Semantic Frames to Predict Stock Price Movement

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 873–883, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Semantic Frames to Predict Stock Price Movement Boyi Xie, Rebecca J. Passonneau, Leon Wu Center for Computational Learning Systems Columbia University New York, NY USA (bx2109|becky|leon.wu)@columbia.edu Germ´an G. Creamer Howe School of Technology Management Stevens Institute of Technology Hoboken, NJ USA gcreamer@stevens.edu Abstract Semantic frames are a rich linguistic re- source. There has been much work on semantic frame parsers, but less that applies them to general NLP problems. We address a task to predict change in stock price from financial news. Seman- tic frames help to generalize from spe- cific sentences to scenarios, and to de- tect the (positive or negative) roles of spe- cific companies. We introduce a novel tree representation, and use it to train predic- tive models with tree kernels using sup- port vector machines. Our experiments test multiple text representations on two binary classification tasks, change of price and polarity. Experiments show that fea- tures derived from semantic frame pars- ing have significantly better performance across years on the polarity task. 1 Introduction A growing literature evaluates the financial effects of media on the market (Tetlock, 2007; Engel- berg and Parsons, 2011). Recent work has applied NLP techniques to various financial media (con- ventional news, tweets) to detect sentiment in con- ventional news (Devitt and Ahmad, 2007; Haider and Mehrotra, 2011) or message boards (Chua et al., 2009), or discriminate expert from non- expert investors in financial tweets (Bar-Haim et al., 2011). With the exception of Bar-Haim et al. (2011), these NLP studies have relied on small corpora of hand-labeled data for training or evalu- ation, and the connection to market events is done indirectly through sentiment detection. We hy- pothesize that conventional news can be used to predict changes in the stock price of specific com- panies, and that the semantic features that best represent relevant aspects of the news vary across On Wednesday, April 11th, 2012, Google Inc announced its first � � � � quarterly earnings report, a week before the April 20 options contracts expiration in contrast to its history of reporting a day before monthly options expirations. The stock price of Google surged 3.85% from April 10th’s $626.86 to 12th’s $651.01. On Friday, April 13th, news reported Oracle Corp would sue � � � � Google Inc , claiming Google’s Android operating system tramples � � � � its intellectual property rights . Jury selection was set for the next Monday. Google’s stock price tumbled 4.06% on Friday, and continued to drop in the following week. Figure 1: Summary of financial news items per- taining to Google in April, 2012. market sectors. To test this hypothesis, we use price information to label data from six years of financial news. Our experiments test several doc- ument representations for two binary classification tasks, change of price and polarity. Our main con- tribution is a novel tree representation based …

Content truncated to 500 words