RESEARCH

Informing climate risk analysis using textual information - A research agenda (2024)
With Malte Schierholz, Bolei Ma, Jacob Beck, Andreas Dimmelmeier, Hendrik Christian Doll, Maurice Fehr, Frauke Kreuter, and Alex Fraser. 

This project is part of the larger research agenda GIST - Greenhouse Gas Insights and Sustainability Tracking, a research collaboration between Deutsche Bundesbank and LMU Munich to generate high-quality, granular firm-level emissions and sustainability data. More information can be found here. 

Abstract:
We present a research agenda focused on efficiently extracting, assuring quality, and consolidating textual company sustainability information to address urgent climate change decision-making needs. Leveraging technological advancements, particularly in large language models (LLMs) and RAG (Retrieval-Augmented Generation) pipelines, we aim to unlock the potential of underutilized textual data, such as sustainability reports, alongside global administrative collections and proprietary sources. Key challenges include ensuring the quality of automatically extracted data, for which we propose leveraging domain knowledge and addressing human labeler variations. We discuss the complexities of evaluation due to loosely defined concepts and emphasize the importance of integrating extracted results with existing data to create gold standard datasets. By incorporating novel variables, such as firms' future commitments, we envision enabling new use-cases. Ultimately, our approach aims to provide unified, easily accessible, and FAIR (Findable, Accessible, Interoperable, Reusable) climate-related data to inform decision-making efficiently, thereby contributing to tackling climate change in the short term.

Houston, we have a problem: Can satellite data fix the climate-related data gap? (2024)
With Andrés Alonso Robisco, José Manuel Carbó Martínez, and Elena Triebskorn.

Presentations: 

• BIS-IFC Workshop on "Addressing climate change data needs: the global debate and central banks' contribution", Central Bank of the Republic of Turkey, Izmir, 2024. 

Abstract:
Central banks and international supervisors have identified the difficulty of obtaining climate information as one of the key obstacles impeding the development of green financial products and markets. To bridge this data gap, the utilization of satellite imagery from Earth Observation (EO) systems may be necessary. In this paper, we analyze the potential of applying satellite data to green finance. To assess this, we first explore the policy debate from a central banking perspective. We then briefly describe the main challenges for economists in dealing with the EO data format and quantitative methodologies for measuring its economic materiality. Finally, using topic modeling, we perform a systematic literature review of recent academic studies to uncover in which topic areas satellite data is currently being used in green finance. We find the following research topics: physical risk materialization (including both acute and chronic risk), deforestation, energy and emissions, agricultural risk and land use and land cover. We conclude providing a comprehensive analysis on the financial materiality of this technology, mapping these research topics with new green financial instruments and markets under development, as well as some key considerations for policy discussion.

Extracting Data Citations with Large Language Models  (2024)
With Sebastian Seltmann and Hendrik Christian Doll.

Presentations: 

• Conference on Foundations and Advances of Machine Learning in Official Statistics
  (DESTATIS - Federal Statistical Office of Germany, Wiesbaden, 2024)

• 3rd IFC Workshop on Data Science in Central Banking 2023
  (Bank of Italy, Rome)

Abstract:
Empirical researchers and research data centers (RDCs) face challenges in efficiently understanding and categorizing data sources and methodologies used in scholarly papers. This process currently relies on human readers and is time-consuming and prone to errors. To address this, we explore the potential of using Large Language Models (LLMs), specifically GPT-3.5, to automate the identification and categorization of research data sources. We analyze the accuracy of GPT-3.5 in detecting and summarizing data sources and methods in economics and finance papers. By employing web-scraping techniques, we collect a comprehensive sample of research papers and create human-labeled validation datasets. We evaluate the detection and prediction accuracy and address the issue of false answers provided by the model. Additionally, we assess the pre-processing requirements of GPT-3.5 for cost-effective implementation. Our paper also provides a guide for implementing our proposed solution at research institutions and RDCs worldwide, aiming to enhance data analysis and research data provision services.

Climate-related data needs for central banks and beyond - Leveraging novel textual and image data (2024)
With Susanne Walter and Hendrik Christian Doll.

Abstract:
Central banks and supervisors urgently need climate-related data for incorporating evidence-based climate change considerations. Besides limited structured administrative and proprietary data sources, a wealth of dispersed climate-related information is readily available in the form of unstructured texts (e.g., company sustainability reports) and images (e.g., satellite images). We describe how central banks may leverage this easily accessible information to provide structured climate-related data to an interested wider public. In doing so, central banks can enhance data usability, validate quality, and extend the scope of existing data sources. Furthermore, such approaches can help overcome remaining restrictions from licensing commercial vendors. We investigate the resulting benefits, challenges, and the potential for international collaboration. As data quality remains a challenge with automatic extraction from novel data sources, we describe evaluation attempts. In a case study for Deutsche Bundesbank, we outline early-stage results and lessons learned from ongoing innovation projects with collaborating academic institutions. We propose that the rapid speed in technological developments (such as large language models and multimodal learning) enables value-creating data generation and provides a unique opportunity to help close data gaps in the short and medium term.

Includes upcoming presentations. Unpublished papers available upon request.