Exploiter les réseaux neuronaux et le traitement du langage naturel pour relever les défis des fondamentaux des marchés financiers.
Auteur / Autrice : | Abderraouf Birem |
Direction : | Vincent Bouvatier |
Type : | Projet de thèse |
Discipline(s) : | Sciences économiques |
Date : | Inscription en doctorat le 01/09/2023 |
Etablissement(s) : | Paris 12 |
Ecole(s) doctorale(s) : | École doctorale Organisations, marchés, institutions (Créteil ; 2010-) |
Partenaire(s) de recherche : | Laboratoire : ERUDITE - Equipe de Recherche sur l'Utilisation des Données Individuelles en lien avec la Théorie Economique |
Mots clés
Résumé
This project aims to challenge the Efficient Market Hypothesis (EMH) by applying Natural Language Processing (NLP) techniques in combination with neural networks to improve market efficiency. The project focuses on forecasting stock market indexes by extracting valuable insights from various textual sources, including news articles, social media data, and financial reports. It begins by formalizing the EMH and examining its validity in stock market predictive models, considering the influence of big data. The project then explores the use of Neural Networks and NLP to test the validity of EMH, using data from clean price APIs and scraping NLP data from multiple sources. It addresses the challenges of preprocessing and feature selection in NLP techniques. Furthermore, the project broadens the scope of NLP techniques by incorporating sentiment analysis beyond individual company data, investigating the influence of external factors on the stock market and analyzing the impact of news sentiment. Finally, it investigates the integration of NLP techniques into financial market volatility models, aiming to enhance the effectiveness of volatility models by converting textual data into quantitative information and considering the impact of news shocks on market volatility.