Higher-Order Moment Spillovers And Interpretable Prediction In Commodity Markets Using Arcd, Tvp-Var-Ejc, And Graph Neural Networks

Document Type

Article

Source of Publication

Risk Management-An International Journal

Publication Date

4-25-2026

Abstract

The ever-changing nature of financial markets underscores the need for early warning mechanisms to prevent and mitigate systemic financial risks. This paper proposes a novel approach to measure and predict the higher-order moment risk spillovers, offering early warning risk detection signals in commodity markets by integrating machine learning and traditional quantitative modeling. We employ a combination of the Autoregressive Conditional Density (ARCD) model, the Time-Varying Parameter Vector Autoregression Extended Joint Connectedness (TVP-VAR-EJC) model, and the improved Graph Convolutional Network (IGCN) model with an edge-deletion method. Our results show significant heterogeneity between volatility and higher-order moment risk spillover. We document that energy and precious metals are the main net risk transmitter and receiver of the moment-based spillovers. The pairwise net spillover between energy and precious metals contributes the most to the total commodity risk spillover prediction. Our results show that the proposed IGCN model outperforms alternative deep learning models such as LSTM, GRU, and Transformer.

ISSN

1460-3799

Publisher

Springer Science and Business Media LLC

Volume

28

Issue

2

Disciplines

Business

Keywords

Commodity market, Higher-order moment, Risk spillovers, Explainable machine learning, IGCN

Indexed in Scopus

no

Open Access

no

Link to Full Text

Share

COinS