I recently started up a company that does in depth text analysis of all 10-K and 10-Q filings using a new field - sentiment analysis.

The sentiment categories we use are based on the copyrighted dictionaries of Loughran and McDonald.  After parsing the filing into a vector of words, we generate a count of words for each category and report the portion of words appearing in each of the categories.  The dictionaries can be found at www.nd.edu/~mcdonald.  From prior literature, negative word counts generally are most closely associated with firm performance, however Loughran and McDonald show that the other categories are related to stock price volatility, shareholder lawsuits, and earnings surprises.

We tabulate word counts for the current document, and where available, compare them with counts from the 10-K/Q filing from the preceding quarter.  For comparison, we report the average proportion for each sentiment category across all 10-K filings over the past five years.  We also create a measure based on the regression results in Loughran and McDonald that correlates the collective sentiment proportions  on a 1-10 scale with day [0,3] ("4-Day") and subsequent one year ("1-Year") price performance.  These scores are reported as part of the research on correlations in textual analysis and are not intended as a forecast of future performance.

http://www.finif.com

Please take a look and I'm looking for input and suggestions.

Thanks
Mike