Options
Sentiment and econometrics : toward a unified framework of textual sentiment analysis for economic and financial applications
Maison d'édition
Neuchâtel
Date de parution
2020
Mots-clés
- Agrégation
- Analyse de Sentiment
- Analyse Textuelle
- Données Qualitatives
- Econométrie
- EPU
- ESG
- Investissement Durable
- Régression Pénalisée
- R
- Sentométrie
- sentometrics
- Séries Temporelles KEYWORDS : Aggregation
- Econometrics
- EPU
- ESG
- Penalized Regression
- Qualitative Data
- R
- Sentiment Analysis
- Sentometrics
- sentometrics
- Sustainable Investment
- Textual Analysis
- Time Series
Agrégation
Analyse de Sentiment
Analyse Textuelle
Données Qualitatives
Econométrie
EPU
ESG
Investissement Durabl...
Régression Pénalisée
R
Sentométrie
sentometrics
Séries Temporelles K...
Econometrics
EPU
ESG
Penalized Regression
Qualitative Data
R
Sentiment Analysis
Sentometrics
sentometrics
Sustainable Investmen...
Textual Analysis
Time Series
Résumé
Cette thèse de doctorat apporte trois contributions clés. Premièrement, il donne un aperçu des méthodologies actuelles et des applications empiriques à l'intersection de l'analyse de sentiment et de l'économétrie, tout en proposant de nouvelles pistes d'amélioration. Ce faisant, il formalise ce domaine de recherche émergent, le qualifiant “sentométrie” (<i> sentometrics </i>), qui est un portemanteau de sentiment et d'économétrie (<i>econometrics </i>). Deuxièmement, il décrit un logiciel open source d'un flux de travail complet pour passer de données textuelles qualitatives à des variables (séries temporelles) quantitatives de sentiment et extraire des informations (économétriques) de celles-ci. La première et la deuxième contribution comblent le manque existant auparavant d'une approche unifiée de l'utilisation de données de sentiment alternatives pour obtenir des informations pour une analyse économique et financière. Troisièmement, le flux de travail est adopté et ajusté pour deux applications. La première se situe dans la gestion d'actifs durables. Un grand corpus de nouvelles presse belges et néerlandaises est transformé en signaux quotidiens qui suivent les articles aux sujets d'Environnement (<i>Environmental </i>), de Social et de Gouvernance d'entreprise (<i>corporate Governance </i>), en abrégé ESG. Les signaux textuels se montrent utiles pour sélectionner d'un univers de portefeuille un plus petit sous-ensemble d'entreprises plus durables mais (au moins) aussi performantes. La deuxième application s’occupe de la construction et de l’analyse des indicateurs mensuels et quotidiens sur la base des articles de presse belges à propos de l’incertitude de la politique économique (<i>economic policy uncertainty </i>; ou EPU).
ABSTRACT:
This doctoral thesis has three key contributions. First, it overviews the current methodologies and empirical applications at the intersection of sentiment analysis and econometrics, while also proposing new avenues for improvement. Doing so, it formalizes this emerging research field, terming it “sentometrics”, which is a portmanteau of sentiment and econometrics. Second, this thesis describes an open-source implementation of a complete workflow to go from qualitative textual data to quantitative (time series) sentiment variables and extract (econometric) insights from those. These two contributions fill in the previously existing lack of a unified approach to using alternative sentiment data to obtain insights for economic and financial analysis. Third, the workflow is adopted and adjusted for two applications. The first one is in sustainable asset management. A large corpus of Belgian and Dutch news is transformed into daily signals that track news reporting relevant to Environmental, Social and corporate Governance (ESG) topics. The textual signals prove useful to restrict an investment portfolio universe to a smaller subset of more sustainable yet at least equally performing companies. The second application deals with the construction and analysis of monthly and daily news-based Economic Policy Uncertainty (EPU) indices for Belgium.
ABSTRACT:
This doctoral thesis has three key contributions. First, it overviews the current methodologies and empirical applications at the intersection of sentiment analysis and econometrics, while also proposing new avenues for improvement. Doing so, it formalizes this emerging research field, terming it “sentometrics”, which is a portmanteau of sentiment and econometrics. Second, this thesis describes an open-source implementation of a complete workflow to go from qualitative textual data to quantitative (time series) sentiment variables and extract (econometric) insights from those. These two contributions fill in the previously existing lack of a unified approach to using alternative sentiment data to obtain insights for economic and financial analysis. Third, the workflow is adopted and adjusted for two applications. The first one is in sustainable asset management. A large corpus of Belgian and Dutch news is transformed into daily signals that track news reporting relevant to Environmental, Social and corporate Governance (ESG) topics. The textual signals prove useful to restrict an investment portfolio universe to a smaller subset of more sustainable yet at least equally performing companies. The second application deals with the construction and analysis of monthly and daily news-based Economic Policy Uncertainty (EPU) indices for Belgium.
Notes
Doctor of finance, Doctor of business economics, Université de Neuchâtel, Vrije Universiteit Brussel, Institut d'analyse financière
Identifiants
Type de publication
doctoral thesis
Dossier(s) à télécharger