Three essays on the informational efficiency of financial markets through the use of Big Data Analytics.

Summary The massive increase in the volume of data generated every day by individuals on the Internet offers researchers the opportunity to approach the question of the predictability of financial markets from a new angle. Without claiming to provide a definitive answer to the debate between the proponents of market efficiency and behavioral finance researchers, this thesis aims to improve our understanding of the price formation process in financial markets through a Big Data approach. Specifically, this thesis focuses on (1) measuring investor sentiment at intraday frequency, and the relationship between investor sentiment and aggregate market returns,(2) measuring investor attention to economic and financial information in real time, and the relationship between investor attention and the dynamics of large-cap stock prices, and finally, (3) the detection of suspicious behaviors that may diminish the informational role of financial markets, and the relationship between the volume of activity on social networks and the stock prices of small-cap firms. The first essay proposes a methodology to construct a new indicator of investor sentiment by analyzing the content of messages posted on the social network Stock-Twits. By examining the specific characteristics of each user (level of experience, investment approach, holding period), this essay provides empirical evidence that the behavior of naive investors, subject to periods of excessive optimism or pessimism, has an impact on stock market valuation, in line with theories of behavioral finance. The second essay proposes a methodology to measure investors' attention to news in real time by combining data from traditional media with the content of messages sent by a list of experts on the Twitter platform. This test shows that when news attracts investors' attention, market movements are characterized by a sharp increase in traded volumes, increased volatility and price jumps. This essay also demonstrates that there is no significant information leakage when information sources are combined to correct a potential timestamp problem. The third essay investigates the risk of informational manipulation by examining a new dataset of Twitter posts about small-cap companies. This essay proposes a new methodology to identify anomalous behavior in an automated manner by analyzing user interactions. Given the large number of suspicious buy recommendations sent by certain groups of users, the empirical analysis and findings of this essay underscore the need for greater regulatory oversight of information posted on social networks as well as the value of better education of individual investors.