Part 3: The week and the news · Chapter 10 of 14 · 13 min
Exceptional moves
The widest 5% of hourly candles start at 55.1 $ (3.3 times the median), the widest 1% at 101.2 $ (6.0 times). They do not fall at calendar times: among the 30 widest five-minute candles only 4 are at 08:30 or at the Sunday reopening. The extreme days (11, from 251 $ up) all sit in three episodes between January and March 2026 (p < 0.001 for the clustering), and after one of them the following days stay wide for about a week (318 $, 240 $, 188 $... against 98 $).
What we call "exceptional"
An exceptional move is a candle, hour or day in the widest group. We prefer percentiles to fixed dollars: 50 $ in an hour was a lot when gold was worth 2,000 $, today it is less. In our sample (15 December 2025 - 6 October 2026, market open):
| Measure | Median | Widest 5% threshold | Widest 1% threshold | Maximum |
|---|---|---|---|---|
| H1 candle (4,865 hours) | 16.9 $ | 55.1 $ (3.3×; 244 candles) | 101.2 $ (6.0×; 49 candles) | 391.5 $ (29 January, 15:00 UTC) |
| Trading day (211 days) | 98 $ | 251 $ (11 days) | 717 $ (30 January) | |
| Five-minute candle | 5.4 $ | 100th widest: 50.6 $ (9.4×) | 213.6 $ |
In practice: a 55 $ hour comes about once every twenty market hours; a 100 $ one about once every hundred; a 250 $ day about once every twenty days (in our data, 11 days out of 211).
When they arrive
The first idea is that exceptional moves arrive at the times of data and the Fed. In part that is true: the widest 5% of hours fall in the two hours 08:00-10:00 New York time in 18% of cases, against the 9% you would get if they were evenly spread. But the full map shows they arrive at all hours:
| Block (New York time) | Share of the widest 5% | Share if uniform |
|---|---|---|
| 08:00-10:00 (data and stock-market open) | 18% | 9% |
| 10:00-14:00 | 22% | 17% |
| 14:00-16:00 (Fed and afternoon) | 9% | 9% |
| 18:00-02:00 (evening and Asia) | 34% | 35% |
| 02:00-08:00 (London) | 14% | 26% |
The block with the most excess is the American morning (double), then late morning. The evening and the Asian night weigh as much as their length: 34% in eight hours out of twenty-three. The only block quieter than expected is London's morning, 14% against 26%. For the widest 1% (49 candles) the picture is similar: 39% in the evening and Asia (18:00-02:00), 18% in London, 14% between 08:00 and 10:00, 20% between 10:00 and 14:00. An hour of over 100 $ gives no warning: the night is not a safe harbour.
Five-minute candles: the calendar is not enough
The 30 widest five-minute candles in the sample are a snapshot of sudden moves. A few: 29 January at 10:25 New York time, 214 $; 1 February at 18:00, the Sunday reopening, 166 $; 23 March at 07:05, 134 $; 1 February at 20:05, 126 $; 28 January at 18:35, 124 $; 4 September at 08:30, 110 $.
How many are at a calendar time (08:30, 10:00, 14:00, 14:30, 09:30, 08:20, 13:30 and the 18:00 reopening)? Four out of 30: two at 08:30 (4 September and 14 July) and two at 18:00, at the reopening (Sunday 1 February and Wednesday 28 January). Among the 100 widest, 18. The rest arrive at no known time. Another fact: the 30 candles fall in just 12 days, that is they are concentrated in the episodes we discuss shortly.
The calendar helps you avoid the normal-strong moves, those of data and the Fed; the biggest ones no, because they come without warning. Any risk plan must survive a 100-200 $ candle in five minutes that is not on the agenda.
Read the chart in words
H1 gold chart of 29 and 30 January 2026 with the widest candles marked and, below, the bars of each candle's range.
| When | What | Price |
|---|---|---|
| 29 Jan 2026 15:00 UTC | 29 January 10:25 NY: 214 $ in 5 minutes | 5,504.06 |
| 30 Jan 2026 18:00 UTC | 30 January 13:00 NY | 4,936.79 |
Times in UTC.
Read the chart in words
H1 gold chart of 23 March 2026 with the 07:00 New York candle marked and, below, the bars of each candle's range.
| When | What | Price |
|---|---|---|
| 23 Mar 2026 11:00 UTC | 23 March 07:05 NY: 134 $ in 5 minutes | 4,468.02 |
Times in UTC.
Extreme days come in clusters
Trading days with a range of at least 251 $ (the widest 5%) number 11. Here is when:
| Date | Range | Day's return |
|---|---|---|
| 28 January | 270 $ | +238 $ |
| 29 January | 513 $ | −40 $ |
| 30 January | 717 $ | −504 $ |
| 2 February | 441 $ | −178 $ |
| 3 February | 326 $ | +278 $ |
| 5 February | 269 $ | −188 $ |
| 6 February | 318 $ | +183 $ |
| 3 March | 373 $ | −231 $ |
| 19 March | 354 $ | −171 $ |
| 20 March | 251 $ | −160 $ |
| 23 March | 417 $ | −68 $ |
Three facts. First: they are all between January and March (3, 4 and 4 days). Since 23 March, in over six months, no day has exceeded 251 $. Second: they come in clusters: 7 pairs of consecutive days are both in the widest 5%, whereas with the days shuffled at random we would expect 0.5 (p < 0.001); 10 days out of 11 have another extreme day within three days. Third: the independent sample is small. The 11 days are really three episodes (28 January - 6 February, 3 March and 19-23 March), and eight days out of eleven are down days. This says nothing about future direction: a single episode, the one at the end of January, holds seven of them.
Read the chart in words
Daily gold chart from January to March 2026 with, below, the bars of the trading-day range and two lines at 98 and 251 dollars.
The widest day, 30 January, has a range of 717 $: from 5,397 to 4,893. It is not a single hour: the five hours with the most range are spread over the day (13:00 New York 196 $, 21:00 183 $, 04:00 182 $, 12:00 133 $, 03:00 121 $). This is a result too: the share of hourly range in the three widest hours is 25% on extreme days and 26% on others (if uniform it would be 13%). An extreme day is not an extreme hour: the whole day is wide.
After the extreme day
The most useful fact in the chapter is this. After an extreme day the range of the following days stays high and falls slowly:
| Day after | +1 | +2 | +3 | +4 | +5 | +6 | +7 | +8 | +9 | +10 |
|---|---|---|---|---|---|---|---|---|---|---|
| Median range | 318 $ | 240 $ | 188 $ | 188 $ | 178 $ | 157 $ | 130 $ | 155 $ | 101 $ | 102 $ |
| In multiples of the median (98 $) | 3.2× | 2.4× | 1.9× | 1.9× | 1.8× | 1.6× | 1.3× | 1.6× | 1.0× | 1.0× |
It takes about nine days to return to normal. We know the general reason: volatility clusters (chapter 6). Here you see how much: for a whole day the market is worth more than three normal days, for the three days after still almost double.
On direction, however, nothing. After an extreme day the next day goes in the same direction in 55% of cases (± 29 points: a coin), and the mean return of the five days after, measured in the direction of the extreme day, is −20 $ ± 161 $: inside the error, and the error is huge. After a half-day crash you cannot say whether the price will continue or bounce. You can say it will move a lot.
What to do with it
- Use a measure that adapts. A stop and size based on recent range (the ATR, or the range of the last three days) adjust to clustering on their own. A fixed dollar stop does not.
- After an extreme day, rescale for a week. Apply the table's multipliers: the next day stop and position by 3.2 times (stop 3.2× wider, position 3.2× smaller, at the same dollar risk), then 2.4×, 1.9× and so on. These are three episodes: use them as an order of magnitude.
- Do not trust the calendar for the biggest moves. The 30 widest five-minute candles are almost all off the agenda. Your plan must withstand a 100-200 $ candle in five minutes: with a 10,000 $ account risking 1% (100 $ per trade), a 5-ounce position with a 20 $ stop loses 100 $ if the stop is filled, and far more if it is jumped. The same as we saw for the weekend: size on the worst plausible move, not on the stop.
- Do not anticipate a bounce and do not chase a crash. The data do not say which. They only say the swings stay large.
- An extreme day is the start of an episode, not an end. Of the 11, eight were followed by another within three trading days. The first day, 28 January, was worth 270 $, just above the threshold: nothing set it apart from the others.
Procedure in four steps
- Compute the median range of the last 20 days and the threshold of its widest 5%.
- When a day exceeds the threshold, for the next five to nine days multiply stops by a factor between 3 and 1.5 and reduce positions by the same amount.
- Always size the position on the worst plausible move, not on the stop.
- Do not use the direction of an extreme day as a signal: the data do not support it.
Read the chart in words
H1 gold chart from 30 January to 3 February 2026 with four candles marked with the letters A, B, C and D.
| When | What | Price |
|---|---|---|
| 30 Jan 2026 18:00 UTC | Point A | 4,936.79 |
| 31 Jan 2026 12:00 UTC | Point B | 4,908.97 |
| 1 Feb 2026 23:00 UTC | Point C | 4,868.98 |
| 2 Feb 2026 13:00 UTC | Point D | 4,805.89 |
Times in UTC.
Exercises
Chapter exercises
1. On the chart the letters A-D are on four H1 candles between 30 January and 2 February 2026 (UTC hours). Which letter is the Sunday evening reopening candle, with the 166 $ five-minute candle at 18:00 New York time?
Read the chart in words
| When | What | Price |
|---|---|---|
| 30 Jan 2026 18:00 UTC | Point A | 4,936.79 |
| 31 Jan 2026 12:00 UTC | Point B | 4,908.97 |
| 1 Feb 2026 23:00 UTC | Point C | 4,868.98 |
| 2 Feb 2026 13:00 UTC | Point D | 4,805.89 |
Times in UTC.
2. After a day in the widest 5% the next day has a median range of 318 $, the second day 240 $ and the third 188 $, against 98 $ on a normal day. What do you do with the stop?
3. Out of 30 of the widest five-minute candles, only 4 are at a calendar time (08:30, 10:00, 14:00, 14:30, 09:30, 08:20, 13:30 or the 18:00 reopening). What do you take from it?
4. The extreme days are 11, in 7 consecutive pairs against 0.5 expected by chance, and all between January and March. How do you read it?
Have a question or want to share your exercise?
Post in the community, or join the free signals room on Telegram.
Educational content, not financial advice. Trading involves risk.