Financial Filings Are Less Depressing Than They Seem

One of the way that quant sorts and academic researchers mess about with financial filings is by studying the relationship between words found in a financial filing and subsequent performance. While the relationship is loose, at best, it sometimes points to important changes faster than human readers detect the same thing, especially in long filings.

But there is a problem. The kinds of words typically used to do this sort of analysis are often the sorts of words we normal humans use in everyday life, but financial lingo re-purposes many of those words for different purposes (e.g., tax, board, foreign, vice, and liability). Downbeat lingo in normal language can be less downbeat when found in a financial filing

A new paper wrestles with the issue in good, geek-ish fashion. It shows that almost three-quarters of the negative-sounding context in the reports is actually just financial people using their usual depressing language. Read on:

When is a Liability not a Liability? Textual Analysis, Dictionaries, and 10-Ks

Previous research uses negative word counts to measure the tone of a text. We show that word lists developed for other disciplines misclassify common words in financial text. In a large sample of 10-Ks during 1994-2008, almost three-fourths of the word count identified as negative by the commonly used Harvard Dictionary represents words that typically do not have negative meaning in a financial context. Words like tax, board, foreign, vice, and liability, simply describe company operations. Two potential solutions are explored. First, we develop an alternative negative word list that better reflects the tone of financial text. Second, we show that using a common term weighting scheme reduces the noise introduced by misclassifications. Without term weighting, our list generally outperforms the Harvard list; with weighting the performance appears comparable. However, we also find evidence that some of the power of the Harvard list could be attributable to misclassified words that proxy for other effects. [Emphasis mine]