FRANKFURT SCHOOL

BLOG
Treat or threat: Can Artificial intelligence really help close the ESG Data Gap?
FS-UNEP Centre / 31. Oktober 2024

  • Teilen

  • 766

  • 0

  • Drucken
Mathilde Bossut is a researcher at the Frankfurt School - UNEP Collaborating Centre for Climate and Sustainable Energy Finance and a PhD candidate at the Chair for Weather and Climate Risks at ETH Zurich. She holds a diploma in economics and finance from Sciences Po Strasbourg and a master's degree in energy economics. Prior to joining the FS-UNEP Centre, Mathilde gained experience in sustainable finance at the Climate Bonds Initiative and the Deutsche Bundesbank. Her research focuses on the impact of climate risks on corporate financial performance. In particular, she examines the contagion effect of economic damage from natural disasters on supply networks.

Autorenprofil

Mehr Blog Posts
Turning Ideas into Impact: The Case of Ceres FieldCheck
Entrepreneurship / 6. Mai 2024
Cooperation Company Project: A Journey Through Retail Location Analysis
Master in Applied Data Science / 24. April 2024
Erforschung der Auswirkungen von KI auf die Zukunft der Arbeit
Weiterbildung / 6. März 2024

Artificial Intelligence (AI) and Machine Learning (ML) have made their way into every aspect of our work life. By showing an incredible ability to identify complex data patterns, AI bears the promise of simplifying and improving daily business operations and has multiple implications in finance. So, what can we expect in the field of sustainable finance? Can AI help close the ESG data gap?

In this blog post, I reflect on our experience within the safe Financial Big Data Cluster. Since 2021, the Frankfurt School leads the working group on Sustainable Finance and seeks to leverage ML algorithms to bridge the ESG Data Gap for credit institutions.

 

AI – a game changer in sustainable finance

Amidst climate change and social challenges, ESG matters have become an important component of traditional financial practices. Credit institutions worldwide are considering methodologies in assessing and managing the environmental and social performance in their loan portfolios.

However, despite tremendous efforts to improve data availability, little information is available to inform adequate decision-making without increasing the administrative burden on creditors and portfolio managers. Artificial intelligence could provide time-efficient solutions and facilitate the use of alternative data sources.

We identify four major ways AI/ML algorithms could help bridge the ESG Data gap:

  • Structuring existing data: Today, large undertakings are generally reporting on their ESG performance – however not always in a machine-readable format. AI could assist portfolio managers in automatically reviewing TCFD or CSRD reports, classifying information in different ESG metrics and providing additional metrics on confidence and sentiment analysis.
  • Tapping into alternative data sources: There are more than one source of information that could provide ESG information on a firm beyond self-disclosure. AI developments in text, audio and image analysis could help make use of satellite imagery, news, speeches, and advertisement.
  • Dealing with missing data: By analysing existing data patterns, AI could help filling in missing values in ESG datasets – for example, with historical weather data – as well as estimating values for non-disclosing firms – widely used for carbon emissions estimates.
  • Better assessment of ESG Raw data: AI/ML algorithms can help us better understand the link between ESG metrics and financial performance using methods such as classification and prediction.

 

At the Frankfurt School, we looked at different ways AI can be explored to close the ESG Data Gap, by looking at Scope 3 estimates, climate resilience predictions and use of ML in climate stress-testing.

 

Promising in theory, challenging in practice

While AI/ML represents a massive opportunity for better consideration of ESG issues in the financial industry, there are still many challenges to tackle.

  • Limited legal flexibility. While ML often outperforms traditional statistics in handling complex datasets, it often comes with a cost: uncertainty, lack of transparency, and difficulty in retracing complex decision processes. For this reason, ML and the use of proxies in (ESG) credit risk models have received particular attention from European regulators. Both, the EBA and the ECB call for a careful and proportionate consideration and require banks to build the capacity to properly interpret results, understand the decision process, and identify potential model inaccuracies.
  • Heavy reliance on high-quality data. However, the current ESG raw data universe is still scarce, provides only narrow coverage of the whole company universe and overwhelmingly relies on estimates. Training AI models on such spartan foundation will introduce additional bias and yield inaccurate results. Alternative data solutions could play a critical role in strengthening AI applications.
  • Unequipped financial institutions. While satellite imagery or CEO speeches analysis are a golden mine for time-efficient firm environmental performance assessment, credit institutions do not have yet the capacity to collect such format, store, and process heavy datasets.

Credit institutions will have to weigh-in by comparing the increased time and resources efficiency as well as gained knowledge, with existing trade-offs: investments in data infrastructure and capacity building, and increased uncertainty. Banks that are willing to go down this path could yield an early-mover advantage in better grasping the link between sustainability and the assets they finance.

 

Conclusion

Yes, there are no doubts that AI could be an ally in navigating a topic as complex as the transition to a sustainable and climate-neutral economy. Yet, we need to carefully consider the role we want such models to play in planning our future. AI is about having a better understanding of historical data patterns, which may make it inherently incompatible with sustainability issues.

Our ability to navigate the transition depends on humanity’s ability to prepare and adapt for a large-scale unprecedented challenge. Relying solely on AI to build a liveable future might be as good as “driving forward while only looking through the rear mirror”.

0 Kommentare

Senden