Part 3: The week and the news · Chapter 6 of 14 · 9 min
The days of the week
In gold's data the five market days look alike: neither the range (p = 0.80) nor the return (p = 0.76) depends on the day of the week. What changes is the candle of the 08:30 data, which is worth 1.5× on Monday and 3.1× on Friday, and wide days come in clusters: after a day in the widest 5% the next day has a median range of 318 $, against 94 $.
Why people think about the day of the week
Many ideas circulate in trading about the day of the week: "Monday is slow", "Friday closes higher", "Wednesday moves". An idea like this can have a real cause: macroeconomic data come out on fixed days (on Thursday the weekly US jobless claims, on the first Friday of the month the employment report), the Fed decides on Wednesdays, the market reopens on Monday after the weekend. But a possible cause is not enough: it has to be measured.
Here the "day" is the COMEX trading day, from 18:00 New York time of the previous day to 17:00. Monday therefore starts on Sunday evening. There are 211 days: 43 Mondays and 42 of each of the other days.
The day changes neither the size nor the direction
| Day | Median range | Mean range | In % of price (median) | Mean return ± error | Up days |
|---|---|---|---|---|---|
| Monday | 97.2 $ | 113.2 $ | 2.14% | −8.8 ± 20.5 $ | 21 (49%) |
| Tuesday | 101.1 $ | 115.6 $ | 2.17% | −1.0 ± 26.7 $ | 21 (50%) |
| Wednesday | 111.4 $ | 120.5 $ | 2.32% | +10.6 ± 25.1 $ | 24 (57%) |
| Thursday | 94.3 $ | 122.0 $ | 2.26% | −13.0 ± 22.1 $ | 19 (45%) |
| Friday | 96.2 $ | 119.9 $ | 2.16% | −0.5 ± 32.4 $ | 26 (62%) |
The numbers vary, but they are all inside the errors. To see whether the difference between days is bigger than what chance would produce, we shuffle the day labels 4,000 times and recompute: the effect of the day on the range has p = 0.80 (getting it equal or larger by chance is a normal event), that on the return p = 0.76. Up days are 53% of the total (±7 points): Friday, with 26 out of 42 (62%), seems to stand out, but with 42 days the error is ±15 points.
In practice: in our ten months there is no day on which gold moves more, nor one on which it rises more. Whoever picks the "best" day is reading noise.
What does change: the data calendar
One thing really does change from day to day, and it is the 08:30 New York candle, that of the American data. Its five-minute range, relative to the median:
| Day | Monday | Tuesday | Wednesday | Thursday | Friday |
|---|---|---|---|---|---|
| 08:30 New York candle | 1.5× | 2.0× | 2.3× | 2.5× | 3.1× |
| Median H1 range 08:00-09:00 New York | 22.0 $ | 22.9 $ | 26.9 $ | 23.7 $ | 28.3 $ |
| Median H1 range 15:00-16:00 New York | 14.3 $ | 16.0 $ | 14.4 $ | 15.2 $ | 12.7 $ |
The plausible explanation is the calendar: on Monday few important data come out, on Thursday the weekly jobless claims, on Friday (the first of the month) the employment report. We did not identify the single events, so this is a reading and not a measure. By contrast, Friday afternoon (15:00-16:00 in New York, 21:00-22:00 in Rome) is the calmest hour of the five days (12.7 $): the market heads into the weekend. The day does not make the movement: the data that come out that day do. We look at them closely in chapter 8.
Read the chart in words
Daily gold chart from 6 April to 3 May 2026 with, below, the bars of the 08:30 New York candle by weekday.
Wide days come in clusters
The thing that matters most in the data is another. Volatility clusters: the correlation between the range of a day and that of the next day is 0.62. In the widest 5% of days (11 days, all with a range of at least 251 $) the following day has a median range of 318 $, against 94 $ after all the other days. A huge day is not an isolated fact: another usually follows.
For someone trading this is concrete, easy information: yesterday's range is the best clue to today's range. If yesterday was big, stops must be wider and positions smaller. If yesterday was small, expect a normal day (the overall median is 98 $).
The alternation of days
Another measure, all about direction: after an up day, the next day is up in 43% of cases; after a down day, in 63%. The difference (20 points) has p = 0.004 in a permutation test: it is hard to think of it as chance. In this period, the direction of one day tends to be followed by the opposite one on the next day.
Read it with caution. It is a measure on 211 days and a single period, and in this chapter we made several comparisons (days, hours, sequences): with so many comparisons, some will come out low by chance. It is not a rule: it is a question to retest on a second period before using it. And "alternation" does not mean "sell after an up day": with a range of 100 $ a day, cost and stop weigh a lot.
Read the chart in words
Daily gold chart from 6 April to 3 May 2026 with four candles marked with the letters A, B, C and D.
| When | What | Price |
|---|---|---|
| 6 Apr 2026 | Point A | 4,703.87 |
| 7 Apr 2026 | Point B | 4,833.73 |
| 10 Apr 2026 | Point C | 4,794.24 |
| 12 Apr 2026 | Point D | 4,752.50 |
Times in UTC.
What to do with it
- Do not choose a "good" day. In the data range and direction do not depend on the day. If you backtest by day, remember that 42 days per group give very wide errors.
- Plan on the data calendar, not on the day. On Thursday and Friday the 08:30 candle is the widest (2.5× and 3.1×): that is where caution is needed. Check the calendar every week.
- Use yesterday's range. After a wide day (over 250 $ in our data) the next day is worth triple an average day. Widen stops or reduce size.
- Friday evening is calm. Friday afternoon (from 21:00 Rome time) is the quietest part of the week. If you hold a position over the weekend, consider that the spot market reopens on Sunday evening with possible jumps (chapter 7).
- Do not be swayed by a run of days. The alternation between up and down days in this period is strong, but it is one period.
Procedure in four steps
- Before saying a day is different, shuffle the labels and look at the p-value.
- Check the week's data calendar: that is where the days change.
- Look at yesterday's range to size stop and position.
- Retest every rule on a second period before trusting it.
Exercises
Chapter exercises
1. On the chart the letters A-D are on four daily candles of April 2026. Which letter is a Friday, the day with the widest 08:30 data candle (3.1× the median)?
Read the chart in words
| When | What | Price |
|---|---|---|
| 6 Apr 2026 | Point A | 4,703.87 |
| 7 Apr 2026 | Point B | 4,833.73 |
| 10 Apr 2026 | Point C | 4,794.24 |
| 12 Apr 2026 | Point D | 4,752.50 |
Times in UTC.
2. For the five days of the week the day's range is 94-111 $ as a median and the permutation test gives p = 0.80. What does it mean?
3. After one of the widest days (the largest 5%, at least 251 $) the next day has a median range of 318 $, against 94 $ after the other days. What does it mean for the stop?
4. After an up day the next day is up in 43% of cases, after a down one in 63% (p = 0.004). How do you treat 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.