IA & Sustainable Finance – PHASE II

Directors
- Marc-Antoine Dilhac, Algora Lab (Mila), OBVIA
- Manuel Morales, University of Montreal, Fin-ML
- Rheia Khalaf, Fin-ML
Researchers
- Emanuel Lemus-Monge – Algora Lab
- François Hu – University of Montreal
- Patricia Gautrin – Algora Lab
Partners:
OBVIA, Fin-ML and Algora Lab are jointly carrying out PHASE II of the “AI and Sustainable Finance” project
Development of a natural language processing research application for the evaluation of corporate transition plans
As a follow-up to a preliminary study on the different uses of artificial intelligence to accelerate the transition to sustainable finance (i.e. E.S.G), conducted jointly by researchers from Algora Lab, Fin-ML and Cirano, an application based on language patterns has been identified. This model will be trained for fact-checking to automatically assess companies’ transition plans.
Implementing this model requires conducting data collection (i.e. corporate transition plans), labeling relevant snippets, modeling fact-checking, training and evaluating model performance. The architecture used will be the ‘Retrieval Augmented Generation’ (i.e. R.A.G), considered the state of the art for fact-checking, the model combines different components based on ‘transformers’ to retrieve evidence from a large corpus of documents and other component capable of generating an accurate answer based on the evidence found.
The main purpose of this application is to support investors in their decision-making related to the transition of companies. In order to achieve the goals needed to combat climate change, a deep reallocation of investment capital must take place quickly (Pörtner et al., 2022). However, in the light of our analysis of the sustainable finance ecosystem, we note an informational gap between companies and investors that hinders the allocation of capital necessary for the transition.
This gap can be explained by a lack of upstream regulation leading to a mismatch of downstream frameworks and standards. This lack of harmonization complicates and raises the barrier to entry for companies wanting to start their disclosure efforts, reducing the number of
voluntary disclosures available (Bartlett et al., 2022). To overcome this lack of information, investors should use alternative sources of data (financial press, networks social and other third party information) obtained through data providers intermediaries centralizing this data. The economic model of these suppliers is based on data collected and the analysis of this information in order to build financial indices on the performance of companies in their transition. The collection and analysis methodologies are opaque and often inconsistent owners from one supplier to another preventing a real comparability of companies present in the ecosystem (Berg et al., 2019).
Our app aims to help investors navigate this ecosystem more easily by centralizing the information and evaluating it automatically. To do this, a model of text retrieval and natural language processing has been trained to extract corporate reports related to climate disclosure and assess the information contained under the prism of a given disclosure framework. Additionally, our solution can also be used by companies wishing to obtain feedback on their transition plan before disclosing it publicly or for anyone wishing to analyze the transition plan from a company. Examples of usage will be provided later in the report. This project does following a previous publication by our researchers on a model for analyzing the alignment of climate disclosures to the SASB framework (Kheradmand et al., 2021).