Do Traditional Technical Patterns Survive Out of Sample? Evidence from 39 Mechanically Defined Patterns Across 16 Foreign-Exchange Markets

This working paper evaluates whether traditional technical chart, candlestick and indicator patterns provide stable and economically meaningful directional information in foreign-exchange markets. Thirty-nine mechanically defined patterns were tested across 16 currency pairs using four-hour data from 2015 to 2026.

A chronological development and holdout design was used, incorporating causal pattern confirmation, a 20-bar embargo, matched unconditional benchmarks and estimated transaction costs. The 598 combinations selected exclusively from training data generated 101,199 signal–horizon observations in the holdout period.

The selected patterns achieved a gross win rate of 52.03%, but the net win rate fell to 50.27% after estimated costs. Average gross return was +1.41 basis points per observation, while average net return was −0.72 basis points. Training performance had almost no relationship with subsequent performance, and monthly-block uncertainty intervals included zero at every tested horizon.

The results indicate limited gross predictability without reliable average economic exploitability and are consistent with weak-form market efficiency in a practical sense.

This is a working paper and has not undergone formal peer review. Generative AI assistance used during code auditing, statistical checking and manuscript preparation is disclosed within the paper.