Part 3: Time, stop and size · Chapter 7 of 14 · 7 min
The time scale
The size of movements grows with the square root of time: on gold the 4-hour deviation is 0.725% against 0.737% expected from the rule (ratio 0.98), the 24-hour one 1.78% against 1.81%, and the ratio stays between 0.98 and 1.07 from 1 to 48 hours. The range (high minus low) over 12 hours is 3.4 hourly ATR on median, over 24 hours 5.1 and over 48 hours 7.4; a day's range is 4.9 times the previous day's hourly ATR on median (from 3.7 to 6.7 in the central half of days).
The square-root rule
If the movements of one hour are not tied to those of the hour before (as we saw for direction, chapter 3), the changes add up like random steps. For random steps the typical size after N steps equals that of one step multiplied by the square root of N. It is the square-root-of-time rule:
> deviation over N hours ≈ deviation over 1 hour × √N
Four hours are worth double one hour (√4 = 2), not four times. You need sixteen hours to quadruple it. The rule holds if successive movements have no recurring direction; if a market continues (trend) or bounces back (mean reversion), the ratio departs from 1.
Does it work on gold?
On hourly returns, with non-overlapping windows, the real deviation compared with that of the rule:
| Horizon | Real deviation | Rule (0.369% × √N) | Ratio | Samples |
|---|---|---|---|---|
| 1 hour | 0.369% | 0.369% | 1.000 | 4,864 |
| 2 hours | 0.523% | 0.521% | 1.003 | 2,432 |
| 4 hours | 0.725% | 0.737% | 0.984 | 1,216 |
| 8 hours | 1.041% | 1.043% | 0.998 | 608 |
| 12 hours | 1.259% | 1.277% | 0.986 | 405 |
| 24 hours | 1.782% | 1.806% | 0.987 | 202 |
| 48 hours | 2.724% | 2.554% | 1.066 | 101 |
From 1 to 24 hours the ratio stays within 2% of 1; at 48 hours it rises to 1.07, with only 101 samples (the statistical error at this length is about 7%). The rule works: on gold, between one hour and two days, there is no trend or mean reversion that moves it visibly. It matches the result of chapter 3: direction has no memory.
From the hourly ATR to the expected movement
The deviation describes a typical movement; for stops and targets what is usually needed is the range (high minus low) the price covers in a certain number of candles. Starting from every candle, we measured the range of the next N candles divided by the ATR of that moment:
| Next candles | Range in ATR: median | 10th percentile | 90th percentile |
|---|---|---|---|
| 1 | 0.87 | 0.45 | 1.78 |
| 4 | 1.86 | 1.04 | 3.72 |
| 12 | 3.43 | 2.01 | 6.85 |
| 24 | 5.07 | 3.09 | 10.14 |
| 48 | 7.41 | 4.41 | 14.31 |
Two readings.
- The range grows roughly like the square root of time. Between 4 and 48 candles time is multiplied by 12 and the range by 4: an exponent of about 0.55 instead of 0.5, slightly above the rule.
- The band is wide: in 12 hours the median is 3.4 ATR, but the 10th percentile is 2.0 and the 90th 6.9. The 24-hour range, 5.1 ATR on median, can be worth from 3.1 to 10.1.
From the hour to the day
A market day's range (about 23 candles) divided by the hourly ATR of the last candle of the day before:
| Measure | Value |
|---|---|
| Median | 4.87 |
| 25th percentile | 3.68 |
| 75th percentile | 6.69 |
The square root of 23 is 4.80: the daily range is 4.9 times the hourly ATR on median, almost what the rule says. With an hourly ATR of $20 the expected range of the day is about $97 (with the band from $74 to $134 in the central half of days).
Read the chart in words
Daily gold chart with, below, the daily range bars and a pink line that follows them.
The direct comparison with the ATR calculated on days (the mean of the previous 14 ranges) gives a median ratio of 0.90, in line with chapter 5.
What it is for
- For a realistic target: if you hold a position for 12 hours, the price covers 3.4 ATR on median, so a target 5 ATR away in a single direction is rarely reached.
- For changing timeframe: the 4-hour deviation is about double the one-hour one (0.725% against 0.369%).
- For going from one unit to another without recalculating: hourly ATR × 4.9 gives the daily range; × 7 gives the two-day one.
- For checking the data: a 4-hour deviation that is not about double the one-hour one is a sign of data or a calculation to review.
What to do
- Use the square root of time to convert: from one hour to N hours multiply by √N, from a day to N days the same.
- Use the range of the next N candles, not just the deviation, to choose stops and targets: the range is the number the price really touches.
- Write the band, not just the median: the 10th and 90th percentile change the plan.
- Check the ratio on your instrument: the rule holds if there is no trend or mean reversion at that scale.
- Do not extend it beyond two days without testing it: at 48 hours the ratio shifts and the sample is small.
Procedure in four steps
- Calculate the deviation (or the ATR) over one hour.
- Multiply by the square root of the number of hours in your horizon.
- Compare with the range measured for the same horizon (median, 10th and 90th percentile).
- Use the band in the plan.
Exercises
Chapter exercises
1. The standard deviation of one-hour returns is 0.369%. What deviation do you expect over 4 hours with the square-root-of-time rule?
2. The hourly ATR is $20. What is the expected range of the day (median)?
3. On median, how many hourly ATR does the price cover (high minus low) in 12 hours?
4. Why does the square-root-of-time rule work on gold between 1 and 24 hours?
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.