Is the monthly flip a floor?

The short answer

Not that we could show. On SPY and QQQ, 2018–2020, a bounce from the monthly flip (Flip M on a Prism by GEXfreak chart) won 61.0% of the time; the same trade at a random line, one with no options data behind it, the same distance from price, won 64.1%. Both run high because trades are booked at the line, not where price turned, so only the gap between them matters.

That gap is the lift. For the bounce, the top of its range is about +4.1 points, so a large floor effect looks unlikely.

The other three situations (a turn down from a flip above price, a close through it down or up) read “not detected”, not “no effect”: their ranges could hide a real one.

Two limits matter: the flip was rebuilt after the fact, so this is not a claim about the live level, and trades ran inside one day, so no month-long hold was tested. An earlier morning study found a bigger gap; see Limits.

What to do with this

• Ask of any claim that a level “held” 75% of the time: what did a random line the same distance from price hold on the same days? Here the flip did not beat one.
• Do not count on Flip M to stop price, or read it as a floor because it is the monthly flip. Treat it as a reference line, not a prediction. Among the rebuilt levels the monthly flip is one of the least accurate: within $1 of a true intraday map on 66.9% of marks, against 73.7% for the weekly flip (see Limits). This result is about the rebuilt level.
• Your own log would need about 100 logged events (about 1.8 years) to show even a gap of about 17 points, so for Flip M expect “not detected”. The same sheet tests any other level or claim.
• Only flips 0.10 to 1.50 expected moves from price at 09:35 (a tenth of a typical day’s move up to one and a half times it) were tested; check where today’s Flip M sits first.

How we tested

GEXfreak rebuilt the monthly flip for every SPY and QQQ session from 1 May 2018 to 31 December 2020 and kept the 845 of 1,344 sessions (one symbol on one day) with a flip in reach; 38 had one on each side of price and count on both. Four situations were tested: a bounce from a flip below price (a support fade), a turn down from one above (a resistance fade), and a close through it down or up (a break down, a break up), each a same-day trade with a target and stop 0.35 expected moves away (a bracket). That gave 427 events.

Each flip trade was set against random lines (see Terms). A random fade wins more than half the time because it is booked at the line while price may be up to 0.10 expected moves short of it, which puts the target nearer than the stop. Random lines below price beat those above, but the flip and its random lines share the same side and days, so that gap does not tilt the comparison.

The 90% range says what this data leave room for. The smallest detectable lift is a different, planning size, what a fresh test this big would flag four times in five: 10.6 points for the bounce and 14.7 to 20.4 for the other three, all above the 5 points the study set in advance as the smallest lift it wanted to see. A real floor would show a positive lift on the bounce and a negative one on a break down.

A second test on seven names (AMD, BAC, QQQ, SPY, TSLA, WMT and XOM) scored trades in R (multiples of the stop distance) with the same finding; the resistance fade came closest to clearing zero.

Check it yourself with Prism and a spreadsheet

You need a spreadsheet, the Prism chart at 09:35 ET and a broker chart with one-minute bars. It is rougher than the study: fades only (breaks need a different exit rule), one unweighted random line per row, and your own yardstick. Past flips cannot be read off a chart, so the log starts the day you begin. Every session with a flip in reach is a row scored twice (J and M); score it the same day, while your broker chart has the bars. To practise first, use yesterday’s high or low as the level on the last 20 sessions and ignore the result.

Steps show one recorded row (31 May 2018, SPY), where D is the study’s expected move, 2.0859; your yardstick will differ in size.

The steps

• 1. Fill A to D at 09:35 ET. Recorded row: B 272.0900, C 271.6841 (the rebuilt flip), D 2.0859.
• 2. E =ABS(C2-B2)/D2 and F =IF(C2<B2,"support","resistance"). Keep the row only if E is above 0.10 and at most 1.50; delete it otherwise. Recorded row: E =ABS(271.6841-272.0900)/2.0859 = 0.195, F support.
• 3. G =0.35*D2, H =IF(F2="support",C2+G2,C2-G2) and I =IF(F2="support",C2-G2,C2+G2). Recorded row: G 0.7301, H 272.4142, I 270.9540.
• 4. Score J on the one-minute chart in three parts. Start: the first minute from 09:36 to 15:30 ET in which price comes within 0.10*D2 of the level (support: the low reaches C2+0.10*D2; resistance: the high reaches C2-0.10*D2); leave J blank if none does. Result: from there J is W if price reaches H2 before I2, L if it reaches I2 first, T if neither by 15:55 ET. Ties: a minute that reaches both counts as L, and on the start minute only the stop counts, because one bar cannot show which came first. Recorded row: price came within 0.10*2.0859 = 0.2086 of the flip at 09:37 and the stop came first, so J is L.
• 5. K =INDEX($E$2:$E$1000,RANDBETWEEN(1,COUNT($E$2:$E$1000))) and L =IF(F2="support",B2-K2*D2,B2+K2*D2). Paste K and L as values at once (Copy, Paste Special, Values), because F9 redraws them; if ABS(K2-E2) is under 0.10, type another row’s E into K2; in the first rows, type K from your practice run’s E values. Recorded row: K 0.4917, L =272.0900-0.4917*2.0859 = 271.064.
• 6. Score M with the step 4 rules around L, on every row, including rows where J is blank; leave M blank if price never comes within 0.10*D2 of L by 15:30 ET. Recorded row: M W, the line reached its target.
• 7. For the support rows: n1 =COUNTIFS(F$2:F$1000,"support",J$2:J$1000,"?*"), p1 =COUNTIFS(F$2:F$1000,"support",J$2:J$1000,"W")/n1, n2 and p2 the same over M, and the lift in percentage points =(p1-p2)*100. Repeat with "resistance" for the other side. Recorded bounce: p1 =89/146 = 0.6096, p2 0.6412 (the study’s weighted rate; yours is unweighted), lift −3.16 points.
• 8. The smallest lift your log could see is =2.487*SQRT(p1*(1-p1)/n1+p2*(1-p2)/n2)*100. Recorded bounce: p1 0.6096, n1 146, p2 0.6412, n2 450 give 2.487 × 4.63 = 11.5 points (the study’s own was 10.6; 2.487 makes it a gap a test this size flags four times in five). A lift below your figure is “not detected”.

What a small log can say

The table is your step 8 formula with equal counts on both sides, at the study’s bounce win rates. Choose your count first and stop there. SPY and QQQ move together, so pooling is worth less than double.

Limits

• The trades were same-day, but the flip is a monthly level and a month-long hold is the more natural use; that was not tested. One bracket size, 0.35 expected moves, was used, no commissions, spreads or slippage are modeled.
• Every flip is a tier-0 reconstruction from each morning’s open interest and the prior close’s option prices, re-computed every five minutes; your sheet freezes it at 09:35. Through 28 March 2023 an option’s price is a stored last trade, or a modeled value where none traded (the fallback-premium era); after that date prices are native. Every session in this study falls in that first era. Prism’s build may differ from the study’s 90-day one. The result is about the rebuilt level, not a claim about the live Flip M.
• The monthly flip is among the least accurate of the tier-0 levels. Against a true intraday SPY map (204 sessions, October 2025 to August 2026) it was within $1 on 66.9% of marks, about as often as the daily flip (68.9%) and less often than the weekly flip (73.7%), the call wall (77.5%) and the put wall (83.3%). By month it ranged from 22% in March 2026, a quarterly-expiry month, to 85% in May 2026; the other quarterly months in that window scored 83% (December 2025) and 65% (June 2026). QQQ and the stocks were not checked, and the check does not reach this study’s 2018–2020 sessions, which also carry the older option prices. A real effect at the live monthly flip could be hidden wherever the rebuilt line sat elsewhere.
• The seven-name test is not independent of the first: SPY and QQQ supply 23% to 44% of its paired sessions.
• 2021–2024 was kept back for a confirmation test a situation had to earn by passing rules set in advance; none did, so it was never opened. A separate Resource, “Thirty-six level tests, none passed every rule”, covers all 36 combinations.
• A separate GEXfreak study of the 10:00 ET morning map, 2023–2024, saw the flip below price hold 56.3% of touches against 49.5% at random lines (+6.81 points, 167 touches). It was the best of 20 combinations with no adjustment for looking at 20, its “held” label mostly records the day’s direction, and this page’s test, on a different window and outcome, did not find it, so it is not a replication.
• The other three situations would need 633 to 1,363 events to reach 5 points, about 23 to 45 years of SPY and QQQ at the study’s pace.