Part 4: Strategies, system and reference · Chapter 11 of 14 · 13 min

Time-based strategies

We tested three clock-based rules (breakout of the Asian range, breakout of the New York opening, going against the first candle of the 08:30 data), with rules written beforehand, one trade a day, costs and a comparison with random direction. None beats chance: Asia gives +0.05 ± 0.10 R over 169 trades (random direction −0.01), the New York opening +0.03 ± 0.12 over 208. The "best hour" is also a selection illusion: the best t among 23 hours (2.08) is what chance produces (median 2.16).

What you can ask of the clock

In the previous chapters we measured size: at what time, on what day, around which events gold moves the most. Those measures hold, because volatility follows people's schedules. A strategy asks for something more: direction. "Buy if it breaks the Asian range" is a direction rule. For that the clock says almost nothing, and it is worth understanding why before wasting money.

Testing a time-based rule properly takes eight things, and we applied them to every test in the chapter:

  1. The rule written beforehand. Entry, stop, target, time exit, all decided before looking at the results.
  2. At most one trade a day. That way trades are independent.
  3. Costs included. 0.3 $ per trade, as in the previous books.
  4. The result in R, that is in multiples of the risk (1 R = the distance between entry and stop).
  5. A comparison with random direction: the same entries, the same risk, but a random direction. It is the true starting point: with symmetric stop and target the expected result is just below zero (the cost).
  6. The two halves of the sample kept apart.
  7. Random series: the same candles shuffled in blocks of six hours, to see what the rule would do without time structure.
  8. Varied parameters: a rule that works only with one exact value does not work.

The three rules

RuleEntryStop and targetExit
AsiaThe Asian range is the high and low of the H1 candles between 00:00 and 07:00 UTC. Enter at the close of the first H1 candle that closes beyond an extreme, between 07:00 and 15:59 UTCStop at the opposite extreme, target at 1 R21:00 UTC
New York openingThe range is the first 30 minutes from 09:30 New York time. Enter at the close of the first five-minute candle beyond an extreme, until 15:00Stop at the opposite extreme, target at 1 R16:55 New York
Against the 08:30 dataOn days when the 08:30 candle is at least 3 times the reference (07:00-08:30) and has a body of at least 3 $, enter at 08:35 against its directionStop and target at half a body09:25 New York

The UTC times correspond, in Rome, to 01:00-08:00 in winter and 02:00-09:00 in summer.

What happened

RuleTradesMean R ± errorWinnersRandom direction (same entries)
Asia (range 00:00-07:00)169+0.05 ± 0.1051%−0.01
New York opening (30 minutes)208+0.03 ± 0.1249%−0.01
Against the data, ratio ≥ 3×31−0.17 ± 0.3448%−0.18
Against the data, ratio ≥ 5×14+0.21 ± 0.4771%−0.18

One reading: no rule has a result distinguishable from zero. For the first two, the error interval (± 0.10, ± 0.12) contains zero and also contains the random-direction result. For the third the sample is too small to say anything (± 0.34 and ± 0.47).

Asia. Over the 169 days, 33 trades reach the target, 21 the stop and 115 (68%) exit on time: most days decide nothing. The cost is worth 0.005 R (0.3 $ on a median risk of 62 $). In the two halves of the sample: +0.05 ± 0.15 and +0.06 ± 0.13: no difference. Comparison with random series: over 300 series shuffled in blocks of six hours the same rule gives on average −0.01 R (between −0.10 and +0.09 for 90% of the series); our +0.05 is exceeded by 12% of random series: it cannot be told apart.

New York opening. 208 trades, 77 at the target, 62 at the stop, 69 on time; median risk 26 $. In the halves: +0.05 ± 0.16 and +0.01 ± 0.17.

Against the data. With the ratio ≥ 3× (31 days) mean R is −0.17; with ≥ 5× (14 days) it is +0.21. Changing the threshold from 3 to 5 flips the sign, but the sample falls from 31 to 14 days and the error rises from ± 0.34 to ± 0.47: both estimates contain zero. On the 14 days, going with the first candle instead of against gives −0.57 ± 0.38: consistent with the partial return seen in chapter 8, but it is a hypothesis, not proof.

Figure 1. An Asia trade that hit its target (22 January 2026, UTC hours): the Asian range is 56.3 $, the 07:00 close above the high is the entry, the stop is the low of the range and the target is 1 R.Data: gold, XAUUSDT perpetual future (real candles) · 1 hour (H1) · from 22 Jan 2026 00:00 UTC to 23 Jan 2026 05:00 UTC.
Read the chart in words

H1 gold chart of 22 January 2026 with the Asian range, the long entry, the stop and the target.

WhenWhatPrice
22 Jan 2026 00:00 UTC → 22 Jan 2026 06:00 UTCAsian range 00:00-07:00 UTC
22 Jan 2026 07:00 UTClong entry4,832.23
22 Jan 2026 07:00 UTCstop4,771.55
22 Jan 2026 07:00 UTCtarget 1R4,892.91

Times in UTC.

Figure 2. An Asia trade stopped out (2 January 2026, UTC hours): long entry at 08:00, price returns below the high and touches the stop.Data: gold, XAUUSDT perpetual future (real candles) · 1 hour (H1) · from 2 Jan 2026 00:00 UTC to 3 Jan 2026 05:00 UTC.
Read the chart in words

H1 gold chart of 2 January 2026 with the Asian range, the long entry, the stop and the target.

WhenWhatPrice
2 Jan 2026 00:00 UTC → 2 Jan 2026 06:00 UTCAsian range 00:00-07:00 UTC
2 Jan 2026 08:00 UTClong entry4,383.57
2 Jan 2026 08:00 UTCstop4,344.74
2 Jan 2026 08:00 UTCtarget 1R4,422.40

Times in UTC.

Varying the parameters

Changing the parameters moves the result, and not in the same direction:

Rule and parameterTradesMean R ± error
Asia, range until 06:00178+0.10 ± 0.10
Asia, range until 07:00169+0.05 ± 0.10
Asia, range until 08:00161+0.01 ± 0.10
Asia, target at 2 R169+0.07 ± 0.11
New York opening, 15 minutes211+0.01 ± 0.12
New York opening, 30 minutes208+0.03 ± 0.12
New York opening, 60 minutes179+0.05 ± 0.10

All the rows are inside the error, and every variant could be "the best": depending on the parameter you can pick +0.10 or +0.01 with the same rule. It is the most common way of fooling yourself: trying ten variants and presenting the one that does best.

Another distinction: Asia, split by direction, gives −0.02 over the 89 long trades and +0.14 over the 80 short ones. It is not an advantage of shorts: in our period the price fell (from about 4,300 $ to about 4,138 $, with peaks) and a short-only rule benefits. Random direction (the average of the two sides) exists precisely to remove this effect.

Figure 3. The cumulative R of Asia over the ten months: the dashed line is random direction on the same entries (almost flat, −0.9 R), the bars are the strategy (+9.3 R over 169 trades, with an irregular path).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 from December 2025 to October 2026 with, below, the cumulative R of the Asian strategy and of random direction.

The illusion of the best hour

The most popular idea among beginners is: "find the hour when gold rises the most and always buy it". We did it with the right rules. For each of the 23 New York hours we computed the mean return of the H1 candles (close minus open) and its t, the ratio between the mean and its error.

The best is the 02:00 hour in New York: +2.13 $ on average, t = 2.08, over 212 candles, the only one with |t| above 2. It looks like a result. But we built 300 random series (the same candles shuffled in blocks of six hours) and for each we looked for the best hour: the maximum |t| among the 23 hours has a median value of 2.16 and a 95th percentile of 3.11. Our 2.08 is exactly what chance produces: 57% of random series have an equal or higher maximum. If you run 23 tests, at least one comes out near "2 sigma" almost every time.

The out-of-sample check says the same. Choosing the best hour in the first half of the sample (until 11 May) gives the 07:00 hour (mean +2.59 $, t = 1.31). In the second half, the same hour, in the chosen direction, is worth +1.75 ± 2.13 $ over 105 candles: positive, but inside the error. On random series the same procedure gives a positive mean in the second half in 54% of cases: a coin.

How many trades are needed

The standard deviation of the Asian strategy's R is about 0.65. With it, to measure a mean R with an uncertainty of ± 0.10 about 162 trades are needed; for ± 0.05, about 645. With one trade a day that is about ten months (our whole sample) in the first case and about three years in the second. The "against the data" rule cannot be judged with 14 days: the error (± 0.47) is larger than any plausible result.

The practical rule is this: before trusting a result, compute how wide its error is. A +0.2 R over 14 trades is worth nothing.

Figure 4. Exercise: one day (6 January 2026, UTC hours) with the Asian range in grey and its two lines; four candles marked with the letters A-D.Data: gold, XAUUSDT perpetual future (real candles) · 1 hour (H1) · from 6 Jan 2026 00:00 UTC to 7 Jan 2026 05:00 UTC.
Read the chart in words

H1 gold chart of 6 January 2026 with the Asian range, the high and low lines and four candles marked with the letters A, B, C and D.

WhenWhatPrice
6 Jan 2026 00:00 UTC → 6 Jan 2026 06:00 UTCAsian range
6 Jan 2026 03:00 UTCPoint A4,466.09
6 Jan 2026 13:00 UTCPoint B4,465.47
6 Jan 2026 14:00 UTCPoint C4,487.99
6 Jan 2026 20:00 UTCPoint D4,492.45

Times in UTC.

What to do with it

  1. Use the clock for size, not for direction. Chapters 2-10 show times that repeat (the American morning concentrates the day's extremes, the 08:30 data and the Fed move a lot). They give no direction.
  2. If you want to test a time-based rule, apply the eight checks. Rule written beforehand, one trade a day, costs, R, random direction, two halves, random series, varied parameters.
  3. Count the trades before judging. With 14 or 31 nothing can be said. You need hundreds.
  4. Distrust "the best". The best of 23 hours, 10 variants or 5 instruments is almost always the effect of selection. Always ask: how many things did I try to find this one?
  5. Use hours for risk management. The things that hold are practical: avoiding entries on data and the Fed, sizing the stop on the hour (chapter 9), keeping positions small over the weekend (chapter 7).

Procedure in five steps

  1. Write the rule (entry, stop, target, exit) before looking at the data.
  2. Run one trade a day, with the cost, and measure the result in R.
  3. Compare with random direction on the same entries, with the two halves of the sample and with random series.
  4. Vary the parameters and see whether the sign holds.
  5. Compute the error (± 1.96 times the standard deviation divided by the square root of n) and compare it with the result.

Exercises

Chapter exercises

1. On the chart the Asian range (00:00-07:00 UTC) is in grey with the two lines of the high and the low. Which letter is the candle whose close triggers the Asia rule's entry, that is the first H1 candle closing beyond an extreme after 07:00?

Chart exercise. Data: gold, XAUUSDT perpetual future (real candles) · 1 hour (H1) · from 6 Jan 2026 00:00 UTC to 7 Jan 2026 05:00 UTC.
Read the chart in words
WhenWhatPrice
6 Jan 2026 00:00 UTC → 6 Jan 2026 06:00 UTCAsian range
6 Jan 2026 03:00 UTCPoint A4,466.09
6 Jan 2026 13:00 UTCPoint B4,465.47
6 Jan 2026 14:00 UTCPoint C4,487.99
6 Jan 2026 20:00 UTCPoint D4,492.45

Times in UTC.

2. On the same 169 trades the Asia rule gives +0.05 ± 0.10 R and random direction −0.01 R. What does it mean?

3. Among the 23 New York hours the best has t = 2.08 and looks like proof of an advantage. On 300 random series the maximum |t| has a median of 2.16. How do you read it?

4. The "against the data" rule gives +0.21 R with ± 0.47 over 14 trades (ratio ≥ 5×) and −0.17 with ± 0.34 over 31 (≥ 3×). What do you do?

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.