Daily Broken-Wing Butterflies on SPY and XSP: Six Rules, 678 Days and a Test Against Luck

A broken-wing butterfly (BWB) is a butterfly with one wing wider than the other, so it can open for a credit. The version tested here has three options on each side of the price. On the put side: sell one put near the price, buy one put a little lower (the body wing, about half the expected move away) and buy one more put lower still (the broken wing, twice as far from the short). The call side mirrors it above the price. Unlike the textbook butterfly (buy 1, sell 2, buy 1), every option bought here sits further out than the one sold. On a one-day SPY butterfly the wings came out about 3 and 6 points from the short.

How a side pays: if the price stays on the right side of the short, the credit is kept. Past the short the loss grows until the body wing, then stays flat at the body width minus the credit — the most the side can lose. Past the broken wing the loss shrinks again and can turn into a profit, because the side then holds one more bought option than sold. A side’s breakeven is its short strike minus (puts) or plus (calls) the credit it took in.

Every rule here opens both sides at once — a put BWB and a call BWB, the shorts at the price — so it earns when the price stays near where it opened and the options lose their time value. Short-dated means days, not weeks: most rules open a butterfly that expires the next trading day (1-DTE), which becomes a 0-DTE position (zero days to expiration) on its last day; one rule opens and closes the same day’s expiration.

Rolling means closing a side that is in trouble and opening it again in a later expiration — trading a loss now for more time. Mid price is halfway between the bid and the ask; every trade here is filled at the mid, which is better than most real fills.

Two wing sizes were tested. “Market wings” are sized from the market’s own implied volatility (read from the at-the-money straddle). “Wide wings” use a fixed 30% volatility instead, which made them about 2.5 times wider: on XSP about 7 and 14 points from the short, where the market’s own volatility, near 11%, gave 2–3.

Each rule opens one put-and-call butterfly a day on XSP, the mini S&P 500 index option (replayed on SPY option prices), at 10:30 am New York time unless noted, the shorts at the price unless noted. A side opens only for a credit of $0.05 or more at mid, and on its expiration day whatever is still open is closed at 3:42 pm unless noted. “Every day” is the rule with no filter; those results come from an earlier run on 674 of the days, and “six runs” means two wing sizes × a decision every 5, 10 or 15 minutes.

A rule counts as a find only when its hindsight result is above the luck range — the best the same search found once the day-by-day results were slid along the calendar, so the readings could no longer know anything. Only T4 with market wings after a weak close cleared it.

Per 1 contract over 678 days, after $0.56 per contract in fees, at mid prices; every T4 row opens at 11:30 am. The honest pick is the best rule on the first half of the days, run on the second half: with 3 conditions it lost in 8 of 12 series and made at most +$614 in the rest. The model’s two winners were +$2,151 (T1, wide wings, inside the luck range) and +$2,316 (T2, wide wings).

Starting point: T4 opened at 11:30 am with market wings on the weak-close days (the two-reading version): +$21,233 per contract. We replayed 1,152 variants of the wing size, take-profit and roll rules on the same days. Today’s rules were already the best — and almost any change cut the result in half or worse.

A few butterflies carry it. Never rolled: 95 butterflies, +$15.9k. Rolled once: −$0.1k. Twice: −$3.1k. Three or more times: 31 butterflies, +$8.5k. The best 20 butterflies made the whole profit. With a $2 per contract spread the wide-wing version falls to −$1.7k and the best every-day variant to −$16.5k; only 1 of the top 20 every-day variants picked on the first half made money later.

Two kinds of slices: one side of the butterfly instead of both, and the market’s mood or the day of the week. The weak-close rows are T4 at 11:30 am with market wings. The mood and weekday rows are T4 opened every day at 11:30 am with market wings — no weak-close filter — with the Fear & Greed index as it stood before the session.

Every day, the call side lost less than the put side in 10 of 12 theory-and-wing series (one put-side exception: T5 with wide wings, +$2,403, both halves up). Across all 432 replays split by mood, every rule with market wings lost on greed days at all 12 entry times (T4: −$19.8k on average there, against +$5.3k in fear). Only T1 with wide wings made money in greed at every time (+$4.3k on average), but at 10:30 shifted mood labels beat it 221 times in 638, and at about 23 contracts a day a $2 spread turned every time negative (−$4.3k to −$11.0k). On greed days SPY simply moved less: a daily swing of 0.63% against 1.61% in extreme fear (2022–26). Fear & Greed and the weekday made no top-10 list in the full search.

Search enough conditions on 678 days and some will look great by chance. These checks separate a pattern from a lucky pick.

Each side ties up its body-wing width × $100 — the most it can lose before its credit. T4 keeps rolling sides out while a new butterfly opens every day, so 7 to 11 sides can be open at once. Per 1 contract over 678 days:

By this arithmetic $10,000 covers 1 contract of the weak-close version. Its three-reading form (T7) had a worst week of −$1,477 and 12 losing weeks of the 54 it traded.

• !Found by looking: the weak-close pattern first appeared in earlier searches of the same 678 days, and every “best” rule here was picked in hindsight. Its call side held up on days it was not picked on (+$5,123 on the second half), but the weak-close idea itself came from looking at all the days.
• !A narrow spike: moving the entry an hour (10:30 or 12:30) cut the result to about half (+$11.5k, +$11.0k); rolling from 2:00 pm left +$8.8k, capping rolls at 2 left +$6.8k, and wings at a full expected move or wider lost in all 576 variants.
• !A few trades carry it: the best 20 butterflies made the whole profit, and butterflies rolled once or twice lost money.
• !Mid prices are optimistic: measured XSP half-spreads on these legs were about $1.5–$2.7 per contract. A $2 per contract spread cut the weak-close rule from +$21,233 to +$12.8k and left the best every-day T4 variant at −$16.5k.
• !Uneven years: 2025 made +$2.0k against +$9.5k in 2024 and +$9.7k in 2026 (to October 9); at the other entry times tried (10:30 to 12:30), 2025 was slightly negative.
• !Rolling keeps losers open: T4 rolls keeping its strikes and pays whatever debit the market asks. Every day with wide wings its deepest drawdown was −$17,062 per contract.
• !Small test choices swing results: T1’s after-fee result went from +$2,033 to −$117 just by deciding every 10 minutes instead of every 5.
• !A stand-in: XSP was replayed on SPY option prices, because XSP’s own minute prices are too sparse. On the days compared they agreed within $72 per rule.

• Prices: real SPY option prices on 678 trading days, January 29, 2024 → October 9, 2026 — 5-minute bars through the day and 1-minute bars from 3:35 pm, strikes within 4.5% of the open, the day’s expiration and the next three — as a stand-in for XSP (on the days compared, within $72 per rule of real XSP quotes).
• Fills: every leg at mid (the options’ traded prices), 1 contract per leg, a decision every 5 minutes, $0.56 per contract in fees; a $2 per contract spread as a stress test.
• Wings: “market” = body wing half the expected move from the at-the-money straddle’s implied volatility, broken wing twice as far; “wide” = the same with a fixed 30% volatility.
• Conditions: 94 chart indicators on SPY 1-, 5-, 15-minute and daily bars, each reading taken at the last finished bar before entry and ranked against its own past (the past 252 sessions for daily bars, the same time of day over 120 sessions for minute bars), plus the weekday and the Fear & Greed index as known before the session. Lines that change when later bars arrive (Ichimoku’s lagging span, ZigZag) were dropped.
• Luck benchmark: the day-by-day results slid along the calendar so readings and trades no longer line up; the best the same search finds there is what luck produces.
• Honest tests: rules picked on the first half of the days and judged on the second; a gradient-descent model fitted on the first half (its penalty chosen on the last quarter of that half).