break-even.Neutral — positive centre, negative floor
−0.10 … +0.79
pessimistic end touches zero
Uplift availableUplift availableWhat it isThe improvement from changing the exit rules only — same trades entered, same money risked, different management after entry.What it tells youThe size of the prize for a change that costs nothing to implement.How it is usedIt is the business case. No new capital, technology or risk is required to capture it.Our number+80%. The tested exit rules return +0.39R against +0.216R for the rules currently written in the desk checklist.Good — free improvement
+80%
from the exit rule alone
SampleSample sizeWhat it isHow many historical trades the conclusions rest on, and over what period.What it tells youWhether the result could be a fluke. Small samples prove nothing; large ones still mislead if the trades are not independent of each other.How it is usedJudged alongside the number of independent time periods, which is the real constraint here.Our number4,142 trades, 150 stocks, four years — but only 20 independent months in the out-of-sample window. The month count is the honest sample size.Neutral — many trades, few months
4,142
tradeable trades · 150 stocks · 4 years
Exit methodNet per tradePer trade, netWhat it isThe average profit per trade after subtracting commission and spread, expressed in R.What it tells youWhether the strategy makes money once real-world trading costs are paid. Gross figures flatter every strategy; this is the one that counts.How it is usedThe single headline number. Multiply by expected trade count to project a period.Our number+0.39R. The average trade returns 39% of what it risked. Costs consume 44% of the gross result, which is normal for a strategy with three exit points.Good — solid, not spectacularStatus
What the desk did on CRWD (no written rule) +2.00R on a 10.63R moveThe CRWD tradeWhat it isA live trade the desk took on 31 July 2026, used here as the worked example.What it tells youIt shows the cost of the missing rule concretely rather than in the abstract.How it is usedThe trade is documented in full in the linked closeout review.Our numberThe move was worth 10.63R; the desk banked 2.00R. The difference was a hold rule that had never been written down.The problem, illustratedAd hoc
Current written ladder+0.216RIn the checklist today
Best tested ladder+0.389R Recommended

RR (risk unit)What it isThe amount of money put at risk on a single trade — the distance between the entry price and the price at which the trade is abandoned.What it tells youIt converts every trade to a common scale. A $200 stock and a $20 stock both risk 1R, so results can be averaged without one stock dominating.How it is usedAll results on this page are quoted in R. Multiply by whatever is risked per trade to get money: +0.39R returns 39 cents per dollar risked.Our numberMedian risk per trade in the study was 0.99% of the share price.Unit of measure is the risk unit — one R is the money risked per trade. A result of +0.39R means the average trade returns 39% of what it risked — 39 cents for every dollar. The rules currently written down return about 22 cents per dollar. These figures are net of estimated commission and spread, and exclude setups whose stop is too tight to be tradeable.

Gross Versus Net

Trading costs are not a rounding error here — they consume between 41% and 67% of the gross result depending on the exit rules used. This is the single strongest argument for the recommended ladder, separate from its higher return.

Exit ladderGrossCostNet Cost as % of grossBreak-even cost
Current checklist ladder+0.519−0.304 +0.21658%0.257%
Trail the 30%, 5-DMA runner+0.598−0.304 +0.29451%0.296%
Recommended — 17-DMA runner +0.692−0.304 +0.38944%0.342%
Slowest runner — 50-DMA+0.736−0.304 +0.43241%0.364%
Two tranches only+0.450−0.304 +0.14767%0.222%

Break-even cost is the round-trip cost at which that ladder returns exactly zero. Our assumption is 0.15%; the recommended ladder keeps working up to 0.34%, more than twice our estimate. The current checklist ladder breaks at 0.26%.

If Our Cost Assumption Is Wrong

Net return per trade at different round-trip costs. Read across the row to see how much execution quality matters.

Exit ladder0.05%0.10%0.15%
assumed
0.20%0.30%
Current checklist ladder+0.418+0.317 +0.216+0.115−0.088
Trail the 30%, 5-DMA runner+0.497+0.396 +0.294+0.193−0.009
Recommended — 17-DMA runner +0.591+0.490+0.389 +0.288+0.085
Slowest runner — 50-DMA+0.634+0.533 +0.432+0.331+0.129
Two tranches only+0.349+0.248 +0.147+0.045−0.157
The finding that matters here At a 0.30% round trip — double our assumption, and plausible in less liquid names — the current checklist ladder and the two-tranche version both lose money. Only the ladders that trail the first tranche survive. The recommended change is therefore not just higher-returning, it is the one that stays profitable if execution is worse than assumed.

How Much To Trust It

What is solid

  • Tested on data never used to design the rulesOut of sampleWhat it isTesting the rules on a period deliberately held back and never used to design them.What it tells youWhether the rules found something real or were tuned to fit past noise. This is the standard guard against self-deception in strategy research.How it is usedRules were chosen on Oct 2022 – Dec 2024 and scored on Jan 2025 – Aug 2026. Only the second number is quoted as the result.Our number+0.692 out of sample against +0.830 in the design period. Close together, which is the good outcome — it did not collapse.Good — held up — results held (+0.692 vs +0.830)
  • Survives a statistical test that accounts for trades clusteringClustering testWhat it isA check that corrects for many trades happening at the same time in stocks that move together.What it tells youStandard statistics assume each trade is an independent bet. When 200 positions all open in the same week, they are really one bet, and ordinary confidence measures overstate certainty badly.How it is usedWhole calendar months are resampled 4,000 times so correlated trades stay bundled together.Our numberThe correction widened the uncertainty substantially, and the edge still survived it before costs.Good — survived the correction — before costs
  • All four years positive, between +0.55R and +1.00R — no single year carries it
  • Profitable after realistic commission and spread on the central estimate
  • Stock universe selected in advance and frozen — no hindsight in the sample
  • 73% of months profitableMonths profitableWhat it isThe share of calendar months in which the strategy made money.What it tells youHow often a losing stretch should be expected. It sets the tolerance needed to keep following the rules through a bad run.How it is usedUsed to judge whether drawdowns are normal or a signal that something has broken.Our number73% of 45 months profitable, so roughly one month in four loses money.Good for this style

What is thin

  • No bear-market evidence. The setup stops firing in downtrends, so it sits out rather than losing — but that is untested
  • The test uses a close variant of the desk's entry, not the exact one. Resolving this is the top open item
  • Only 20 independent months in the out-of-sample window
  • Large-cap stocks only; the desk also trades small caps
  • Over half of individual trades lose — profits come from a minority of large winnersMean versus medianWhat it isThe average trade versus the typical trade. When they differ a lot, results depend on a few outsized winners rather than steady gains.What it tells youHow the returns will actually feel, and what discipline is needed to capture them.How it is usedIt dictates position sizing and, critically, that every qualifying signal must be taken.Our numberAverage +0.69R, typical trade +0.34R. Over half of all trades lose. The profit comes from a minority that run a long way.Caution — hard to sit through
  • After costs the pessimistic end of the range is slightly negative (−0.10R). The central estimate is clearly positive, but the floor is not
  • Results exclude 7% of setups whose stop was too tight to tradeUntradeable setups excludedWhat it isHistorical trades whose stop was so close to the entry that they could not have been executed in reality — the bid-offer spread alone would have closed them.What it tells youWhether the backtest is measuring achievable trades or arithmetic artefacts. A stop 0.002% away produces an enormous return as a multiple of risk while being impossible to trade.How it is usedAny setup with a stop tighter than 0.25% of the share price is removed before results are calculated.Our number7% of setups removed. Including them had inflated the headline from +0.39R to +0.82R — the correction noted on this page.Neutral — conservative treatment — a blunt correction that likely understates the true figure somewhat
Correction, 18 Aug An earlier version of this page showed +0.82R per trade. That figure was inflated by a small number of backtest trades whose stops were a fraction of a percent from entry — untradeable in practice, but producing enormous returns when expressed as a multiple of risk. Excluding them halves the result to +0.39R. The edge remains, at roughly half the previously stated size.
The risk management point More than half of these trades lose money. The positive average comes from a minority that run a long way. This requires consistent position sizing and taking every qualifying signal — a run of six or eight consecutive losers is normal and is not evidence the pattern has stopped working. A trader who skips signals to avoid losers will underperform the tested result.

Assumptions Behind These Numbers

Every backtest rests on assumptions. These are ours, with an honest note on whether each one flatters the result optimistic, understates it conservative, or is unlikely to matter neutral.

Costs and execution

Commission + spread 0.15% round trip, charged twice per trade because the ladder exits in three pieces. At the median stop of 0.99% of price that is 0.304R deducted from every trade — the difference between the +0.69R gross and +0.39R net figures. Neutral
Stop slippage None. Stops are assumed to fill at exactly the stop price. In reality stops gap through, especially on the news days that cause them. Optimistic
Entry fills At the closing price of the trigger bar, in full, with no market impact and no partial fills. Optimistic
Overnight financing Not modelled. Positions are held for days to weeks; borrow and margin costs are real but small relative to 0.15%. Slightly optimistic

Data and universe

Stocks 150 US common stocks, ranked by dollar volume on 30 Dec 2022 with a $10 minimum price, then frozen — never re-ranked. No stock was chosen because it later did well. Neutral
Period October 2022 – August 2026. Rules chosen on Oct 2022 – Dec 2024; results quoted from Jan 2025 – Aug 2026 only. Neutral
Bars 30-minute bars, regular hours only (09:30–16:00 ET). Pre-market and after-hours activity is invisible to the test. Neutral
Market regime The window is essentially one large-cap bull market. There is no bear-market segment of any length in the sample. Optimistic
Company size Large caps only. The desk also trades small caps, which gap more and carry wider spreads. Optimistic

How each trade was constructed

Stop placement The lowest low of the trigger bar and the two bars before it. Neutral
Minimum stop Setups with a stop tighter than 0.25% of the share price are discarded entirely — 7% of all setups. They are not executable, but discarding rather than widening the stop throws away trades that could have been taken with more risk. Conservative
Maximum stop Setups risking more than 25% of the share price are discarded. Neutral
Maximum hold 60 trading days, after which the position is marked at the last close rather than sold. A mark is not a fill. Slightly optimistic
Position sizing Every trade is weighted equally, as if the same dollar risk were taken each time. No limit on how many positions run at once. Optimistic

What the setup filter did — and did not — include

Included Price above a rising 50-day average; a close back above the 5-day average after being below it; and a pullback that touched the 10, 20 or 50-day average within the prior eight sessions. Neutral
Not included The desk's actual qualification filters. Bollinger-width percentile, coil bar count, volume pocket and moving-average cluster width are all absent. The tested filter is looser than the one traders use, so it admits setups the desk would reject. Unknown direction
Entry trigger A break of the prior day's high. The desk's CRWD trade was entered on a VWAP pull-in instead — a related but different trigger. Resolving which is canonical is the top open item on this project. Unknown direction

Not modelled at all

Excluded Taxes; trading halts; order rejections; capital or buying-power limits; earnings gaps treated as ordinary bars; and any discretionary decision a trader would make in the moment. Optimistic
Net direction of the assumptions More of these flatter the result than understate it — chiefly no stop slippage, no bear market, large caps only, and unlimited concurrent positions. The one materially conservative assumption is discarding 7% of setups outright rather than trading them with a wider stop. A fair reading is that +0.39R is closer to a ceiling than a floor, and that the honest figure after slippage and a full market cycle would be lower. That is why the recommendation is to adopt the exit rules at reduced size rather than to scale.

What We Need To Finish This

1
Four desk decisions Which stop definition is the standard, which entry is canonical, the per-trade risk figure in dollars, and the policy for holding through earnings. All four are ambiguous in the current documentation and each changes the reported numbers.
2
Scanner history including the failures The trade library contains only setups that worked. No entry filter can be validated without the ones that failed. This is the single largest blocker.
3
Three remaining tests Re-run on the desk's exact entry, add realistic stop slippage, and extend to a small-cap universe. All use tooling that already exists.

Recommendation

Adopt the tested exit rules now and trade the play at reduced size while the remaining tests complete. The exit change is worth roughly +80% per trade against the rules currently written down, and that conclusion is unaffected by the correction above — both figures moved together. The exit change is low-risk — it alters how existing positions are managed, not what is bought or how much is risked — and it is the difference between capturing half the edge and most of it.

Do not increase size or capital allocation until the entry question is resolved and the failed-setup data is in hand. The statistics are strong; the coverage is not yet wide enough to justify scaling.

Glossary — Every Metric Explained

The same text appears as a pop-up wherever a term is used on this page. Hover it, or tap it on a phone.

R (risk unit)What it isThe amount of money put at risk on a single trade — the distance between the entry price and the price at which the trade is abandoned.What it tells youIt converts every trade to a common scale. A $200 stock and a $20 stock both risk 1R, so results can be averaged without one stock dominating.How it is usedAll results on this page are quoted in R. Multiply by whatever is risked per trade to get money: +0.39R returns 39 cents per dollar risked.Our numberMedian risk per trade in the study was 0.99% of the share price.Unit of measure
Per trade, netWhat it isThe average profit per trade after subtracting commission and spread, expressed in R.What it tells youWhether the strategy makes money once real-world trading costs are paid. Gross figures flatter every strategy; this is the one that counts.How it is usedThe single headline number. Multiply by expected trade count to project a period.Our number+0.39R. The average trade returns 39% of what it risked. Costs consume 44% of the gross result, which is normal for a strategy with three exit points.Good — solid, not spectacular
95% rangeWhat it isThe band the true long-run average is very likely to fall inside — not the range of individual trades, which is far wider.What it tells youHow much is still unknown. A narrow band means the size of the edge is well pinned down; a wide one means it could be much better or much worse than the headline.How it is usedPlan around the bottom of the range, never the middle. If the bottom is negative, the strategy could in principle lose.Our number−0.10 to +0.79 after costs. Before costs the range is entirely positive (+0.21 to +1.10). After costs the pessimistic end dips just below break-even.Neutral — positive centre, negative floor
Uplift availableWhat it isThe improvement from changing the exit rules only — same trades entered, same money risked, different management after entry.What it tells youThe size of the prize for a change that costs nothing to implement.How it is usedIt is the business case. No new capital, technology or risk is required to capture it.Our number+80%. The tested exit rules return +0.39R against +0.216R for the rules currently written in the desk checklist.Good — free improvement
Sample sizeWhat it isHow many historical trades the conclusions rest on, and over what period.What it tells youWhether the result could be a fluke. Small samples prove nothing; large ones still mislead if the trades are not independent of each other.How it is usedJudged alongside the number of independent time periods, which is the real constraint here.Our number4,142 trades, 150 stocks, four years — but only 20 independent months in the out-of-sample window. The month count is the honest sample size.Neutral — many trades, few months
Out of sampleWhat it isTesting the rules on a period deliberately held back and never used to design them.What it tells youWhether the rules found something real or were tuned to fit past noise. This is the standard guard against self-deception in strategy research.How it is usedRules were chosen on Oct 2022 – Dec 2024 and scored on Jan 2025 – Aug 2026. Only the second number is quoted as the result.Our number+0.692 out of sample against +0.830 in the design period. Close together, which is the good outcome — it did not collapse.Good — held up
Clustering testWhat it isA check that corrects for many trades happening at the same time in stocks that move together.What it tells youStandard statistics assume each trade is an independent bet. When 200 positions all open in the same week, they are really one bet, and ordinary confidence measures overstate certainty badly.How it is usedWhole calendar months are resampled 4,000 times so correlated trades stay bundled together.Our numberThe correction widened the uncertainty substantially, and the edge still survived it before costs.Good — survived the correction
Mean versus medianWhat it isThe average trade versus the typical trade. When they differ a lot, results depend on a few outsized winners rather than steady gains.What it tells youHow the returns will actually feel, and what discipline is needed to capture them.How it is usedIt dictates position sizing and, critically, that every qualifying signal must be taken.Our numberAverage +0.69R, typical trade +0.34R. Over half of all trades lose. The profit comes from a minority that run a long way.Caution — hard to sit through
Untradeable setups excludedWhat it isHistorical trades whose stop was so close to the entry that they could not have been executed in reality — the bid-offer spread alone would have closed them.What it tells youWhether the backtest is measuring achievable trades or arithmetic artefacts. A stop 0.002% away produces an enormous return as a multiple of risk while being impossible to trade.How it is usedAny setup with a stop tighter than 0.25% of the share price is removed before results are calculated.Our number7% of setups removed. Including them had inflated the headline from +0.39R to +0.82R — the correction noted on this page.Neutral — conservative treatment
The CRWD tradeWhat it isA live trade the desk took on 31 July 2026, used here as the worked example.What it tells youIt shows the cost of the missing rule concretely rather than in the abstract.How it is usedThe trade is documented in full in the linked closeout review.Our numberThe move was worth 10.63R; the desk banked 2.00R. The difference was a hold rule that had never been written down.The problem, illustrated
Months profitableWhat it isThe share of calendar months in which the strategy made money.What it tells youHow often a losing stretch should be expected. It sets the tolerance needed to keep following the rules through a bad run.How it is usedUsed to judge whether drawdowns are normal or a signal that something has broken.Our number73% of 45 months profitable, so roughly one month in four loses money.Good for this style