Follow the Words: Forecasting Stock Price Movement and the Predictive Power of Earnings Calls

Abstract

This thesis explores the expanding intersection of finance, Natural Language Processing, and linguistic analysis. This paper examines if a linguistic analysis of a company’s earnings call holds predictive power in forecasting the movement of the company’s stock price on the day after the earnings conference call. Our question is answered through extracting linguistic features from the transcripts of earnings conference calls and building machine learning models to predict stock price movement. First, an analysis of 13 different machine learning classifier’s performances on predicting stock price movement on days both with and without earnings calls is provided. Then, an analysis is conducted on samples the day after an earnings call. A discussion of the effectiveness of several different linguistic analysis techniques in a financial context concludes the research. Ultimately, it is found that while predicting stock price movement is a difficult task due to the amount of noise in financial data, with observed accuracy improvements over baseline models as evidence, there is predictive power in linguistic analyses of earnings conference calls, and linguistic features can be used to supplement financial data in predicting stock price movement.