MEASURED · 11 CPI RELEASES
How much does gold move when US CPI comes out?
In the 11 US CPI releases from August 2025 to July 2026, gold’s 8:30 a.m. ET minute (12:30 or 13:30 GMT) moved a median $15.81 from high to low. That is about 4 times the same minute on an ordinary weekday. Within 15 minutes price traveled a median $20.09 from where it stood at 8:29, and never less than $11.38.
The first minute did not predict the next hour. Once it closed, price kept going the same way an hour later in 6 of 11 releases.
Which releases count, and the one that never came
CPI has no fixed weekday, so the dates come from the BLS CPI release archive, and each release page says it was embargoed until 8:30 a.m. ET.
The 2025 and 2026 lapses in federal appropriations changed four of them. The BLS rescheduled the September 2025 CPI to October 24. It canceled the October 2025 CPI, due on November 13, because it could not collect that month’s data; that release never existed, so nothing stands in for it. The November 2025 CPI moved from December 10 to December 18, and the January 2026 CPI from February 11 to February 13 (BLS revised release dates). Each release is measured on the day it came out.
8:30 ET is 12:30 GMT under US daylight time and 13:30 GMT in winter. US clocks went back on November 2, 2025 and forward on March 8, 2026, so the December, January and February releases sit at 13:30 GMT. The bar clock on this Exness feed is GMT, and the CPI days confirm it: the 8:30 ET minute was the largest of its whole day in 7 of the 11 releases, and in the top three in 9.
Every release, measured
| Release (8:30 ET) | CPI for | First minute | vs ordinary 8:30 | Reach in 15 min | Minutes elevated |
|---|---|---|---|---|---|
| Aug 12, 2025 | July 2025 | $13.23 | 3.3× | $14.83 | 5 |
| Sep 11, 2025 | August 2025 | $21.54 | 5.5× | $20.16 | 5 |
| Oct 24, 2025 | September 2025 | $32.66 | 8.3× | $32.16 | 16 |
| Dec 18, 2025 | November 2025 | $15.20 | 3.8× | $17.68 | 4 |
| Jan 13, 2026 | December 2025 | $13.38 | 3.4× | $19.05 | 3 |
| Feb 13, 2026 | January 2026 | $34.52 | 8.7× | $34.39 | 24 |
| Mar 11, 2026 | February 2026 | $12.06 | 3.1× | $17.97 | 9 |
| Apr 10, 2026 | March 2026 | $15.81 | 4.0× | $20.09 | 1 |
| May 12, 2026 | April 2026 | $9.59 | 2.4× | $11.38 | 8 |
| Jun 10, 2026 | May 2026 | $33.82 | 8.6× | $32.21 | 14 |
| Jul 14, 2026 | June 2026 | $60.31 | 15.3× | $73.21 | 18 |
First minute: high minus low of the 8:30 ET bar. Ordinary 8:30: the median of the same minute on 249 other weekdays ($3.95). Reach: the farthest price got, up or down, from the 8:29 close between 8:30 and 8:44 ET. Elevated: consecutive minutes with a range of at least twice the median minute ($3.59).
Bigger than what?
Against the yardstick our NFP study used, the median of all 356,223 one-minute bars ($1.80), the median CPI minute was 8.8 times normal and the largest, on July 14, 2026, 33.6 times.
That flatters CPI. The median minute includes the quiet Asian night, and 8:30 ET is busy on any weekday: the Labor Department’s weekly jobless claims come out then on Thursdays (DOL claims release). On the other 249 weekdays the same minute had a median range of $3.95. The median CPI minute was 4 times that, and 10 of the 11 beat the busiest tenth of ordinary 8:30 minutes ($9.80 or more). The exception was $9.59 on May 12, 2026.
CPI was still not the biggest 8:30 minute of the year. That was $63.53 on February 11, 2026, when the BLS released the January 2026 jobs report.
How far price runs in the first 15 minutes
For a position held into the release, what matters is how far price runs from where it stood just before: the farthest move from the 8:29 close, either way, from 8:30 to 8:44 ET.
| Within 15 minutes | CPI releases (n = 11) | Ordinary 8:30 (n = 249) |
|---|---|---|
| Median reach | $20.09 | $8.02 |
| More than $10 | 11 of 11 | 36.1% |
| More than $20 | 6 of 11 | 11.2% |
| More than $5 both ways | 3 of 11 | 12.9% |
The smallest reach was $11.38 (May 12, 2026), the largest $73.21 (July 14, 2026). In 3 releases price went more than $5 each way, so a buy stop and a sell stop $5 either side of the 8:29 price would both have filled. Elevated volatility lasted a median of 8 minutes, at most 24 (February 13, 2026).
Does the first move keep going?
Measured as our NFP study measures it, every time: an hour after the release, price was still on the release minute’s side of that minute’s open in 11 of 11 releases, against 53.9% of random minutes and 60.2% of ordinary 8:30 minutes.
That test has a head start built in. A $15 first minute needs a $15 reversal to fail it, while a median minute’s whole range is $1.80. The useful question is what happened after the first minute closed, the earliest anyone could act.
| After the first minute | CPI releases | Random minutes | Ordinary 8:30 |
|---|---|---|---|
| 5 minutes later | 5 of 11 | 49.3% | 51.4% |
| 15 minutes later | 6 of 11 | 49.2% | 51% |
| 60 minutes later | 6 of 11 | 49.4% | 49.4% |
From that close, price kept going the same way in 6 of 11 releases an hour later; across all 11, the median move in the first minute’s direction was a further $3.44. The 95% interval on 6 of 11 runs from 28% to 78.7%. That is a coin flip, not a signal: once the first minute closed, its move was already in the price.
The spread changed in February, CPI or not
MT5 stores one spread value per one-minute bar, so the widest tick can be wider than shown here. In the release minute the stored spread was 160 points ($0.16), the median of the whole archive, in 7 of 11 releases. The widest spread from 8:25 to 8:40 ET splits the sample: at most 160 points at all 5 releases to January 2026, 280 to 360 at all 6 from February.
Ordinary mornings broke the same way: that window’s spread topped 160 points on 0% of other weekdays before February 2026 and on 98.4% after. So it looks like a change in this account’s spreads from February 2026, not a CPI effect. Our spread and volatility by hour study reports a flat $0.16 median; that is a median over the whole year, and a change that starts halfway through a year does not show in it.
What to do with this if you run a bot
Put the release in your platform’s clock. On a GMT server CPI is 12:30 in summer and 13:30 in winter. On a server at GMT+2, add two hours; many servers move to GMT+3 with summer time, so check your server’s offset on the release day. The BLS CPI schedule lists dates more than a year ahead.
Pause new entries from 8:25 to 9:00 a.m. ET. The longest elevated run, 24 minutes, ended before 8:55. A Fed decision needs a longer pause, because the 2:30 press conference follows the 2:00 statement: see how much gold moves on Fed decision days.
Decide about open positions by 8:29. Every release moved price more than $10 within 15 minutes, and 6 of 11 more than $20. A stop does not prevent that loss; it sets roughly where the trade ends, and a fast minute can fill it worse. If you hold through, size for $73.21, not the $20.09 median.
Do not chase the first minute. After it closed, the direction held an hour later 6 times in 11.
Check the log, not the settings. On the next CPI day, confirm in your EA’s log that it stopped opening trades by 8:25 ET in server time. A news filter that reads the calendar in the wrong time zone looks right in its settings and pauses at the wrong hour. That goes for any bot, the Tech Kick Trading Bot included.
Trading from the US? Read whether US traders can use MT5 trading bots first: the instrument measured here is a spot gold CFD.
Method and limits
356,223 one-minute XAUUSDm bars from an Exness MetaTrader 5 feed, August 1, 2025 to August 4, 2026. The baseline, the twice-the-median rule for “elevated” and the first continuation test are those of our NFP release study. Here the dates come from the BLS rather than a calendar rule, 8:30 ET is converted with the IANA time zone database, and the ordinary-8:30 control and the after-the-first-minute test are added. The random control is 30,000 minutes drawn with a fixed seed.
11 events is a small sample. Two fell on a Thursday (September 11, 2025 and December 18, 2025), when jobless claims normally share the minute; what claims do to gold on their own is measured separately. One symbol, one broker, one year, a spot CFD rather than futures. The CPI surprise, actual against forecast, is not measured, so nothing here says why a release went up or down. Every figure comes from one script, research-gold-cpi-release.py, run on the same bar archive as our other studies.
Common questions
What time is the US CPI release in GMT?
8:30 a.m. Eastern: 12:30 GMT while the US is on daylight saving time, 13:30 GMT in winter. In the year measured, US clocks went back on November 2, 2025 and forward on March 8, 2026. On a broker server at GMT+2, add two hours; many servers move to GMT+3 with summer time, so check your server's offset on the release day.
Was there a US CPI release in November 2025?
No. The October 2025 CPI, due on November 13, was canceled because the lapse in federal appropriations stopped data collection, and it was never published. The November 2025 CPI came out on December 18, 2025 at 8:30 a.m. ET.
Does gold go up or down when CPI comes out?
Either. 9 of the 11 release minutes here were up, but the direction depends on the number against what the market expected, which this study does not measure, and 11 releases are too few to call a bias.
Should a trading bot trade the CPI release?
Nothing in this sample says it should. The first minute ranged from $9.59 to $60.31 with no way to know which in advance, and after it closed the direction held an hour later in only 6 of 11. The data supports a rule about exposure instead: flat, or sized for a $73.21 move, by 8:29 ET, and no new entries until about 9:00.