Overview

A systematic researcher who wants a 10-K or earnings-call tone feature has already accepted the common-core discipline on hypotheses, timestamps, and validation design. This topic is the measurement decision that comes first: which documents enter the corpus, and which numerical representation of those documents could carry financial tone. Representation is a compression. There is no single correct compression for every research question, and a general-purpose negative dictionary is not a finance-ready tone measure.

By the end of this note you should be able to:

• state the four selection and representation principles that govern a filing corpus;
• build a bag-of-words document-feature matrix and a cosine comparison of two filings;
• explain why a Harvard negative list misclassifies common 10-K language, and what the Loughran–McDonald lists measure instead.

1. Contents
2. The big picture