Generative AI offers partial solution for banks' emissions data gaps
A Banca d'Italia paper investigates if Generative AI can address significant data gaps and inconsistencies in euro area banks' emissions reporting, particularly for scope 3. The study finds GenAI offers a partial solution but notes similar quality issues and concerns regarding replicability and transparency.
Persistent gaps in banks' emissions data
A Banca d'Italia paper highlights significant limitations in emissions data for major listed banks in the euro area, drawing from four leading providers.
The study identifies pervasive data gaps, inconsistencies across providers, and high volatility over time, particularly for scope 3 emissions.
These indirect emissions, often inexplicably lower than scope 2 and lacking correlation with banks' exposure to high-emitting sectors, hinder effective climate risk assessment and policy formulation.
Scope 3 emissions are critical for evaluating banks' climate risk exposure and green transition opportunities, yet their computation for financial institutions presents unique challenges beyond internal operations.
The paper notes that current data issues make these emissions largely unreliable for research and policy purposes.
This context motivates the exploration of Generative Artificial Intelligence (GenAI) tools as a novel and valuable source of information to potentially address these documented data gaps and improve data quality.
GenAI's mixed results in emissions data
The study tested three GenAI tools (Claude, ChatGPT, Gemini) to generate scope 3 emissions data for euro area banks in 2022.
GenAI-based data correlates with traditional sources, potentially filling gaps and identifying anomalies.
However, it suffers from similar quality and consistency issues as professional providers, suggesting shared data limitations.
Replicability and transparency are also concerns due to output variability.
Despite current drawbacks, GenAI shows future potential as a complementary data source, especially with fine-tuned models and improved underlying climate data.
AI's promise hinges on better data
Banks' scope 3 data, from any source, demands caution due to quality issues.
GenAI's potential relies on better underlying climate data and fine-tuned models.
Standardized disclosure for all firms is crucial to unlock AI's capabilities.