Abstract: In this talk we discuss how to build a Twitter sentiment model in Python using Word2Vec and long short-term memory networks (LSTMs), comparing and contrasting with more conventional statistical models. We cover basic Natural Language Processing (NLP) techniques, providing different ways to extract features from text data for use in modeling. We describe a potential use of this sentiment model in developing cross-sectional algorithmic trading signals for factor models, expanding upon previous work using Twitter data.
Bio: Max's background is in applied mathematics, statistics, and quantitative finance. He runs the online lecture series at Quantopian and is responsible for workshop curriculums and educational content. In addition to having experimented with algorithmic trading of cryptocurrencies and Bayesian estimation of covariance matrices, Max has published work in theoretical mathematics. He works with top universities including Columbia, U Chicago, and Cornell and holds a MS in Mathematical Finance from Boston University.