Correlations

This section covers how we locate stretches of an instrument's history that resemble the current one, and how to read what followed them.

The method is a pipeline of five decisions: how long a sample to take, how to make prices comparable, what cheaply removes the obviously unrelated, how to constrain amplitude, and where to put the similarity threshold. Every one of them changes the output, and none is universally correct.

It is worth stating early what the result is not. The output is a list of places with a comparable outline plus a picture of what came next. That is a description of the past. A share of outcomes does not become a probability because we computed it.