A tone score can satisfy the measurement
dossier and
still be unusable as a backtest feature. The text has an availability time. The research path that produced the score
has a width: which list, which section, which window, which model. This note applies the common-core rules to the
running case. A 10-K or earnings-call tone measure is not a tradable signal until it survives the same point-in-time
and multiple-testing discipline as any other equity feature.
Learning objectives. By the end of this note you should be able to:
- time-stamp a 10-K tone score and a call score using the availability rule;
- list the researcher degrees of freedom that a text pipeline adds;
- state the 2011 one-year portfolio result in a way that blocks an accidental trading claim;
- separate a valid measurement from a signal that has not been confirmed.
In this lesson
- Availability is part of the feature
- The search is part of the evidence
- What has to be in the signal file
- A timestamp ledger for one firm-year
Before you start
The full lesson, with worked examples and exercise solutions, is for members.
Read the full lessonNew to SelfMadeQuant.com? See membership options or browse the Text as Data and NLP for Finance syllabus.