A1378
Title: Geometric deep learning for realized covariance matrix forecasting
Authors: Andrea Bucci - University of Macerata (Italy) [presenting]
Michele Palma - Bocconi University (Italy)
Chao Zhang - The Hong Kong University of Science and Technology (China)
Abstract: Traditional methods employed in matrix volatility forecasting often overlook the inherent Riemannian manifold structure of symmetric positive definite matrices, treating them as elements of Euclidean space, which can lead to suboptimal performance and difficulties in parameter interpretation. Moreover, they often struggle to handle high-dimensional matrices. A novel approach for forecasting realized covariance matrices of asset returns is proposed using a Riemannian-geometry-aware deep learning framework. In this way, the geometric properties of the covariance matrices account for both possible non-linear dynamics and efficient handling of high-dimensionality. The efficacy of the approach is demonstrated using daily realized covariance matrices for the 20 most capitalized companies in the S\&P 500 index, showing that the method outperforms traditional approaches in terms of forecasting accuracy.
