Exit Rules Study: What Should Close a Breakout Trade, and How Big Should the Book Be?

Twenty-six sell rules over every S, A+ and A graded breakout in NASDAQ and NYSE history, replayed through a slot-limited book, then sized, then filtered by market regime and relative strength. The exit matters less than the order you take signals in.

How the rules compare

Two views of the same eight rules across all 97,789 breakouts. Profit factor is gross wins over gross losses. Return per month is mean return divided by average holding period, the number that matters when a five-slot book has to be refilled.

The 20% trail earns the most per trade but holds 95 bars on average, and two-thirds of its trades still end at the hard stop, since the trail only rises above entry after a +25% move. On S-grade names it is worth that wait.

Findings

• Close below the 50-day average is the best single fast sell rule. Same mean as a fixed 42-bar hold (1.10% vs 1.04%) with the hard-stop rate cut from 45% to 25%: it exits losers at −2 to −3% before they reach −7%. On S-grade breakouts: 1.77% mean, profit factor 1.91, 8% stops.
• Selling into +20% does not help. Half off at +20% then MA50: 0.99% mean. All out at +20%: 0.88%. Plain MA50: 1.10%. The rest of the return lives in the tail the partial removes.
• Trailing stops make the most per trade and tie up the slot. A 20% trail: 3.73% mean, profit factor 1.80, 95 bars. Per month it is 0.83% against 0.68% for MA50. Adding a 15% trail to MA50 is worse than MA50 alone (0.89%).
• The 20-day average is too tight. Highest turnover, lowest profit factor (1.16). It sells ordinary pullbacks.
• O'Neil's composite adds nothing over its MA50 component. 0.99% vs 1.10%. Only 12% of trades reach +20% inside 15 bars, and when they do the MA50 exit was already doing the work.
• Grade dominates the exit. S beats A on every rule by two to four times in profit-factor terms. The exit rule moves results by tenths of a percent; the grade moves them by whole percents. See how bases are graded.
• 2010 to 2026 is weaker across the board, with mean returns roughly half of pre-2010, but the ranking of rules is unchanged.

All rules, all breakouts

Win = share of trades closing above entry. PF = gross wins ÷ gross losses. Bars = mean holding period in trading days. %/mo = mean return ÷ bars × 21. Ends at stop = share of trades ending at the 7% hard stop. A −7.00% median means more than half of trades ended there.

S grade only: 4,072 breakouts

Peter Brandt's 3-day trailing stop

Stop under the lowest low of the prior three sessions, ratcheted daily and never lowered. Brandt uses it to protect parabolic runs, so it is measured armed from entry, armed after +10% and +20%, close-based, and layered on top of the 50-day rule.

• From entry it is harmful. A fresh breakout's 3-day low is the base it just left. Ordinary post-breakout pullbacks take it out within a week.
• Armed after +20% it is a middle path. Profit factor 1.50, between MA50 and the 20% trail, with shorter holds than the trail. On S grade it earns 5.6% mean at PF 2.48, and armed after +10% on S it has the best median of any rule, +6.9%.
• Stacked on MA50 it hurts. 0.83% against 1.10%: it clips winners the MA rule would have carried.

Portfolio simulation: which rule grows the book

Per-trade averages ignore the slot constraint. This replays the signal stream day by day with a five-slot book and 1% of equity at risk per trade, which with a 7% stop is a 14.3% position, capped at 20%, no leverage. Free slots are filled from that day's breakouts, best grade first. Every position is marked to market daily from its own price path.

Slots are full almost every day, so which of a day's signals get taken is a tie-break, and a single ordering is noisy. These figures average twelve random within-day orderings and show the spread.

• The 20% trail is the best rule for the book, not just the best per trade. Highest growth, lowest drawdown, best Sharpe, a third of the trades. It holds in 2010 onward and with grade priority switched off.
• MA50's advantage was slot turnover, and turnover buys nothing here. The refilled slot gets an average signal, and average signals barely beat zero after the first month. Fixed 21 and MA50 land in the same place.
• Brandt after +20% is second with grade priority on, but drops to 5.1% with it off. It leans on catching S-grade runners.
• The +20% partial is last or near last in every configuration.
• The composite tracks MA50 because S is 4% of the stream. Trailing every grade is the setting the evidence supports.
• Read the CAGRs as relative, not forecasts. The universe is today's listings, so survivorship flatters it. Fills are frictionless. The raw signal stream is broader than what the scanner emails.

Sizing: how much the book can make

The same simulation swept slots, risk per trade, position cap and trail width, twelve orderings each, unlevered. A position is the smaller of risk ÷ 7%, the cap, and the cash left.

• The book caps out near 9 to 9.5% a year once it is fully invested. Five slots at 1.5% risk already demand 107% of equity. Raising risk beyond that changes nothing but the drawdown. Without margin, 2% risk on five slots gives 9.3%, not the 14% a doubling would suggest.
• Diversification is the free lever. Ten slots at 1% risk earns the same 9.0% as five slots at 2%, with the best Sharpe in the grid and the best worst-case ordering, at half the concentration.
• Drawdown is set by exposure, not by risk per trade. Every fully invested row sits at 37 to 39%. The only way under that is to hold cash: five slots at 1% is 71% invested and 31% drawdown, three slots is 20%.
• Trail width: 15% and 20% are close over the full history, 20% is best from 2010 onward, 25% and 30% give up Sharpe for nothing. 20% stays.

Selection and regime

Two mechanical pieces of what a discretionary trader does, added to the fully invested profile and run over twelve orderings. Regime: the S&P 500 above its 200-day average, either blocking entries only or also flattening the book below it. Relative strength: each trade's six-month return ranked against the whole universe at the start of its month, the scanner's RS rating replayed, used as a floor or as the order in which free slots are filled.

• Ranking by strength is the biggest lever in the whole study. The same signals taken strongest-first lift the book from 9% to 14% a year with the same drawdown and a tight spread. A floor alone does little. The preference is what matters.
• The regime filter earns its keep only if it also exits. Blocking entries below the 200-day average loses the early re-entries. Flattening below it and re-entering above halves the worst drawdown at a higher return.
• Together: about 14% a year, 26% worst drawdown, Sharpe 0.92. The index's return with half its worst drawdown, from a rule set. From 2010 onward the same filters give 12.7% with a 29% drawdown.
• On a trading championship's own metric, single ordering with the best filters: best calendar year +67%, best rolling twelve months +99%, worst year −15%. The MA50 variant under the same filters posts a +142% year and a −30% year. That is the shape of a contest entry, from the same book.
• Caveats sharpen here. The strength rank leans on today's listings, turnover roughly doubles so frictionless fills flatter it more, and regime whipsaws are counted but their slippage is not.

The time backstop and the market-health score

Two follow-ups. A 90-day backstop looked like a safety net, but with the trail rule it is the exit: the trail's average hold is 95 bars and a fifth of its trades were still open at the study's 250-bar limit.

The dashboard's Market Health score, trend plus distribution days plus breadth, was rebuilt back to 1971 and tested as an entry floor and an exit trigger on the invested profile with the strength filters. Risk-on is 70 and above, caution 45 to 69, risk-off below 45.

• Caution is not a sell. A 56 out of 100 tape is where much of the return is made. Requiring risk-on cuts the book to 9%, requiring 55 cuts it to 11%. Breakouts that work often start while the gauge is still rebuilding.
• Risk-off is the line. No entries below 45 and flattening below it is the best full-history result in the study, but every health-based rule degrades sharply after 2010, while the S&P 200-day rule holds. Distribution days and breadth said more in the 1990s tape than they do in the ETF era.
• So the switch worth building on is the 200-day rule, with the health score's risk-off line as an optional extra brake.

Circuit breakers: stopping when it isn't working

Four families of equity-based breakers on the best setup, twelve orderings. Halted is the share of days the breaker kept the book switched off.

• Equity-curve breakers destroy this strategy. Its profit arrives in bursts after stretches of small stop-outs. A drawdown halt or an equity-average filter switches the book off in exactly those stretches and misses the burst that pays for them. The 15% halt keeps the book idle 77% of the time and earns 4.4% a year.
• Loss-streak pauses cost a third of the return and remove no bad years. Streaks of stops are the normal texture of a 35% win-rate rule, not a sign something broke.
• The only breaker that does not hurt is sizing down with drawdown. Scaling risk to zero at 40% drawdown gives up one point of return for three points of drawdown. It never halts, so it never misses the recovery. A lever, not a default.
• The deep losing years are already handled by the market switch. With it, the worst year is −11.6% and one year in 36 lands below −10%. Without it, −25% and nearly four. Breakers cannot replace the switch, because they react to the book's own losses, which lag the market by weeks. The 200-day rule reacts to the market itself.

Rotation: leave when momentum fades, take the next breakout

The discretionary version of this style does not wait through sideways tape. Tested three ways on the best setup, now with a transaction-cost model of 10 basis points a side, roughly a market order in a liquid name: stall exits after N days without a new closing high, clock rotation on fixed holds, and swapping a stalled laggard only when a stronger signal is waiting.

• Rotation raises raw return. With strength-ranked refills, recycling the slot every one to two months beats holding through the trail. The next strong name is worth more than a stalled one.
• The stall signal carries no information. No new high in 15 to 20 days performs like a plain 15 to 20 day clock. The gain comes from redeploying into the current strongest breakout, not from detecting the fade. Swapping a laggard only when a stronger signal appears adds 0.2 points.
• Costs and drawdown eat most of it. Turnover is two to four times the trail's. At 10 basis points a side the 21-day clock loses 3 points and the 10-day clock collapses. The fast rules' worst drawdowns run 45 to 68% against the trail's 27%.
• After 2010, at realistic cost, the trail is the best risk-adjusted rule. Only the 42-day clock beats it on return, and it pays with a 55% drawdown.

Is a base under 10% deep the real top tier?

Prompted by a stock that graded A off a 15.5%-deep, 21-bar base. Blue-sky bases above the 200-day average, at least 25 bars long, bucketed by depth:

Where the numbers come from

This is a full-history study, not the live signal table. DataQuant's graded alerts only began carrying a grade in September 2026, which is far too few trades to rank exit rules. Instead the production base detector was replayed over every symbol's complete price history, and each close that resolved a blue-sky pivot above the 200-day average with at least 100k average volume became one trade, graded by the production cut. Live signals will be measured against these numbers as they accumulate, and reported the way the honest backtest reports: wins and losses both.

Entry is the breakout day's close. The 7% stop fills on the low, at the open if price gaps through. Close-based signals fill at the next open. Open-ended rules are capped at 250 bars. No slippage or commissions except where a cost model is stated. Blue-sky is a pivot within 2% of the 52-week high. Yahoo daily bars, split-adjusted; warrants, rights and units excluded. The universe is today's listings, so survivorship flatters every absolute return; the rankings between rules are the finding.