In Saty Swing mode, the ATR levels are built from the monthly 14-period Wilder ATR and the prior month's close (PMC) — they reset on the 1st and apply all month.
The Golden Gate is the zone from ±38.2% to ±61.8% of that monthly ATR away from PMC. A gate opens at ±38.2% and completes at ±61.8%. The trigger is ±23.6% (call trigger above, put trigger below) and the monthly pivot is the 0-ATR line — the prior month's close itself.
For each month and each direction we find the first day price reaches 38.2%, record which week of the month that is, then ask whether the gate completes by month end — and how far price retraced back toward the pivot before it did. Directions are scored independently; a month can open an upside gate, a downside gate, or both.
The two baselines, first. A gate opens when price reaches ±38.2% of the monthly ATR away from the prior-month close, and completes when price reaches ±61.8%. Of the 126 downside gates that opened with a fair window, 83 completed — a 66% completion rate. Of the 166 upside gates that opened with a fair window, 101 completed — a 61% completion rate. (Gates that opened with fewer than 5 trading sessions left in the month are excluded throughout — see the note below.) Every figure here is measured against those two base rates.
A shallow pullback behaves oppositely by direction. Of the downside gates that opened and then bounced back up to the put trigger (−23.6%) before completing, 39 of 45 still completed (87%) — well above the 66% downside base rate, so a bounce to the put trigger did not hurt. Of the upside gates that opened and then dipped back to the call trigger (+23.6%), only 32 of 60 completed (53%) — below the 61% upside base rate. This shallow-pullback split (it spares puts, it hurts calls) is the finding that survives the daily-bar sensitivity check.
A deeper pullback to the pivot erases the edge both ways. The monthly pivot is the prior-month close (the 0-ATR line). Of downside gates that retraced all the way back to the pivot, 15 of 23 completed (65%) — essentially equal to the 66% downside base rate. Of upside gates that retraced to the pivot, only 6 of 22 completed (27%) — under half the 61% upside base rate. And once price crossed the pivot to tag the opposite trigger, completion collapsed: 3 of 32 downside gates (9%) and 3 of 18 upside gates (17%). The pivot is the invalidation line, not a discount.
Separately (Question A below), among gates that had a fair window the completion rate is still highest for early-month opens. Note: the exact retrace percentages above are sensitive to a daily-bar limitation explained in Method & caveats.
The test. Before a downside gate opens (−38.2%), price must first pass the put trigger (−23.6%). We measured how many trading sessions that traversal took, then asked whether the gate went on to complete (−61.8%) by month end. Of the 126 downside gates that opened with a fair window (clock-truncated opens excluded — see the note below), 83 completed — the 66% base rate. Splitting by traversal speed:
It is not just "opened earlier." Fast gates do tend to open earlier in the month (median 18 sessions left vs 14 for slow ones), and the parent study shows earlier opens complete more simply because more days remain. But the speed edge survives holding days-left fixed: among gates with more than 10 sessions left, fast still beat slow 70% (54 of 77) to 63% (20 of 32). And the same pattern shows up — stronger — on the upside gate (fast ≤1 session completes 73% (40 of 55) vs 55% (61 of 111) slow), independent confirmation that momentum into the gate, not the calendar, is doing the work.
Stacking with weekly context. The other condition that helped was a stretched-up weekly Phase Oscillator (≥61.8) when the downside gate opens — a reversal from a weekly top, which completed 73% (19 of 26). Combining the two: a fast traversal while the weekly PO is extended-up completed 77% (13 of 17), versus 58% (22 of 38) for a slow traversal with a non-stretched weekly. Thin cells (n=17), but consistent with both signals on their own.
Exploratory follow-up · same clock-truncated exclusion as the rest of the page (gates with <5 sessions left in the month at open are dropped). Daily indicators read from the bar before the open day and weekly from the last week ending before it (no look-ahead); same daily-bar ordering limit as below applies to the same-day/gap bucket. Generated from backtest_swing_gg_wom_predictors.py.
n= on the right is how many gates are in that bucket. The grey tick marks this direction's base rate (the percentage of all opened gates that completed). A bar right of the grey tick = retracing that far raised the completion rate; a bar left of it = retracing that far lowered the completion rate.completed-and-continued / completed count, because these percentages are conditional on completing first, so later-week rows rest on only a handful of completions. A 0.0% there means "none of a handful," not "impossible."Data: FirstRateData SPX cash index, daily bars. Levels from the prior month's close + one-month-lagged monthly Wilder ATR(14), matching the lag convention used elsewhere on this site. Warmup leaves 293 analyzed months (2002-01 → 2026-04). A gate opens in 177 of them on the upside and 139 on the downside; after the exclusion below, 166 upside and 126 downside gates remain in the reported rates.
Week of month = calendar week of the open day: days 1–7 = week 1, 8–14 = week 2, 15–21 = week 3, 22–28 = week 4, 29–31 = week 5.
⚠ Clock-truncated opens are excluded — and why. A gate that first opens with fewer than 5 trading sessions left in the month cannot fairly reach 61.8% before the monthly level resets on the 1st, so its non-completion measures the calendar, not the setup. These opens were dragging the late-week completion rates toward zero (their by-month-end completion was 9% on the upside / 39% on the downside, but 36% / 62% within a fair 5-session window). We drop them everywhere: 11 upside and 13 downside gates (all week-4 and week-5 opens). This raises the base completion rates from 58% to 61% (up) and 63% to 66% (down), removes the artificial week-5 collapse, and — because every kept gate now has a full in-month 5-session window — guarantees the within-5-day tick never exceeds the by-month-end bar.
Retrace depth is the deepest bounce back toward the pivot measured from the open day up to (but not including) the completion day — so the retrace genuinely precedes completion. Buckets are nested by depth: same-side trigger (±23.6%) → pivot (0 line) → opposite trigger. "No bounce" = never retraced as far as the same-side trigger.
Open / complete detection uses daily highs and lows, so within a single day the order of a high vs a low is unknown; gates that open and complete the same session are split out separately (no retrace window). Each direction is scored independently per month.
⚠ Daily-bar limit on the retrace ladder (Question B): the retrace window includes the open day, whose intraday high and low may have occurred before the gate opened — so the bucketing is approximate. Excluding the open day from the window reclassifies 54 of the 292 fair-window events, and the completion rates in the thin buckets move materially: the upside-trigger bucket goes from 53% to 35%, the upside-pivot bucket from 27% to 11%, the downside-trigger bucket from 87% to 76%, and the downside-pivot bucket from 65% to 56%. What survives that sensitivity check — and the part you should trust — is the shallow-pullback asymmetry (a bounce to the same-side trigger is tolerated on the downside but lowers the upside completion rate) and the opposite-trigger collapse (under 20% completion either way). The exact pivot and opposite-trigger percentages are noisy on daily data; the definitive test needs intraday (1-minute) bar ordering.
Independent check: re-running this exact analysis on Yahoo Finance ^GSPC daily bars (a different data vendor) over the same window reproduced every completion rate to within about 1 event per bucket — so the numbers are not an artifact of one data source. This does not address the daily-bar ordering limit above, which applies to any daily OHLC source.
The clock confound: week-of-open and sessions-left-in-month are nearly the same axis, which is exactly why the clock-truncated opens are excluded above. Among the gates that remain (all with a fair 5-session window) the late-month completion collapse is gone; the residual early-month edge — and the upside week-3 dip — are real, not mechanical. The retrace ladder is the cleanest time-independent signal.
Caveats: the pivot and opposite-trigger retrace buckets run thin (n<50, flagged faded) — the direction of the effect is robust, the exact percentages are noisy. SPX is the cash index (no overnight gaps inside the month's range the way a tradable gaps), so map onto SPY/ES with care.