Credit risk prediction identifies default risk by establishing a nonlinear relationship between feature data and default status. The feature system is the crucial factor affecting the accuracy of credit risk prediction. This paper addresses how to identify customers within a certain range of cost budgets. A case study on farmers in China shows that the characteristics of incomplete loan information, dispersed residences and strong liquidity indicate that bank lending should consider not only the identification of default customers but also the actual cost budget. A modified support vector machine (CCSVM), which considers the cost of obtaining credit information and the budgets of decision makers, is established to optimize the credit evaluation results. This paper performs feature selection and weighting by optimizing the minimum total misjudgment cost as the objective function while constraining the prediction deviation, feature weights and cost budget. The empirical results for Chinese farmers as customers of a large commercial bank demonstrate that the proposed C+SVM model is effective in selecting credit information costs and that decision makers can achieve the necessary prediction accuracy at lower costs. The best features and weights for credit risk prediction can be obtained for different cost budgets.
This study evaluates the robustness and performance of graph-based models for credit default prediction. Two inductive graph neural network (GNN) architectures-GraphSAGE and the graph attention network (GAT)-are implemented within a framework that integrates automated hyperparameter optimization, imbalance-aware loss functions and adversarial stress testing. Borrowers are represented as nodes in a k-nearest neighbor graph constructed from financial and demographic features. Model tuning is performed via Optuna, while robustness is examined under the fast gradient sign method and projected gradient descent perturbations, with adversarial training enhancing stability. Experimental results demonstrate that optimized and adversarially trained GNNs outperform classical baselines such as logistic regression, random forest and gradient boosting in area under the curve and F1 metrics, while maintaining resilience under feature perturbations. These findings highlight the importance of robustness evaluation as part of the broader model assessment process for modern credit risk modeling.
The escalation of trade tensions and tariff shocks has created persistent uncertainty for financial institutions by directly impacting the integrity of traditional bank risk models. Conventional frameworks often fail to capture the nonlinear and sector-specific effects of tariffs on credit, market and operational risk, which results in misestimated portfolio losses and distorted capital allocation. This paper develops a comprehensive methodology, focusing on the validation and enhancement of risk models specifically under trade war conditions. It introduces three key dimensions: embedding explicit tariff shocks into stress scenarios; applying granular sector-level calibration; and adopting adaptive governance with artificial-intelligence-driven recalibration and overlays. Using local projection methods with robust alternative specifications and controls, the study quantifies how tariffs are transmitted through inflation, interest rates and gross domestic product. It then maps these effects into core model parameters such as the probability of default, loss given default, preprovision net revenue and market risk metrics. Empirical tests, including the Diebold-Mariano test, demonstrate that tariff-augmented models significantly outperform traditional specifications in out-of-sample forecasting. A case study of a US regional bank further illustrates the practical application of this framework in recalibrating Current Expected Credit Losses and Comprehensive Capital Analysis and Review models. By integrating empirical stress testing with robustness checks, artificial intelligence tools and conservative governance as well as a reproducible calibration protocol, the framework strengthens resilience, supports regulatory compliance and ensures forward-looking model risk management under sustained geopolitical uncertainty.
This paper explores the critical importance of the interest rate risk in the banking book (IRRBB) regulations set forth by the Basel Committee on Banking Supervision. It emphasizes the need for financial institutions to develop robust models for forecasting demand deposit balances while adhering to regulatory guidelines. To ensure these models meet the stringent requirements of risk model validation, we implement comprehensive validation techniques that assess predictive accuracy and robustness. Using a random sample of demand deposit balances in the Mexican banking sector, our research contrasts the forecasting results to robustly and efficiently construct survival rates and attrition curves (harvests) through traditional time-series methods and the use of artificial intelligence algorithms. While the Glosten-Jagannathan-Runkle generalized autoregressive conditional heteroscedasticity models showed the best goodness-of-fit results, time-series models lack the flexibility and robustness needed to capture the complexities of demand deposit balance behavior, especially in negative interest rate scenarios. In contrast, the use of machine learning techniques significantly improves predictive accuracy. The results demonstrate that our long artificial neural network-long short-term memory and random forest models improve the prediction results for demand deposit balances, allowing us to efficiently construct attrition curves and survival rates, providing better risk management tools for financial institutions facing the challenges posed by the IRRBB.
The Fundamental Review of the Trading Book encourages financial institutions to shift from the value-at-risk (VaR) to the expected shortfall (ES) risk measure when measuring market risk capital. This paper examines the application of exceedancebased validation (or "backtesting") methods, commonly used for VaR model validation, to the validation of ES models. The examined approach includes finding the quantile value corresponding to the ES for four different estimation methods: an analytical delta-normal approach, an analytical generalized Pareto distribution-based approach, numerical Monte Carlo simulation and nonparametric historical simulation. The paper also investigates the stability of this quantile and proposes an adjustment to the traditional backtesting approaches that helps to accommodate an unstable quantile. The application of the approach is illustrated using real-world Baltic equity index (OMXBBGI) returns data for three different confidence levels: 90%, 95% and 97.5%. Our findings show that the direct application of exceedance-based methods to the validation of ES models is possible, even in the case of an unstable ES quantile.
