Part 2: Regimes, forecasts and tails · Chapter 4 of 14 · 8 min

Volatility regimes

Dividing the days into three regimes according to the 10-day mean range (low up to $95, high above $112), today's regime is tomorrow's in 87% of cases (against 78% for shuffled ranges, 95th percentile 84%) and it never jumps two levels. High-regime periods last 11 days on median, low-regime ones 5; in the high regime the mean range is $174, in the low one $84, about double. The thresholds should be calculated on the past: here we fixed them on the whole sample.

What a regime is

In chapter 3 we saw that volatility has memory. A volatility regime is a name given to that fact: a period in which volatility sits in a band, low, medium or high, and stays there long enough to be taken into account.

You need a rule to classify. The simplest is this: take the mean range of the last 10 days and divide the history into three equal parts (terciles).

Regime10-day mean rangeDays in our data
Lowup to $95.468
Mediumfrom $95.4 to $112.067
Highabove $112.066

The three classes are there to give you a number to look at. With two or four classes the conclusions would be the same in substance.

Figure 1. 10-day mean range (pink line, in $) with the two tercile thresholds ($95.4 and $112.0) and the high-regime periods lasting at least eight days (grey bands): late January-February, March-mid April and June.Data: gold, XAUUSDT perpetual future (real candles) · daily (D1) · from 15 Dec 2025 to 5 Oct 2026.
Read the chart in words

Daily gold chart with, below, the 10-day mean range line, two horizontal threshold lines and grey bands in the high-regime periods.

WhenWhatPrice
28 Jan 2026 → 25 Feb 2026high regime
2 Mar 2026 → 16 Mar 2026
18 Mar 2026 → 15 Apr 2026
10 Jun 2026 → 24 Jun 2026

Times in UTC.

How much gold moves in each regime

RegimeMean daily rangeMedian
Low$84$79
Medium$104$104
High$174$138

The high regime is worth more than double the low one on average ($174 against $84) and much more in the tail: the mean sits $36 above the median because it contains the $400-700 days of late January. The difference is what you would expect from a classification made on the range itself; the useful information is that the number changes scale: a stop designed for $80 a day is very tight when the day moves $170.

How long they last and how they change

The probability that the next day is in the same regime:

FromStaysGoes up one levelGoes down one level
Low88%12%–
Medium81%7%12%
High92%–8%

In no case does the regime jump two levels in a day: you move by steps. The durations of the periods, in days:

RegimePeriodsMedian durationMaximum
Low9521
Medium13315
High51121

The high regime lasts longer because volatility, once it has exploded, fades slowly (chapter 3). The medium one is a transit zone: it lasts little. The long periods, eight days or more, are ten: low from 29 December to 26 January (21 days), high from 28 January to 25 February (21), high from 2 to 16 March (11), high from 18 March to 15 April (21), high from 10 to 24 June (11), medium from 25 June to 9 July (11), low from 15 July to 7 August (18), medium from 10 to 28 August (15), medium from 2 to 17 September (12), low from 18 September to 5 October (12).

How much is due to the method

There is an obvious objection. Today's 10-day mean range and tomorrow's share nine days out of ten: the regime would stay the same even if the data had no memory. To measure the part that comes from the data and not from the construction, we shuffle the daily ranges into random order (300 times), repeat the classification and count the share of days in which the regime stays the same.

Share of days in which the regime stays the same
Real data87%
Shuffled ranges, mean of 300 trials78%
Shuffled ranges, 95th percentile84%

The persistence of the real data sits above that of chance (87% against a plausible maximum of 84%), but the margin is modest: nine points. Most of the apparent stability comes from the 10-day window, and only a part from the market's real memory. With a shorter window the classification would be more nervous and with a longer one slower: the same lesson as chapter 2.

The limit of the thresholds

The two thresholds ($95.4 and $112.0) were calculated on the whole sample, so using also days that come after those being classified. In real use the thresholds should be calculated on the past only (for example on the last 250 days), and they change over time. With ten months we do not have enough history to test it honestly: that is why the numbers of the chapter are read as a description and not as a rule already proven.

There is also a bigger limit: almost all the high-regime days come from a single episode (from late January to mid April, with two short pauses), plus one period in June. The «median duration of 11 days» therefore rests on five periods in all.

What to do

  1. Name the regime with a single number: 10-day mean range, with thresholds calculated on the past (for example terciles of the last 250 days).
  2. Write the typical values of each regime for your instrument: in the high regime the mean range can be double.
  3. Adapt stop and size to the regime (chapters 8 and 9), because the same stop in dollars means something different.
  4. Expect the regime to stay for days, but not for ever: after 11 days of high regime, on median, it changes.
  5. Do not trust a median duration calculated on five periods.

Procedure in four steps

  1. Calculate the 10-day mean range at every day.
  2. Calculate the thresholds (terciles) on the previous days only.
  3. Assign the regime and count how many days it stays and how it changes.
  4. Compare the persistence with a shuffled series before believing it is real.

Exercises

Chapter exercises

1. In the high regime the mean daily range is $174, in the low regime $84. What does it imply for a fixed $80-a-day stop?

2. The regime stays the same on 87% of days in the real data and on 78% (95th percentile 84%) with shuffled ranges. What does it mean?

3. Why can the $95.4 and $112.0 thresholds not be used like this in real trading?

4. The median duration of the high regime is 11 days, over how many periods?

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.