False positives from pattern recognition on short timeframes

[Amanhasnoname]

Human brains find patterns in random data. On one-minute charts, where price movement is heavily noise, this tendency leads to many false signals.

Double tops, head and shoulders and flags appear all the time in random walks. Measuring how often a given pattern leads to a particular outcome is the only way to know if it has predictive value.

Most patterns, tested properly on short timeframes, show weak or no edge after costs. The ones that survive tend to be tied to specific conditions, like a particular time of day or volatility level.

A way to check is to define the pattern precisely, in code or in strict rules, then count occurrences and results over a large sample. Visual identification invites bias, because ambiguous cases get interpreted in favour of the expected outcome.

The exercise is humbling. A favourite pattern may turn out to perform no better than random entries.

Which patterns have you tested objectively, and what did you find?