Identification of climate disclosures in European bank reports using Large Language Models
-
SeriesResearch Master Defense
-
SpeakerAna-Maria Tecuci
-
LocationOnline
-
Date and time
August 30, 2023
15:00 - 16:00
In this paper I investigate the capability of Large Language Models to identify and extract climate disclosures information from documents and reports published by European banks. Using a novel and proprietary data set of expert assessments, I find that the pre-trained DistilBERT SQuAD modeI and Climate-BERT are greatly surpassed in performance by the ChatGPT-3.5 model, while a relatively good performance can also be achieved with a fine-tuned BERT. These results suggest that Large Language Models can add value in the identification of passages within very long documents by efficiently helping humans focus on the soundness of the disclosures instead of the existence of the disclosures.