Predicting the French Stock Market Using Social Media Analysis

Predicting the French Stock Market Using Social Media Analysis

Vincent Martin (Université de Toulon, CNRS, LSIS, UMR 7296, 83957, La Garde, France and Aix Marseille Université, CNRS, ENSAM, LSIS, UMR 7296, 13397, Marseille, France & COEXEL, Toulon, France), Emmanuel Bruno (Université de Toulon, CNRS, LSIS, UMR 7296, 83957, La Garde, France & Aix Marseille Université, CNRS, ENSAM, LSIS, UMR 7296, 13397, Marseille, France) and Elisabeth Murisasco (Université de Toulon, CNRS, LSIS, UMR 7296, 83957, La Garde, France & Aix Marseille Université, CNRS, ENSAM, LSIS, UMR 7296, 13397, Marseille, France)
DOI: 10.4018/IJVCSN.2015040104
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

In this article, the authors try to predict the next-day CAC40 index. They apply the idea of Johan Bollen et al. from (Bollen, Mao, & Zeng, 2011) on the French stock market and they conduct their experiment using French tweets. Two analyses are applied on tweets: sentiment analysis and subjectivity analysis. Results of these analyses are then used to train a simple neural network. The input features are the sentiment, the subjectivity and the CAC40 closing value at day-1 and day-0. The single output value is the predicted CAC40 closing value at day+1. The authors propose an architecture using the JEE framework resulting in a better scalability and an easier industrialization. The main experiments are conducted over 5 months of data. The authors train their neural network on the first of the data and they test predictions on the remaining quarter. Their best run gives a direction accuracy of 80% and a mean absolute percentage error (MAPE) of 2.97%. In another experiment, the authors retrain the neural network each day which decreases the MAPE to 1.14%.
Article Preview

Introduction

Stock exchange is a fascinating subject for researchers because it can be studied from many points of views: economics (mother discipline1), computer science (model and analyze data), psychology (understand behaviors), politics, etc. Modern stock market has completely changed the way to buy and sell financial assets. Internet and computers now allow a real-time access to financial markets and the amount of available data (directly related to stock markets or not) is increasing day after day, especially with social medias where people can discuss and react about new information and give their opinions about the market. Therefore, in modern stock market approaches, we have to analyze more and more data, which is quasi-impossible without some kinds of summarization and/or interpretation. And even if we could correctly interpret all available information, the strong version of the Efficient Market Hypothesis (EMH) (Fama, 1965) states that it’s impossible to beat the market over the long term. Does that mean that trying to predict future stock values is a lost fight? Bollen et al. have shown that twitter mood is a good indicator to predict stock market (Bollen, Mao, & Zeng, 2011) and there are many other works that show the lacks of the EMH (see RELATED WORKS).

Our experiments go in this direction: we try to predict the French CAC40 stock market index which represents the 40 more important values in term of market capitalization. The goal of this paper is to study the correlation between public opinions extracted from Twitter and the French stock market. We know that social media information is sometime subjective and therefore we also measure the subjectivity in order to discover if this aspect is correlated with stock market fluctuations.

The remaining part of this paper is organized as follows. Section 1 presents the related works. Section 2 focuses on our methodology for social media (Twitter) analysis and French stock market prediction. Section 3 describes the experiments and presents our results. Finally we conclude and give some perspectives for future works.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 9: 4 Issues (2017): 2 Released, 2 Forthcoming
Volume 8: 4 Issues (2016)
Volume 7: 4 Issues (2015)
Volume 6: 4 Issues (2014)
Volume 5: 4 Issues (2013)
Volume 4: 4 Issues (2012)
Volume 3: 4 Issues (2011)
Volume 2: 4 Issues (2010)
Volume 1: 4 Issues (2009)
View Complete Journal Contents Listing