Filing language density
SentimentAdvanced
Lexical density is a text measure of how much content, rather than function words, sits in a company filing. Research versions treat a change in the language of 10-Ks or similar filings as information, and they rank stocks on that change. It is a document strategy. It is not a headline sentiment score, and it is not a trading tip extracted from a press release.
Not investment advice. Nothing here is a recommendation to buy or sell anything. Backtests and historical studies are not guarantees, and a rule that worked in one sample can fail in the next. Trading can lose money. Links are for education, not endorsements. Do your own research.
- Rules
- On each filing date, compute the lexical-density measure exactly as the specification defines it, using only the filing text available that day. Rank the cross-section and go long one extreme and short the other. Hold until the next filing cycle or for a stated number of months. Align the trade date with the filing timestamp, not with the fiscal year-end.
- Where AI fits
- AI can parse filings, compute the text statistic, and prevent look-ahead from amended filings that appear later. This is a better use of a language model than asking it to 'read the 10-K and decide.' The code that counts words has to be frozen. Prompting a chatbot for a score is not a replication.
- Markets
- Individual stocks that file comparable reports. Comparing a 10-K with a foreign annual report needs a stated normalization.
- Timeframes
- Event-driven on filing dates, with holds measured in weeks or months.
- Risks
- Boilerplate and accounting-standard changes move text statistics without moving the business. Amendments and exhibits confuse parsers. The sample of filings per year is small. A language model that uses knowledge from after the filing date leaks.