This thesis investigates the relationship between Re_nitiv's Environmental, Social and Governance (ESG) Scores and _nancial performance and introduces a new approach for updating ESG Scores more frequently using stakeholder data and leveraging state-of-theart natural language processing (NLP) models. Firstly, we _nd evidence that ESG Scores based on company reportings have a negative e_ect on how a company is valued in the market, whereas ESG Scores based on global media reports have a considerable positive e_ect. Exploring the connection between ESG Scores and _nancial perfomance through a portfolio-based approach, we _nd that the market allocates a higher risk-premium to companies with low sustainability activities and that this phenomenon is concentrated on those _rms that also have a rather positive sentiment portrayed in society by global media. By constructing an ESG Factor, we see a signi_cant outperformance when creating a negative spread on sustainability activities as reported by the company and a positive spread on sustainability activities as reported by the media. Building on these results and leveraging the ever growing body of stakeholder data, we propose an ESG scoring methodology based on Twitter discourse and powered by a pipeline of state-of-the-art NLP models, which dynamically and automatically reects ESG-relevant events in near real-time. Here, we _nd that our framework successfully detects material events in a Twitter sample and adjusts the ESG score accordingly, providing an e_cient, objective and independent assessment of _rms.     «

This thesis investigates the relationship between Re_nitiv's Environmental, Social and Governance (ESG) Scores and _nancial performance and introduces a new approach for updating ESG Scores more frequently using stakeholder data and leveraging state-of-theart natural language processing (NLP) models. Firstly, we _nd evidence that ESG Scores based on company reportings have a negative e_ect on how a company is valued in the market, whereas ESG Scores based on global media reports have a considera...     »