Non-parametric Causal Discovery for EU Allowances Returns Through the Information Imbalance

2025-08-21

AI Summary

This study investigates the causal drivers of EU Allowance (EUA) yield dynamics, focusing on nonlinear relationships that conventional linear methods fail to detect. Method: We propose a nonparametric causal discovery framework grounded in Differentiable Information Imbalance (DII), benchmarking it against linear Granger causality and VAR models using both empirical data (2013–2024) and synthetic datasets. Contribution/Results: We identify coal futures prices and the IBEX35 index as key common causal variables influencing EUA yields. DII robustly uncovers nonlinear causal pathways missed by linear approaches and demonstrates superior accuracy and robustness on synthetic data. To our knowledge, this is the first application of DII in carbon finance, revealing previously overlooked nonlinear causal structures underlying EU ETS price formation. The findings advance methodological foundations for modeling carbon market dynamics and inform evidence-based policy design.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Economics, Online Markets and Human Computation: Sustainability of Web economics
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

We propose to use a recently introduced non-parametric tool named Differentiable Information Imbalance (DII) to identify variables that are causally related -- potentially through non-linear relationships -- to the financial returns of the European Union Allowances (EUAs) within the EU Emissions Trading System (EU ETS). We examine data from January 2013 to April 2024 and compare the DII approach with multivariate Granger causality, a well-known linear approach based on VAR models. We find significant overlap among the causal variables identified by linear and non-linear methods, such as the coal futures prices and the IBEX35 index. We also find important differences between the two causal sets identified. On two synthetic datasets, we show how these differences could originate from limitations of the linear methodology.

Problem

• Identifying causal variables for EU Allowances returns
• Comparing non-linear and linear causal discovery methods
• Analyzing limitations of linear approaches in causality

Innovation

• Non-parametric causal discovery using DII
• Comparing DII with Granger causality methods
• Identifying non-linear relationships in EUA returns