Part 4: Compressions, spikes and the sheet · Chapter 14 of 14 · 17 min
A stop told twice, mistakes, reference, glossary and final test
The same stop told in two ways: «$20 below the entry, 1% risk, 3 R target: a 1 to 3 ratio» and «1.1 hourly ATR, touched by noise in 63% of cases within 12 hours and in 73% within 24, while a random entry hits the target one time in four». Then the most frequent mistakes in chapter order, the quick reference with rules and measurements, the glossary and a final test of 15 questions.
A stop told twice
A trader opens a position on gold with an hourly ATR of $18.6, the median of our data. They choose a $20 stop and a $60 target.
The story that reassures. «Stop $20 below the entry, risk 1% of the account, target $60: a reward-to-risk ratio of 3 to 1. The idea only needs to be right one time in four for me not to lose.» All true, and the plan looks solid.
The story that informs. The same decision, with the numbers of the book:
| Question | Answer (chapter) |
|---|---|
| How wide is the stop in ATR? | 20 ÷ 18.6 = about 1.1 ATR (1, 2) |
| How often does noise touch it? | With random entries, 1 ATR is touched in 63% of cases within 12 hours, in 73% within 24, in 82% within 48 (8) |
| What is the target worth? | $60 is about 3.2 ATR: with a stop at 1.1 ATR you need a movement of 3 units against 1 (8) |
| What base frequency does it have? | A random entry with a 1.5 ATR stop and a 3 R target hits the target about one time in four (23%) and the stop in 70% of cases (8) |
| And in a high regime? | With an ATR of $33 (90th percentile) the same stop is 0.6 ATR: touched in about eight cases out of ten within 12 hours (8) |
| And after a spike? | The hour after is worth double the normal ATR; a stop sized beforehand is touched by a normal candle for that phase (12) |
| And if the candle jumps the stop? | An 8 ATR candle cuts through the stop; the loss can be more than 1 R (6) |
| How much do you really risk? | With a fixed quantity, from 0.64% to 4.99% of the account on the example plan; with recalculated size, the share decided (9) |
| How far will the price move? | Over the next 24 hours, 5.1 ATR on median (about $94), from 3.1 to 10.1 ATR in 80% of cases (7) |
The difference between the two stories is not in the data: it is in the question. The first asks «how much do I risk and how much can I earn», the second «how often does noise stop me before the idea plays out». The 3 to 1 ratio counts if the stop survives the noise often enough; the sheet of chapter 13 exists to make you look at it every time.
The most common mistakes
| Mistake | Chapter |
|---|---|
| Calling a day «quiet» because it closed near the open; quoting the mean range as the normal value | 1 |
| Comparing two «14» ATRs without knowing whether they are simple or Wilder's; optimising the window until you find the one that «predicts best» | 2 |
| Deducing that «after an agitated day tomorrow will surely be agitated»; calculating persistence on a single episode | 3 |
| Fixing the regime thresholds on the whole history; reading the persistence without comparing it with a shuffled series | 4 |
| Taking the 14-day average as «tomorrow's range»; choosing among six methods the one with the lowest error as «the model that works» | 5 |
| Calculating a stop with the normal; talking of a «one-in-a-million event» when faced with a five-deviation candle | 6 |
| Multiplying by N instead of by the square root of N; sizing a target on the deviation instead of the range | 7 |
| Setting a «tight stop to risk little» at 0.5 ATR; reading a 41% target rate as proof of an edge without the 23% of chance | 8 |
| Always opening the same quantity believing you always risk the same; choosing fixed size because it earned more over 79 trades | 9 |
| Saying «the rule does better with high ATR» from a single cut; adding a filter «because it reduces the bad trades» | 10 |
| Saying «after compression the market explodes»; testing a breakout in compression on 42 trades | 11 |
| Entering after a spike with a stop sized on the previous ATR; reading a huge candle as «the market has chosen a direction» | 12 |
| Using the sheet as a signal; adding an unproven filter to the sheet | 13 |
Quick reference: the rules
| Concept | Rule of the book | Ch. |
|---|---|---|
| Measure | Range and true range are enough; declare mean or median, dollars or percent | 1 |
| ATR | Mean of the 14 hourly true ranges; write whether simple or Wilder's; beware of the step | 2 |
| Memory | Volatility clusters, direction does not; check by period | 3 |
| Regimes | One number (10-day mean range, terciles of the past); persistence compared with chance | 4 |
| Forecast | 14-day average, band from 0.5 to 1.5 times (83% of days) | 5 |
| Tails | High kurtosis: extremes are seven times more frequent beyond 3 deviations; a stop is not a ceiling | 6 |
| Scale | Square root of time; range in ATR for 4, 12, 24, 48 candles | 7 |
| Stop | In ATR; touch frequency from the holding time; comparison with the random base | 8 |
| Size | Risk in dollars ÷ stop in dollars, recalculated at every entry | 9 |
| Filters | No volatility filter without a proof written beforehand, comparison with chance and out of sample | 10 |
| Compression | Position of the ATR among the last 500 candles; scale, not direction | 11 |
| Spikes | True range above 3 ATR: a tail of about three hours; direction without signal | 12 |
| Sheet | Eight items at each session; thresholds rewritten every month | 13 |
Quick reference: the measurements
| Ch. | Result |
|---|---|
| 1 | Median hourly true range $17 (mean 22.1, maximum 391.5); median daily range $95.4 (mean 117.7; from 11 to 717); correlation with realised volatility: range 0.905, close-open difference 0.547 |
| 2 | Hourly ATR(14): median $18.6 (0.41%), 10th-90th percentile $13.2-33.0; simple against Wilder: correlation 0.98; correlation with the next 24 hours at 14 candles 0.78 (typical error 18%) |
| 3 | Autocorrelation of hourly true ranges 0.59 (1 hour), 0.38 (24 hours), 0.12 (5 days); of the return 0.00; day above median after a day above 62%; daily range 0.63 at 1 day, 0.05 from April |
| 4 | Regimes by terciles of the 10-day mean range: 84 / 104 / 174 $; same regime the next day 87% (78% shuffled); high regime 11 days on median |
| 5 | 14-day average: typical error 27%, real between ½ and 1½ of the estimate on 83% of days; median of the past 28% and 81% |
| 6 | Kurtosis 21.7; beyond 3 deviations 2.0% (normal 0.27%), beyond 4 0.97% (0.006%); true range above 3 ATR in 2.0% of hours, on 73 days out of 211 |
| 7 | Ratio with the square root of time 0.98-1.07 from 1 to 48 hours; 12/24/48-hour range 3.4/5.1/7.4 ATR; day 4.87 times the hourly ATR |
| 8 | Stop touched within 12 hours: 0.5 ATR 80%, 1 ATR 63%, 1.5 ATR 49%, 3 ATR 21%; random entries with 1.5 ATR and 3 R: target 23%, stop 70%; plan 41% and 53% |
| 9 | Risk per unit $18.2-141.6 (median $28.4); fixed risk 176% (worst −1.02%), fixed size 204% (worst −3.57%, risk 0.64-4.99%) |
| 10 | Plan by ATR: +0.29/+0.90/+0.80 R, by position +0.20/+0.55/+1.32; 480 rules with high ATR only −0.024 (random series −0.019) |
| 11 | After compression range 6.7 ATR (5.0 normal, 4.5 expansion); ATR after 24 candles 1.22 (shuffled 1.31); breakout in compression +0.13 ± 0.55 R (42 trades) |
| 12 | 68 spikes: hour after 2.0 ATR (shuffled 1.23); +0.09 ± 0.31 ATR at one hour in the direction; day after beyond ±2%: $152 against $108 |
| 13 | Last candle: ATR $13.49, 16th percentile, low regime, expected range $41-124, stop $20.2, 4.94 ounces; high-ATR filter: 3 trades in the second half |
All the measurements come from ten months of a single instrument (December 2025 - October 2026, gold perpetual future on Binance, prices only, 4,865 hourly candles with the market open and 211 days), with an episode of very high volatility between January and March. The example plan (79 trades) has an unproven edge (book 12). The random series are built by shuffling the same candles.
Glossary
The terms of the book are in the site glossary, where each one has an example with numbers.
| Term | Definition |
|---|---|
| True range | The largest of three distances of a candle: high minus low, absolute value of high minus previous close, absolute value of low minus previous close. |
| Daily range | The difference between the high and the low of a market day. |
| Realised volatility | Volatility measured after the fact with high-frequency returns: the square root of the sum of squared changes (for example at 5 minutes) over a period. |
| Wilder's ATR | The version of the ATR in which the new value is obtained from the old one: new ATR = (old ATR × (N − 1) + new true range) ÷ N. |
| ATR step effect | The sudden drop of the simple-average ATR when a very large candle leaves the window, without the market having done anything at that moment. |
| Autocorrelation | How much the value of a series resembles the one k steps before: it is 0 if there is no link, 1 if the two always move together. |
| Volatility regime | A volatility band (low, medium, high), defined with a single number such as the 10-day mean range divided into terciles, in which the market stays for a few days. |
| Fat tails | The feature of a distribution in which extreme values are far more frequent than the normal curve predicts, measured by kurtosis (3 for the normal). |
| Square-root-of-time rule | The rule that the deviation of movements over N periods equals that of one period multiplied by the square root of N, if successive movements are not tied to each other. |
| Stop in ATR | A stop whose distance from the price is measured in multiples of the ATR (for example 1.5 ATR) instead of in fixed dollars. |
| Size by volatility | The size derived from the dollar risk divided by the dollar stop (k × ATR), recalculated at every entry, so as to always have the same risk per trade. |
| Volatility compression | A period in which the ATR sits below the 20th percentile of the last 500 candles; expansion is the period above the 80th. |
| Volatility spike | A candle whose true range exceeds three times the ATR of the previous candle. |
| Volatility sheet | A page to fill in at every session with eight items: ATR, ATR position, regime, expected range with band, stop, size, spike alert, day beyond ±2%. |
What to do now
- Fill in the volatility sheet at every session (chapter 13) and rewrite the thresholds every month.
- Choose the stop in ATR from the holding time and the noise you accept, and compare it with the random base.
- Calculate the size from risk and stop at every entry: do not always open the same quantity.
- After a spike or a day beyond ±2%, recalculate ATR, band and size before entering.
- Do not add volatility filters without a proof written beforehand, compared with chance and verified out of sample.
- For the foundations: the «Market structure» book, the «Order blocks from A to Z» book, the «FVG and liquidity» book, the «SMC: the whole system» book, the «Indicators from A to Z» book, the «Elliott and Fibonacci» book, the «Chart patterns» book, the «Japanese candlesticks» book, the «Hours, sessions and news» book, the «Support, resistance and levels» book, the «Risk and position sizing» book, the «Backtesting and overfitting» book and the «Psychology and the trading journal» book. For risk and discipline: the risk and psychology course.
Final test
Fifteen questions on the book. You pass the test with at least 12 right answers.
Final test of the book
1. In the chart the four marked candles are A (17 March 2026, 09:00 UTC), B (15:00), C (22:00) and D (18 March, 04:00). Which has the largest true range?
Read the chart in words
| When | What | Price |
|---|---|---|
| 17 Mar 2026 09:00 UTC | Point A | 5,025.04 |
| 17 Mar 2026 15:00 UTC | Point B | 5,017.99 |
| 17 Mar 2026 22:00 UTC | Point C | 5,007.34 |
| 18 Mar 2026 04:00 UTC | Point D | 4,994.17 |
Times in UTC.
2. In the chart the lines are entry ($4,634.20), stop ($4,661.79) and target ($4,551.44) of a sell on 15 May 2026 (UTC time); the line on the left is the 80-candle low ($4,642.18). Which letter is the candle in which the target is reached?
Read the chart in words
| When | What | Price |
|---|---|---|
| 15 May 2026 00:00 UTC | entry | 4,634.20 |
| 15 May 2026 00:00 UTC | stop: −1 R (27.6 $) | 4,661.79 |
| 15 May 2026 00:00 UTC | target: +3 R | 4,551.44 |
| 14 May 2026 16:00 UTC | 80-candle low | 4,642.18 |
| 14 May 2026 20:00 UTC | Point A | 4,661.01 |
| 15 May 2026 00:00 UTC | Point B | 4,661.13 |
| 15 May 2026 08:00 UTC | Point C | 4,576.40 |
| 15 May 2026 09:00 UTC | Point D | 4,568.00 |
Times in UTC.
3. A candle has high 52, low 47 and the previous one closed at 56. What is the true range?
4. The true ranges of the last three candles are 6, 9, 9 and the new candle is 15. What is the 3-candle ATR with the simple average?
5. The autocorrelation of the hourly return is 0.00 and that of the true range 0.59. What does it mean?
6. The volatility regime stays the same on 87% of days in the real data and 78% (95th percentile 84%) with shuffled ranges. What do you conclude?
7. The mean of the previous 14 days gives $100. Between which values does the day's range sit in 83% of cases?
8. Returns beyond 3 deviations are 2.01% in the data and 0.27% in a normal. What does it imply for a stop?
9. The hourly ATR is $24. What is the expected range of the day (median)?
10. With random entries, a 0.5 ATR stop is touched within 12 candles in 80% of cases. What does it mean?
11. Account $20,000, 1% risk and a stop of $40 per unit. How many units do you open?
12. Across 480 rules, keeping only entries with high ATR changes the mean R by −0.024 and on random series by −0.019. What do you conclude?
13. After a compression the ATR rises to 1.22 times, in a shuffled series without memory to 1.31. What does it mean?
14. What does the volatility sheet NOT contain?
15. The standard deviation of one-hour returns is 0.5%. What deviation do you expect over 9 hours with the square-root-of-time rule?
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.