Machine Learning and the Stock Market

Abstract

Recent advances in machine learning methodologies have improved the usefulness of the technology. This paper examines whether machine learning using only past prices as the input can detect mispricings. Generally searching for mispricings is a slow process and can easily suffer from data-snooping. This paper provides a machine learning algorithm to search for mispricings while controlling for data-snooping. The process generates significant out-of-sample alpha. Overall, the results show that mispricings still exist, but have decreased over time, implying that markets have recently become more efficient.

Keywords: Asset Pricing, Anomalies, Mispricings, Machine Learning, Big Data Analysis

JEL Classification: B26, G12, G14, C58, N20