A0182
Title: Focused SAFE-driven AI approach for volatility prediction in EU ETS
Authors: Emanuela Raffinetti - University of Pavia (Italy) [presenting]
Maria Elena De Giuli - University of Pavia (Italy)
Abstract: The EU emissions trading system (EU ETS) involves regulatory changes and sectoral transformations whose non-linear dynamics require the employment of powerful forecasting techniques. Recent literature showed that artificial intelligence (AI) systems appear more effective than classical econometric approaches in detecting the complexities of the EU ETS. Despite their advantages, the black-box nature of AI systems may give rise to direct outputs without a clear connection with the inputs that generate them. This is the reason why highly complex machine and deep learning models have to be supported by actions addressed to avoid, or at least restrict, the dangerous effects derived from their incorrect use. To this purpose, the European Commission has promoted focused safety principles for a trustworthy AI, specified in terms of sustainability, accuracy, fairness, and explainability. In line with these premises, the contribution is twofold: 1) on the theoretical side, the metrics associated with the principles of accuracy and explainability is formalized in order to investigate the role of historical data in shaping the realized volatility of the EUA prices and the implied volatility inferred from the EUA-related financial instruments; 2) on the empirical side, the proposed methodology is implemented on EUA price data.
