Title: Trading the stock market using Google search volumes: a long short-term memory approach

Authors: Joseph St. Pierre; Mateusz Klimkiewicz; Adonay Resom; Nikolaos Kalampalikis

Addresses: Worcester Polytechnic Institute, 100 Institute Road, Worcester, MA 01609, USA ' Worcester Polytechnic Institute, 100 Institute Road, Worcester, MA 01609, USA ' Worcester Polytechnic Institute, 100 Institute Road, Worcester, MA 01609, USA ' Worcester Polytechnic Institute, 100 Institute Road, Worcester, MA 01609, USA

Abstract: In this paper, we present a methodology for utilising Google search indices obtained from the Google trends website as a means for measuring potential investor interest in stocks listed on the Dow Jones index (Dow 30). We accomplish this task by utilising a long short-term memory network that correlates changes in the search volume for a given asset with changes in the actual trade volume for said asset. Additionally, by using these predictions, we formulate a concise trading strategy in the hopes of being able to outperform the market and analyse the results of this new strategy by backtesting across weekly closing price data for the last six months of 2016. Furthermore, with a success rate of 43%, we believe our algorithm to be scalable beyond the narrow scope of this study and could in fact be applicable to numerous other assets on the market.

Keywords: artificial intelligence; machine learning; long short-term memory; algorithmic trading.

DOI: 10.1504/IJFERM.2019.096669

International Journal of Financial Engineering and Risk Management, 2019 Vol.3 No.1, pp.3 - 18

Received: 16 Dec 2017
Accepted: 01 Mar 2018
Published online: 07 Dec 2018 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article