Multiple testing
Also written: data mining, overfitting
Testing many possibilities and reporting only the ones that worked, which manufactures apparent winners out of pure chance.
If you flip a coin ten times, getting ten heads is very unlikely. If a thousand people each flip a coin ten times, someone almost certainly will. Showing that person alone makes them look gifted.
The same happens with market patterns. A tool that sweeps thousands of combinations of dates will always surface a handful with perfect track records, whether or not any real pattern exists. The results are not fraudulent; they are just what searching hard enough guarantees.
The defence is to ask how many combinations were tested before the winner was shown, and to be suspicious when the same asset appears twice with different date windows.
The EPTA5 seasonality engine sweeps 2,388 calendar patterns and displays the best ranked. Several show a 100 percent success rate. That is the expected outcome of testing 2,388 times, not evidence of a method, and the article says so.
See also
This definition is part of the EPTA5 lexicon, written so that a reader who does not work in finance can follow our market notes without stopping. It is general information and not investment advice. Read the notes.