Gold's third day down: sell the bounce at $3,900

Gold is not drifting. It is being repriced. Spot printed $3,881.70 at 18:36 UTC on Thursday, down 2.7%, its lowest level since October 2025 (CNBC), and Kitco's mid-session print of $3,903.98 at 13:30 UTC shows the day was a one-way auction rather than a single flush (Kitco). The desk's own series tops out at $3,995.96 in the 01:00 UTC bar, floors at $3,853.56 in the 15:00 bar and last trades $3,883.63 — a $142 session range against a 20-day mean of $63.

That is the third straight lower close and $280 off Monday's high. The bias is not a forecast; it is the tape.

What did the damage

• The 10-year went back above 5.00%, the first time in three weeks, and FedWatch put 94% odds on a 25bp December hike with 6% on a half-point (Bullionvault). CNBC's read of the same market was 85% for 25bp, the difference being when the screen was captured — either way December is a hike meeting, not a cut meeting.
• Oil is now an inflation input, not a gold input. Trump said the US does not need a deal with Iran, Brent rose more than 2%, and the market read higher crude as higher inflation and therefore higher rates. "If crude oil prices spike higher on this news, that's negative for gold from an inflation perspective," Jim Wyckoff of American Gold Exchange told CNBC. That inversion is why the war premium stopped helping bullion.
• The dollar is at a 19-month high (CNBC), its strongest since April 2025 on Bullionvault's trade-weighted read.
• Jobless claims printed 200,000 against a 195,000 consensus, and gold made fresh session lows on the release (Kitco). A $5,000 miss in a weekly series is noise; the reaction to it says the positioning was long and nervous.

The structure the plan is built on

Four consecutive lower daily highs on the desk's series: $4,133.99 on 9 November, $4,127.11, $4,058.60, then $3,995.96 today. Nothing in that sequence argues for buying a dip; it argues for selling a bounce.

The bounce has a home. Thursday's 08:00-11:00 UTC consolidation sat at $3,922-$3,950, and that shelf broke at 13:00 UTC. The double top inside it — 10:00's $3,949.90 and 11:00's $3,950.23 — is the level that has to hold for the short to be right.

The entry is deliberately close to the market. The last two sessions both opened and bounced in their first two hours: 11 November ran from a $4,033.95 open to $4,058.60 by 05:00, and 12 November opened $3,974.28, dipped to $3,966.00 and then printed $3,995.96 at 01:00. That is a $22-$25 early bounce as the pattern, so a $3,900 limit sits exactly where the market has been going rather than on a poster $45 away.

Yesterday's order is the cautionary case: the $4,004 sell limit never filled, because the highest print in its whole window on the desk's series was $3,995.96. The direction was right and the order was a spectator. Hence the tighter entry and the explicit open conditions below.

The plan

• Bias: short while $3,950 caps the bounce
• Entry: $3,900 sell limit
• Stop loss: $3,958
• Take profit 1: $3,842 — 1.0R, take 50% off
• Take profit 2: $3,784 — 2.0R
• Risk-reward: 1:1 to TP1, 1:2 to TP2, blended 1.5R if both fill
• Do not take it if: the 00:00 UTC open is above $3,940 (the bounce is already spent and the stop sits inside the opening range) or below $3,855 (a fresh low at the open leaves nothing to sell into — a retest order there is a spectator) or the limit has not filled by 11:00 UTC
• Time stop: flat at 13:15 UTC

Instrument is XAUUSD spot; the window is Asia and London, 00:00-13:15 UTC on 13 November.

The stop distance is $58, which is 0.81 x ATR(14) of $71.76. That is wider than the desk used to publish, and it is the point: the earlier signals carried $19-$23 stops against sessions averaging $71, and three of them were decided by less than $6. The stop also clears the 11:00 bar high by $7.77, a hair over the 0.1 x ATR buffer, so a feed that prints a few dollars higher does not decide the outcome. The reference series for both the ATR and the levels is Binance PAXGUSDT daily and hourly klines; Investing.com is behind a Cloudflare challenge from this host and cannot be used, so the series has changed and the ATR with it.

TP1 at $3,842 sits $11 under Thursday's low. It is a test of that low, not a demand that it breaks — which is the whole lesson of the last month, where first targets were parked below the session low and never got there.

The arithmetic you have to accept

A blended 1.5R breaks even at a 40% strike rate. The caveat is the one that has been costing this desk money: if TP1 fills and the runner is scratched at entry, winners pay +0.5R against -1R losses, and that book needs 66.7%. The measured average win across the published sample is +0.82R, which needs 49.1%. Take the 50% at TP1 into dollars and keep the remainder only while $3,855 keeps rejecting.

Sizing follows from the $58 stop, and it is unforgiving on small accounts. At 0.5% of equity a $10,000 account risks $50; at $100 a point per lot that is 0.0086 lots, under the 0.01 minimum, so that account cannot take this trade inside its own rule and should sit it out. A $100,000 account risks $500 and carries 0.086 lots. The position-size calculator runs the same arithmetic for any balance.

What makes this wrong

• An hourly close back above $3,950. That breaks the lower-high sequence, and the stop is already there.
• A dovish surprise at 13:30 UTC. October PPI lands at 8:30 ET on Friday, and the prior prints were 1.3% headline and 0.7% core (Investrade calendar). A soft number in a market this short is a $40 squeeze — which is why the time stop is 15 minutes before the print and there is no position through it.
• A real-yield reversal. Every leg of this move is a rates move; if the front end stops rising, the reason to be short stops with it. The desk tracks that argument on /topics/real-yields.

Signals are graded on the public ledger at /track-record whether they win or lose. The setup logic sits on the /signals board, and the CPI print that started this repricing is written up in October CPI runs hot.

Not financial advice. Levels are for the 00:00-13:15 UTC session on 13 November 2026 and expire at the time stop; a fill outside the stated open conditions voids the plan. Prices on this page come from Binance PAXGUSDT klines and the wire prints linked above, which can differ by a few dollars from any single broker feed.

Sources: CNBC gold, Kitco claims and gold, Bullionvault rate hawks, Investrade weekly calendar.