The first essay examines news and the cross section of returns. Using a sentiment score provided by Thomson Reuters to measure the tone of news articles, this paper examines monthly portfolio returns constructed from information about past news articles. The sentiment score is obtained from the kind of words and phrases that are used in the news article. Positive tone in news articles in the past months predicts positive returns. Similarly, negative tone in the past months predicts negative returns. Past sentiment predicts future returns even for large stocks. The predictive ability of past sentiment dominates the predictive ability of past returns. After controlling for past sentiment, the predictive ability of past returns (in predicting future return) disappears. The findings are robust to multiple specifications. The predictive ability of past sentiment can be used profitably. When applied to the largest decile of stocks, a strategy that takes a long position in stocks with past positive sentiment score and a short position in stocks with past negative sentiment score generates a statistically significant alpha of 34 basis points per month.

The resulting portfolio is also positively correlated with a long-short momentum portfolio. Within the same time period, a trading strategy using the sentiment scores from the subset of news articles citing analysts is not profitable. The news items that cite analysts have economically significant contemporaneous returns. The findings suggest that (i) the market underreacts to information contained in news articles, (ii) momentum might be related to underreaction to the sentiment information, and (iii) market participants pay attention to sentiment score information in analyst news. The findings are consistent with a model where one trader has private information and others are trading based on past returns and volume information. The paper also shows that after adjusting for firm size, stocks with abnormally high counts of news articles underperform stocks with normal counts of news. Stocks with abnormally low newscounts also underperform.

The second essay examines the relationship between news and trading activity. The theory of trading game invariance of Kyle and Obizhaeva(2009) predicts that for every one percent increase in trading activity, the frequency of news articles should increase two-thirds of one percent. Using news data from 2003 to 2008, we show that the cross-sectional variation in news articles across stocks is related to the trading activity in a manner consistent with the trading game invariance. The relationship is robust to various estimation procedures including models of count data. The relationship is also robust to multiple ways of counting news and excluding various type of firm specific news.