Hybrid LASSO-SVQR Framework for Modeling Nonlinear Systemic Risk Dependencies

Conditional Value-at-Risk (CoVaR) has become a key measure for assessing systemic risk by capturing interdependencies among financial institutions. However, existing CoVaR estimation methods often rely on linear assumptions or non-convex models, limiting their ability to capture complex nonlinear relationships while maintaining stability and interpretability. This study proposes a hybrid LASSO-SVQR framework that integrates sparse variable selection with convex nonlinear quantile regression for CoVaR estimation. The proposed approach leverages the sparsity-inducing property of LASSO to identify significant risk drivers and employs Support Vector Quantile Regression (SVQR) to model nonlinear conditional dependencies. Simulation results demonstrate that the method achieves robust performance across heterogeneous data-generating processes, while empirical analysis on financial market data reveals both quantile-specific and stable systemic risk relationships. Backtesting results confirm the reliability of the proposed model. The findings highlight the effectiveness of combining sparsity and nonlinear modeling for interpretable and accurate systemic risk analysis.