This thesis analyses risk factors defined in 10-K company documents and their impact on daily prices of stocks. First, using Latent Dirichlet Allocation (LDA) a fixed number of topics are extracted from all risk factor documents, and topic weights get assigned to individual documents. After that, using the same LDA model preprocessed historical news get topic weights. Given that weight vectors are available for individual company documents and news articles, the relationship between news articles and companies can be quantified. Based on these values, news sentiments and additional quantitative stock data the daily return is predicted by a supervised learning model.