What we tested
A mass scanner: every combination of 11 windows across the day, 18 signal types (breakout and rejection of the 15-, 60- and 240-minute high, deviation from session VWAP, a big move over 5, 15 and 30 minutes, break and rejection of the Asian and London range), 5 filters (high and low volatility, days of the week), two variants (with the move and against it) and 6 exits (time exits at 15 and 60 minutes, ATR targets and stops). 10,680 rules in total. Entry on the next minute, at the price a buyer really pays (ask) or a seller really gets (bid).
Result
In 2012–2019 no rule passed the correction for multiple testing (with so many tries a few "good" results always appear by chance). We repeated the whole scanner on 2021–2024, when gold was in a completely different regime, with today’s broker costs. Same result: zero. Importantly, the median result before costs was about zero — the problem is not "a good signal eaten by commission", but no direction in the price conditions themselves.
What it means for a trader
- If a simple price rule looks great on a few weeks of chart, it is most likely chance. Among thousands of possible rules, one always "works" in the past.
- Scalping gold needs information beyond the price chart — or holding the move longer, so the profit per trade is clearly larger than the cost.
Limits
The scanner covers price rules on M1–M5 candles. It does not cover exchange data (order flow, options) — those are separate lab studies.