backtest / metals / Gold · COPPERPALLADIUMPLATINUMSILVER
Gold the file, every extreme since 2006
★ FREE SAMPLE FILEseparate extremes on file (back-to-back extreme weeks count once), numbered on the chart, none hidden
3y percentile
84.6
cot index
64.1
z-score · 156w
0.09
weeks in zone
—
report as of
2026-08-04
the Gold report →is Gold stretched now? →
Every GOLD trade on file · 31 since 2006+
Historical analysis for information and education only, not advice, not a recommendation, not a solicitation. Results are model-based (front-month price proxies, weekly data), can contain errors, and would differ with costs, slippage and live execution. Past performance is no indicator of future results.
Every trade on file
31 extremes · back-to-back weeks count once · numbered on the tapearrows = the trade we take (numbered #): ▲ green = fade long (below the bars, crowd is short), ▼ red = fade short (above the bars, crowd is long) · ● = the turn, else the selected trade’s W13 window close marks where it ended · tap a numbered arrow to open its trade file
trade #31 · crowd long → fade short · confirm
report 2024-12-03 · entry 2024-12-09 · no turn within 13w
needed for a “turn”
4.2%
4w outcome
✗ -1.2%
8w outcome
✗ -8.5%
13w outcome
✗ -13.8%
26w outcome
✗ -28.7%
peak fade, best case, in hindsight
—
days to the turn
—
dip to sit through first
—
crowd first cut
week 2
top-4 concentration
96 pctl
VIX regime
32 pctl
dollar / rates 26w
↑ / ↓
● trend w/ crowd ○ divergence ○ OI expanding ● high vol ○ fast build ● H2
the exit is the clock, not a choice: every trade is graded at fixed gates (W1 / W2 / W4 / W8 / W13) and the window simply closes at W13, win or lose · no stops, no targets · 26w = aftermath, tracked but not part of the test
| # | report | entry | crowd → fade | 4w | 8w | 13w | 26w | turned | peak fade (hindsight) |
|---|---|---|---|---|---|---|---|---|---|
| #31 | 2024-12-03 | 2024-12-09 | long → short | ✗ -1.2% | ✗ -8.5% | ✗ -13.8% | ✗ -28.7% | — | — |
| #30 | 2024-09-03 | 2024-09-09 | long → short | ✗ -6.2% | ✗ -9.8% | ✗ -8.3% | ✗ -20.4% | — | — |
| #29 | 2024-06-04 | 2024-06-10 | long → short | ✗ -3.6% | ✗ -4.4% | ✗ -11.4% | ✗ -16.0% | — | — |
| #28 | 2024-03-05 | 2024-03-11 | long → short | ✗ -7.5% | ✗ -6.2% | ✗ -7.1% | ✗ -18.3% | — | — |
| #27 | 2023-11-28 | 2023-12-04 | long → short | ✓ +2.0% | ✓ +1.1% | ✓ -5.0% | ✗ -11.2% | 2023-12-11 | +2.0% |
| #26 | 2023-10-10 | 2023-10-16 | short → long | ✓ +1.8% | ✓ +3.5% | ✓ +6.1% | ✓ +21.8% | 2023-11-27 | +21.8% |
| #25 | 2023-07-18 | 2023-07-24 | long → short | ✓ +3.7% | ✓ +1.7% | ✓ -0.5% | ✓ -3.0% | 2023-09-28 | +3.7% |
| #24 | 2022-08-23 | 2022-08-29 | short → long | ✗ -6.1% | ✗ -4.6% | ✗ +0.8% | ✓ +4.7% | — | +4.7% |
| #23 | 2022-05-17 | 2022-05-23 | short → long | ✗ -0.6% | ✗ -7.2% | ✗ -4.8% | ✗ -5.7% | — | — |
| #22 | 2018-07-17 | 2018-07-23 | short → long | ✗ -2.9% | ✗ -2.1% | ✗ +0.9% | ✓ +7.1% | — | +7.1% |
| #21 | 2017-09-12 | 2017-09-18 | long → short | ✓ +0.8% | ✗ +2.5% | ✓ +3.7% | ✗ -2.4% | 2017-12-08 | +3.7% |
| #20 | 2016-09-27 | 2016-10-03 | long → short | ✓ +2.1% | ✓ +11.2% | ✓ +9.9% | ✓ +2.9% | 2016-11-14 | +11.2% |
| #19 | 2016-06-28 | 2016-07-05 | long → short | ✗ -1.7% | ✗ +2.3% | ✓ +5.7% | ✓ +10.9% | 2016-10-04 | +10.9% |
| #18 | 2016-03-29 | 2016-04-04 | long → short | ✗ -6.0% | ✗ +0.6% | ✗ -11.7% | ✗ -2.3% | — | +0.6% |
| #17 | 2015-11-17 | 2015-11-23 | short → long | ✗ +0.5% | ✗ +2.5% | ✓ +15.3% | ✓ +13.3% | 2016-02-01 | +15.3% |
| #16 | 2015-06-30 | 2015-07-06 | short → long | ✗ -7.0% | ✗ -3.4% | ✗ -2.1% | ✗ -6.3% | — | — |
| #15 | 2015-03-17 | 2015-03-23 | short → long | ✓ +1.8% | ✗ +2.1% | ✗ -0.8% | ✗ -3.0% | — | +2.1% |
| #14 | 2015-01-27 | 2015-02-02 | long → short | ✓ +6.2% | ✓ +7.9% | ✓ +7.3% | ✓ +14.8% | 2015-02-11 | +14.8% |
| #13 | 2013-11-19 | 2013-11-25 | short → long | ✗ -2.1% | ✗ +2.7% | ✓ +9.7% | ✗ +1.0% | 2014-02-24 | +9.7% |
| #12 | 2013-06-18 | 2013-06-24 | short → long | ✗ +2.9% | ✓ +5.8% | ✓ +3.0% | ✗ -6.3% | 2013-08-23 | +5.8% |
| #11 | 2013-01-29 | 2013-02-04 | short → long | ✗ -5.6% | ✗ -6.9% | ✗ -11.6% | ✗ -21.3% | — | — |
| #10 | 2012-07-10 | 2012-07-16 | short → long | ✗ +1.5% | ✓ +9.2% | ✓ +10.0% | ✓ +6.3% | 2012-09-07 | +10.0% |
| #9 | 2012-04-10 | 2012-04-16 | short → long | ✗ -5.5% | ✗ -2.3% | ✗ -4.9% | ✗ +5.6% | — | +5.6% |
| #8 | 2011-12-20 | 2011-12-27 | short → long | ✓ +8.0% | ✓ +11.0% | ✓ +4.4% | ✓ +1.4% | 2012-01-26 | +11.0% |
| #7 | 2011-08-02 | 2011-08-08 | long → short | ✗ -11.4% | ✓ +3.8% | ✓ -7.1% | ✓ -2.6% | 2011-09-26 | +3.8% |
| #6 | 2009-09-01 | 2009-09-08 | long → short | ✗ -4.5% | ✗ -9.1% | ✗ -12.8% | ✗ -13.0% | — | — |
| #5 | 2009-06-02 | 2009-06-08 | long → short | ✗ +3.1% | ✗ -1.0% | ✗ -3.9% | ✗ -17.5% | — | +3.1% |
| #4 | 2009-03-03 | 2009-03-09 | long → short | ✓ +7.3% | ✗ +3.9% | ✗ -1.4% | ✗ -6.8% | — | +7.3% |
| #3 | 2007-06-26 | 2007-07-02 | short → long | ✓ +1.5% | ✗ +1.2% | ✓ +11.0% | ✓ +30.9% | 2007-09-11 | +30.9% |
| #2 | 2007-01-09 | 2007-01-16 | short → long | ✓ +6.5% | ✓ +2.8% | ✓ +9.7% | ✗ +9.2% | 2007-02-23 | +9.7% |
| #1 | 2006-09-26 | 2006-10-02 | short → long | ✗ +1.0% | ✗ +6.1% | ✗ +1.0% | ✓ +13.0% | — | +13.0% |
✓/✗ = counter-crowd move ≥ the vol-scaled threshold within the horizon · green rows = the turns · peak fade = the best the fade paid at any horizon (known only in hindsight, not an entry claim) · worst-case runs live under timing & risk · data: release r2
What the file says
No extreme is active on the latest report. The full historical record, including every case that went nowhere, is on file below.
n=0 de-clustered episodes 2006→today · CI [—, —] · inference split: 0/0 stated on 2006–2018, 0/0 receipted on 2019→ · placebo p=— · FDR family: 930 cells, q=0.10
Anatomy
how the reversals here actually unfolded15 of 31 extremes here ended in a reversal within 13 weeks. The ones that turned took a median 7.6 weeks and first ran 1.0% further against the fade (the bet against the crowd). The ones that never turned cost a median 6.5% over 26 weeks, the price of fading a crowd that was simply right.
THE TURN FINGERPRINT · GOLD
its own 15 real turns vs 16 times the crowd was right · each condition on a 0–100 scale of its own 3-year history
● when it turned ● when it ran on · a map for your read, not a rule
Fingerprint at the extreme
When it turned, the wait was 7.6 weeks (median).
7.6 vs —Before turning, the fade first ran 1.0% against you: the shakeout you had to sit through.
1.0% vs —How dominant were the four biggest traders when it turned, vs when the crowd was right?
16 vs 29How many speculator accounts were crowding the extreme?
92 vs 79How hard were the commercials (the hedgers) leaning the other way?
94 vs 58Where did the small (retail) accounts sit?
19 vs 49Was the market already moving hard (realized volatility)?
61 vs 34How far did price sit from its 52-week extreme?
2.4% vs 2.5%What was the equity-fear backdrop (VIX)?
43 vs 49How often was the dollar rising into the extreme?
47% vs 63%How often did the crowd start cutting within 4 weeks?
100% vs 75%What did fading pay over 26 weeks: when it turned, and when the crowd was simply right?
+2.9% vs -6.5%green = the 15 that turned · red = the 16 that ran on · switch to the raw table for the full grid
This market’s own cases, taken apart: a map for your read, not a rule.
What marked the turns
measured at the actual top or bottom of every reversalFor each of the 489 reversals on file we located the actual top or bottom in the daily data, then measured which everyday events showed up first. The catch, printed with it: these events also fire in normal weeks. Only the crowded extreme makes them worth listening to.
| event after the top/bottom | marked the turn | typically after | move still ahead | false alarms |
|---|---|---|---|---|
| first close beyond the 5-day level, against the crowd | 100% of turns | 2 days | 78% | ~60× / year |
| first close beyond the 10-day level | 100% of turns | 4 days | 71% | ~43× / year |
| first report where the specs cut | 100% of turns | 4 days | 81% | weekly report |
| open interest drains (≥2% in a week) | 87% of turns | 4 days | 80% | weekly report |
this market’s 15 turns · What marked past turns, measured at the actual swing point of every reversal. The same price triggers fire all year · the extreme is what makes them worth listening to. Descriptive forensics, receipted across eras like everything else.
The record
reversals / episodes vs baseline, 2006 → todayA reversal = a move against the crowd too big to be this market’s normal noise: 0.75× its own volatility, scaled to each window, measured from the first open you could trade. Thin cells hold only a handful of cases, read them as war stories, not statistics.
CROWD LONG → FADE SHORT
| definition | 1w | 2w | 4w | 8w | 13w | 26w |
|---|---|---|---|---|---|---|
| COT index ≥95/≤5 | 3/14 vs 32% | 3/14 vs 33% | 2/14 vs 32% | 2/14 vs 32% | 2/14 vs 32% | 3/14 vs 27% |
| net %OI percentile ≥95/≤5 | 6/15 vs 32% | 5/15 vs 33% | 6/15 vs 32% | 5/15 vs 32% | 7/15 vs 32% | 5/15 vs 27% |
| z-score beyond ±2 | 1/6 vs 32% | 1/6 vs 33% | 1/6 vs 32% | 1/6 vs 32% | 1/6 vs 32% | 1/6 vs 27% |
| flush at the extreme | — | — | — | — | — | — |
| extreme + price divergence | 0/3 vs 32% | 0/3 vs 33% | 1/3 vs 32% | 1/3 vs 32% | 1/3 vs 32% | 2/3 vs 27% |
| extreme + 4-week price failure | 3/9 vs 32% | 3/9 vs 33% | 3/9 vs 32% | 5/9 vs 32% | 4/9 vs 32% | 2/9 vs 27% |
| commercial capitulation | — | — | — | — | — | — |
CROWD SHORT → FADE LONG
| definition | 1w | 2w | 4w | 8w | 13w | 26w |
|---|---|---|---|---|---|---|
| COT index ≥95/≤5 | 4/13 vs 36% | 3/13 vs 40% | 3/13 vs 44% | 3/13 vs 47% | 3/13 vs 51% | 3/13 vs 57% |
| net %OI percentile ≥95/≤5 | 3/16 vs 36% | 3/16 vs 40% | 5/16 vs 44% | 5/16 vs 47% | 8/16 vs 51% | 8/16 vs 57% |
| z-score beyond ±2 | 3/9 vs 36% | 3/9 vs 40% | 5/9 vs 44% | 4/9 vs 47% | 4/9 vs 51% | 7/9 vs 57% |
| flush at the extreme | 0/1 vs 36% | 0/1 vs 40% | 0/1 vs 44% | 0/1 vs 47% | 0/1 vs 51% | 1/1 vs 57% |
| extreme + price divergence | 3/5 vs 36% | 3/5 vs 40% | 3/5 vs 44% | 3/5 vs 47% | 2/5 vs 51% | 1/5 vs 57% |
| extreme + 4-week price failure | 3/12 vs 36% | 5/12 vs 40% | 5/12 vs 44% | 5/12 vs 47% | 4/12 vs 51% | 4/12 vs 57% |
| commercial capitulation | 3/14 vs 36% | 3/14 vs 40% | 7/14 vs 44% | 4/14 vs 47% | 7/14 vs 51% | 8/14 vs 57% |
verdict on every row: open, your call · nothing here survived FDR + out-of-sample confirmation in release r2
every rate ships with its sample size and baseline · the correction machinery (FDR, placebo tests, splits) lives in the methodology
Did any condition separate the movers from the rest?
Six conditions, written down before we looked at the results. Each one tested on this market’s cases and across all 43 markets, sample sizes shown, so you can weigh them yourself. None of them is a rule. The separations are thin. The read stays yours.
| scope | condition | side | with | without | raw p |
|---|---|---|---|---|---|
| GOLD | open interest expanding | short | 25% (n=8) | 40% (n=5) | 1.000 |
| GOLD | high-vol regime | short | 50% (n=8) | 0% (n=5) | 0.105 |
| GOLD | second half of year | short | 33% (n=6) | 29% (n=7) | 1.000 |
| all markets | trend with crowd (26w) | long | 34% (n=317) | 35% (n=62) | 0.884 |
| all markets | momentum divergence (13w) | long | 38% (n=48) | 34% (n=331) | 0.628 |
| all markets | open interest expanding | long | 34% (n=219) | 34% (n=160) | 1.000 |
| all markets | high-vol regime | long | 31% (n=144) | 36% (n=235) | 0.373 |
| all markets | extreme reached fast | long | 33% (n=227) | 37% (n=152) | 0.440 |
| all markets | second half of year | long | 35% (n=184) | 34% (n=195) | 0.914 |
| all markets | trend with crowd (26w) | short | 43% (n=327) | 35% (n=74) | 0.240 |
| all markets | momentum divergence (13w) | short | 43% (n=82) | 42% (n=319) | 0.900 |
| all markets | open interest expanding | short | 42% (n=211) | 42% (n=190) | 1.000 |
| all markets | high-vol regime | short | 39% (n=185) | 44% (n=216) | 0.363 |
| all markets | extreme reached fast | short | 43% (n=21) | 42% (n=380) | 1.000 |
| all markets | second half of year | short | 40% (n=193) | 44% (n=208) | 0.479 |
| all markets | trend with crowd (26w) | both | 39% (n=644) | 35% (n=136) | 0.497 |
| all markets | momentum divergence (13w) | both | 41% (n=130) | 38% (n=650) | 0.553 |
| all markets | open interest expanding | both | 38% (n=430) | 39% (n=350) | 0.882 |
| all markets | high-vol regime | both | 36% (n=329) | 40% (n=451) | 0.264 |
| all markets | extreme reached fast | both | 33% (n=248) | 40% (n=532) | 0.069 |
| all markets | second half of year | both | 37% (n=377) | 39% (n=403) | 0.659 |
verdict on every row: open, your call · the read stays yours
Timing & risk
measured the same way on every marketWhen the ones that turned, turned
7.6 weeks
median across the 15 cases that turned within 13 weeks · half of the turns arrived between 5.5 and 9.6 weeks after entry. Descriptive only: no confirmation trigger and no condition survived FDR + out-of-sample confirmation in release r1. Timing stats below describe past movers; they are not a validated entry edge.
What fading this market has cost
4.5%
median worst-case
9.7%
75th percentile
22.5%
worst on record
the run against you before any turn arrived, across 31 cases. Size for the worst row, not the median.
This file is free. The others are Pro.
BITCOIN, DOW EMINI, ETHER, NASDAQ EMINI are stretched right now, their files are one click away.