Journal of Risk

Farid AitSahlia    
Warrington College of Business, University of Florida

This issue of The Journal of Risk presents four studies, one that introduces quantitative frameworks designed to enhance liquidity risk assessment of non-maturity deposits under interest-rate stress scenarios; one that addresses class imbalance in default modeling; one that improves graph-based stock forecasting; and one that integrates international extreme-value information with crude oil volatility spillovers.

In the issue’s first paper, “Non-maturity deposit risk under interest rate stress: a behavioral modeling framework”, Luca Cappellina and Domenico Sartore address the internal inconsistency in bank risk modeling, where behavioral assumptions often conflict with regulatory interest rate shocks. This is critical because misaligned models can underestimate liquidity risk during market stress. The authors contribute a modular multivariate logistic regression behavioral framework relying exclusively on financial market variables that are consistent with the supervisory interest rate shock scenarios prescribed by the Basel Committee on Banking Supervision. Their findings reveal that while short-term rates primarily drive withdrawals, yield curve slopes act as significant amplifiers of depositor responses. Broadly, this work supports an integrated view of interest rate and liquidity risk, providing a practical tool for banks to ensure funding stability under supervisory stress scenarios.

In “Addressing class imbalance in probability-of-default modeling: a comparative study of SMOTE, ensemble learning and explainable artificial intelligence”, the second paper in the issue, Konstantinos Papalamprou investigates class imbalance in probability-of-default modeling, where rare default events often bias models and lead to poor risk sensitivity. His paper contributes a governance-aligned framework that integrates domain-informed feature engineering, ensemble learning and explainability based on Shapley additive explanations analysis. A key finding is that economically grounded feature transformations outperform the synthetic minority oversampling technique in both discrimination and transparency. In effect, this approach bridges the gap between complex machine learning and supervisory requirements, providing a regulator-ready pathway for accurate, auditable and stable credit risk assessment in financial institutions.

Our third paper, “Stock trend forecasting with graph neural networks” by Yao Lu and Zhangxi Chen, presents a novel graph neural network framework that represents sliding windows of stock prices as temporal graphs integrating technical analysis indicators such as the 5-day moving average and the 14-day relative strength index. The authors’ empirical results significantly outperform traditional machine learning and sequential deep learning models, highlighting the potential of graph-based modeling to capture complex structural dependencies and offering a robust alternative for enhancing the accuracy of financial forecasting systems.

Finally, in “Forecasting Chinese crude oil futures’ volatility: a heterogeneous volatility spillover-conditional autoregressive range model”, Xinyu Wu, Rongrong Tu and XiaonaWang detail the challenge of accurately forecasting volatility in crude oil futures on the Shanghai International Energy Exchange, a critical measure of market risk given China’s rising global influence. The authors’ proposed heterogeneous volatility spillover-conditional autoregressive range (HVS-CARR) model integrates intraday extreme-value information with time-varying volatility spillovers from, for example, the US West Texas Intermediate crude oil futures market. The paper’s empirical results demonstrate that Wu et al’s model outperforms traditional return-based and single-market models, particularly during crises such as the Covid-19 pandemic.