Does price end monthly expiry Friday near the walls?

The short answer

No, not that we could detect. On monthly options expiry Fridays from May 2018 to December 2024, price did not finish the last half hour (to the last minute before the closing auction) nearer the closer call or put wall than random lines would, in any of six tests on SPY, QQQ and a pool of six large stocks. The earlier version, on a different level, found nothing either.

In plain counts, 12 of 46 SPY months finished within a tenth of an ATR of the closer wall, and 12 started that close. ATR is the average size of a day’s move.

This does not say expiry has no effect. SPY’s test could only have seen a shift of 20.7 percentage points or more in landing share, QQQ has only 31 months, and months with both walls on one strike were left out (see Limits).

What to do with this

• Treat the closer wall on monthly expiry Friday as a reference level, not a magnet: price ended no nearer to it than random lines would. If both walls sit on one strike, treat that day as unknown; the six tests left those months out.
• When a chart or post says price “pinned” to a wall, or a wall “held” a share of the time, ask how far the wall started from price and what random lines at the same distances would have scored. A wall near price ends near it unless price moves a lot (median SPY move in these 29 minutes: 0.10 ATR): SPY walls within 0.1 ATR (12 months) averaged +0.281 against chance, walls 0.5 to 1 ATR away (13 months) −0.367.
• Until your own log holds about 50 Fridays, keep a wall as a reference, not a prediction: at that size a log can only pick out an average gap from chance of about 0.12 (12 percentage points) or more.

How we tested

A monthly expiry Friday is the third Friday of the month, when the standard monthly options expire. At 15:30 ET we took SPY’s price and the nearer wall in force in the rebuild at that clock (the 15:25 rebuild), and measured how far price was from that wall, in ATR, at the last trade of the 15:59 minute. For SPY the median month started 0.22 ATR from the nearer wall ($1.07) and finished 0.24 ATR from it.

On 19 October 2018 SPY’s nearer wall was 0.0165 ATR from price; price moved 0.1743 ATR away and still finished 0.1908 ATR from it, so the landing share was 0.9778: price finished farther from 44 of the other 45 lines than from the wall. With no pull the average share is 0.5, and the figure reports it minus 0.5.

Six tests were fixed before any result was read: SPY, QQQ and a pool of six large US names, each against chance and against the ordinary Friday (the monthly Friday’s share minus the closest ordinary Friday’s).

What changed from the earlier version

The earlier version’s level was the nearest-expiry wall built from the prior close, usable from the 09:30 open and held all session, on 13 names with no coverage gates. This version reads the wall in force in the rebuild at 15:30, on six. They are different constructions, not a correction: a gap between them shows how much the answer depends on the level source.

Every test is still not detected. On the months both levels measured, the swap moves the average by −3.3 to +11.2 percentage points and every 95% range includes zero. On the months the chance tests share, the levels name the same strike on 40% to 55% and the same kind of wall on 77% to 87%.

Check it yourself with Prism and a spreadsheet

One row per monthly expiry Friday of one symbol; write A, E and the two walls before 15:30, and fill B and C after the close. To practise, copy columns A to E of the SPY table from the figure (under “Tables, method and data source”) into a sheet: the formulas then return the study’s SPY numbers.

The steps, and what a small log can say

• 1. The days. Start with the next monthly expiry Friday. Leave a Friday out, and list it separately, if both walls sit on one strike, the nearer wall is more than 1 ATR from price, or the session closes early (there is no 15:30 bar); the study kept 46 of 80 SPY months.
• 2. Columns A to E. Read the walls once, just before 15:30, and note the time. A smoothed (Wilder) ATR rescales F and G; use one method throughout.
• 3. Distance and move. In F2: =ABS(D2-B2)/E2. In G2: =SIGN(D2-B2)*(C2-B2)/E2, positive when price moved toward the wall’s side. Worked line, SPY 20 December 2024: B 591.79, C 590.87, E 4.85525, D 590.00 (the put wall, 1.79 below price; the call wall sat above), so F is 0.3687 and G is +0.1895.
• 4. The simplest check. =AVERAGE(G2:G47) is the average move toward the wall’s side and =COUNTIF(G2:G47,">0") counts the months price moved toward it. The 46 SPY months give −0.016 ATR and 24 of 46: no lean toward the wall.
• 5. Landing share. In H2, filled down: =(SUMPRODUCT(--(ABS(G2-$F$2:$F$47)>ABS(G2-F2)))+0.5*(SUMPRODUCT(--(ABS(G2-$F$2:$F$47)=ABS(G2-F2)))-1))/(COUNT($F$2:$F$47)-1). Change 47 to your last row; the −1 leaves out the row’s own line. Each other row is a random line at its own distance F; the formula counts those price finished farther from (ties half). Worked line: price finished 0.1792 ATR from the wall and farther than that from 17 of the other 45 lines, so H = 17 ÷ 45 = 0.3778. With the 46 SPY rows, H2 reads 0.5333.
• 6. Against chance. =AVERAGE(H2:H47)-0.5 returns +0.004 for the 46 SPY rows, the study’s SPY figure, and =COUNTIF(H2:H47,">0.5") counts the months above chance (23 of 46).
• 7. Read it against the table. Your log’s spread is =STDEV(H2:H47); one symbol’s months came to about 0.31. Your 95% range is your average ± =1.96*STDEV(H2:H47)/SQRT(COUNT(H2:H47)); the smallest gap your log could see uses 2.80 in place of 1.96.
• 8. Pace and tries. Usable SPY Fridays come to about 7 a year, so 20 takes about 3 years and 50 about 7; ordinary Fridays come about 4 times as fast, on a separate sheet. Several symbols on one Friday are not independent events, and every extra cut of your log is another try.

Limits

• These levels are rebuilt at each five-minute mark; Prism by GEXfreak keeps no five-minute history of its levels. A tier-0 rebuild uses that morning’s open interest and the prior close’s option prices, re-evaluated at that mark’s market price. Only SPY has a true intraday map (from August 2025), used just to check accuracy. Within $1 of it, tier-0 matched the call wall 77.5% of the time, the put wall 83.3%, the daily flip 68.9%, the weekly flip 73.7% and the monthly flip 66.9%. QQQ and stock accuracy is unmeasured. A log made from the Prism chart tests Prism’s walls, which can differ, and on an expiry Friday the expiring options change through the day.
• 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. That era holds most rows: 32 of SPY’s 46 months, 24 of QQQ’s 31 and 55 of the pool’s 76. The figure shows both eras, and quarterly against other months, for every test.
• There is one decision clock, 15:30 ET. The window stops at the last trade of the 15:59 minute, so the closing auction, where a closing pull would finish, is not measured. It ends on 31 December 2024; no later data is used.
• The smallest shift in landing share a test could see was 12.5 to 36.1 percentage points, depending on the test (landing share runs from 0 to 1, so 0.207 is 20.7 points), so a smaller pull could be there unseen. QQQ has only 31 months, just above the 30-month floor for a test, and the ordinary-Friday tests are weaker still, since weekly options also expire on ordinary Fridays.
• Only months with two different walls were measured: SPY 46 of 80, QQQ 31 of 80 and 178 of 480 pool name-months. The rest had both walls on one strike (32, 49 and 272), a nearer wall over 1 ATR away (1, 0 and 3) or no usable data (1, 0 and 27). Added back, the shared-strike months leave the averages near zero (+0.016, +0.018, +0.004).
• The pool is six names (AAPL, AMD, INTC, MSFT, NVDA and TSLA), about two per month row; seven of the earlier 13 fall under the 60-month rule once coverage gates apply, so its 178 name-months are not independent events.
• For SPY and QQQ these are the same months, on the same rebuilt level, that the earlier study had already scored once without a test, so a detection would not have been independent evidence. The six tests are also part of 40 across four studies; about 2 would look like detections by chance.
• The era, month-type, year and distance tables are descriptive: no p-values, no correction for the many cells. Of the 48 ranges in the figure’s era, month-type and year views, 2 exclude zero, about what chance gives.
• This is not a live rule: the random lines borrow distances from later Fridays too, and landing share cannot see a steady drift toward one side (step 4).