Content of Annual Reports as a Predictor for Long Term Stock Price Movements

This paper examines the possibility of automatic extraction of future stock price information from the annual Form 10-K produced by publicly traded companies in the United States of America. While previous approaches to automatically interpreting corporate documents have tended to utilize extensive expert knowledge to preprocess and analyze documents, our approach inputs documents verbatim to a compression classifier. We demonstrate the effectiveness of the new approach on a newly constructed dataset based around the Dow Jones Industrial Average over the period 1994-2009. We find statistically significant increase in average returns of stocks recommended by the new system as compared with the Dow as a whole. Also examined are two hypotheses regarding the predictive power of 10-K reports. First, whether congressional attempts to make Form 10-K filings more informative had a measurable impact, and second, whether the filings have long-term predictive value in a dynamically changing