I am exploring the financial-related dictionary of Loughran and McDonald. I want to calculate the TF-IDF weighting of the words within the dictionary. I use the following formula, based on the paper "When Is a Liability Not a Liability":
Q1: Which value should I use for N (the total number of 10-Ks in the sample)? If I use the cleaned (50,115) or the raw (121,217) number of 10-Ks, I get negative values for some words. For example, the word “almost” has a value of 126,738 for Doc Count. However,
log(121,217 / 126,738) yields a negative value which is not possible.
Q2: What does column E and F of the master dictionary represent? Are these values equal to tfi,j the raw count of the ith word in the j-th document, and aj the average word count in the document (page 42 in the paper)?