Rule Collider

Backtests · GOLD · h1_trend_ema20_vs_ema50 · long

Does h1_trend_ema20_vs_ema50 above 0 work on GOLD? 2023 to 2026 backtest

GOLD long when H1 EMA20 above EMA50 is above the threshold, tested at five thresholds taken from the feature’s own distribution on GOLD over this window: its 10th, 25th, 50th, 75th and 90th percentiles. Window 2023-09-30 to 2026-09-30, the last 30 percent held out of sample; stop 2.5 ATR, target 3 R, held overnight, max spread 30 pips, one-minute confirmation 2 pips, no hour filter. Fills and spread as measured on a demo fleet (how a result is judged). The direction and the exit are the ones the research catalogue carried for this pair; the thresholds are not.

Five thresholds, one rule shape

PercentileConditionIn sample R / tradesOut of sample R / tradestTrades to confirmLast quarter R / tradesVerdict
p10, p25h1_trend_ema20_vs_ema50 > 0-75.4 R / 2210-73.2 R / 739-1.71,082-40.1 R / 230No edge shown

R is the risk on one trade. The out-of-sample split carries the verdict: below 100 trades it is “no edge shown” whatever the t; a positive mean with t below 2 is “net positive, not yet distinguishable from chance”; the last column is how many trades that mean would need to reach t = 2.

Each threshold in full

h1_trend_ema20_vs_ema50 > 0 (p10, p25)

Honesty panelNo edge shown

No edge shown out of sample: -73.2 R over 739 trades is -0.099 R per trade, within what chance produces (t = -1.7); about 1,082 trades would be needed to confirm this mean at t = -2, shown: 739.

In sample
2023-09-30 to 2025-11-05
-75.4 R
2210 trades · 33% wins
-0.034 R per trade · t = -0.9, within chance
about 9,810 trades to confirm this mean (-0.034 R per trade) at t = -2 · shown: 2210
spread 0.102 R per trade
Out of sample
2025-11-05 to 2026-09-30
-73.2 R
739 trades · 31% wins
-0.099 R per trade · t = -1.7, within chance
about 1,082 trades to confirm this mean (-0.099 R per trade) at t = -2 · shown: 739
spread 0.060 R per trade
Last quarter
last 91 days of the window
-40.1 R
230 trades · 29% wins
-0.174 R per trade · t = -1.7, within chance
about 337 trades to confirm this mean (-0.174 R per trade) at t = -2 · shown: 230
spread 0.081 R per trade
Costs. Spread 0.091 R per trade at 1.25 times the bar’s minimum (more than the trades made before spread); 483 fills in measured hot hours (01:00 at 3.9 times). Fill offset 0.08 stop-distances on every entry, measured on 732 demo fills. Note: the symbol's quoting precision changed on 2025-08-04; the spread field is converted with the point of each era. At 1.5 times the bar’s minimum the out-of-sample figure would read -82.1 R (-0.111 R per trade over 739 trades).

Entry: stop order 2 pips past the signal close, cancelled after 4 bars; 3,110 signals placed, 2,949 filled, 161 expired; trigger state, an entry on every bar the conditions hold while flat.

Exit: stop 2.5 ATR (mean 84.6 pips, 2.50 ATR, 18 raised to the 10-pip floor); target 3 R; exits 65% at the stop, 22% at the target, 13% at the day end.

Filters: signals on every weekday; no entries 22:00 to 01:00 broker time (6,534 bars); max spread 30 pips (0 bars over it); M1 confirmation 2 pips (568 signals rejected).

Monthly R, h1_trend_ema20_vs_ema50 > 0 on GOLDOne bar per month: the R earned that month; the held-out months in the full colour.2023-10: +29.0 R over 100 trades2023-102023-11: +10.8 R over 77 trades2023-12: -11.9 R over 90 trades2024-01: -23.6 R over 78 trades2024-02: -19.1 R over 76 trades2024-03: +12.0 R over 77 trades2024-04: -20.3 R over 116 trades2024-042024-05: -30.5 R over 95 trades2024-06: -34.2 R over 79 trades2024-07: +14.5 R over 96 trades2024-08: -32.0 R over 110 trades2024-09: +11.6 R over 99 trades2024-10: -2.5 R over 112 trades2024-102024-11: +10.7 R over 45 trades2024-12: -21.5 R over 64 trades2025-01: +3.5 R over 110 trades2025-02: +7.2 R over 111 trades2025-03: +11.1 R over 81 trades2025-04: +16.4 R over 72 trades2025-042025-05: -8.8 R over 66 trades2025-06: -2.3 R over 65 trades2025-07: -12.1 R over 74 trades2025-08: -29.8 R over 98 trades2025-09: +28.9 R over 123 trades2025-10: +19.7 R over 94 trades2025-102025-11: +9.6 R over 79 trades2025-12: +2.8 R over 107 trades2026-01: +7.7 R over 96 trades2026-02: -2.9 R over 62 trades2026-03: -16.3 R over 35 trades2026-04: -18.3 R over 72 trades2026-042026-05: -14.1 R over 41 trades2026-06: -4.1 R over 19 trades2026-07: -37.9 R over 86 trades2026-08: +12.0 R over 107 trades2026-09: -14.2 R over 37 trades+29 R-38 R

What h1_trend_ema20_vs_ema50 measures

H1 EMA20 above EMA50 (flag), on H1 bars. 1 when the H1 20-bar exponential average is above the 50-bar one on the last completed H1 bar.

Test your own variant

Rule Collider runs one rule on one symbol over years of broker M5 data with the fills and the spread a demo fleet measured, and shows the out-of-sample figure first. Early access goes to the waitlist in signup order.

Join the waitlist

Historical simulation on broker M5 data with a measured fill offset, a measured spread factor, swap and a one-minute confirmation model. Past results do not predict future results. This is a research tool, not investment advice. Hours are broker time, the chart time in MT5. Built 2026-11-04 from runs ending 2026-09-30; refreshed monthly.