Warrington College of Business, University of Florida
Explainable artificial intelligence – an emerging approach to banking decisions – figures among the topics addressed in this issue of The Journal of Risk. Other topics include the modeling of time-varying maxima, the measurement of systemic bank fragility, and the estimation of common risk measures in the context of nonstationary volatility.
In “Explainable artificial intelligence for credit scoring in banking”, the first paper in the issue, Borger Melsom, Christian Bakke Vennerød, Petter E. de Lange, Lars Ole Hjelkrem and Sjur Westgaard use data from a Norwegian bank to show that a Light Gradient Boosting Machine model combined with Shapley additive explanations results in a credit scoring model that is competitive relative to logistic regression in both forecast accuracy and predictor interpretability.
In our second paper, “Modeling maxima with a regime-switching Fréchet model”, Keqi Tan, Yu Chen and Pengzhan Chen address the issue of nonstationary returns by developing a Fréchet model with time-varying shape and scale parameters to estimate corresponding series of maxima. Through an empirical illustration, the authors show that their approach captures the presence of switching regimes and time-varying tail risk while also offering better forecast accuracy than the static generalized extreme value model.
In “Assessing systemic fragility: a probabilistic perspective”, the issue’s third paper, Deyan Radev develops a comprehensive procedure to assess the systemic risk of banks and sovereigns. A systemic fragility measure is constructed based on individual (marginal) probabilities of distress, estimated from credit default swap spreads using a bootstrapping procedure, and the corresponding multivariate probability density is recovered by the cross-entropy approach. Using data focused on the euro area between 2007 and 2011, Radev shows that the banking system in the region had started to deteriorate before the collapse of Lehman Brothers in the fall of 2008, and that after 2009 joint sovereign default risk outpaced the rise of systemic risk.
Concluding this issue is “Semiparametric GARCH models with long memory applied to value-at-risk and expected shortfall” by Sebastian Letmathe, Yuanhua Feng and André Uhde. Here, the authors rely on a class of fractionally integrated log-GARCH models with deterministic trend to estimate widely used risk measures. Based on stock index data from Asia, Europe and the United States, Letmathe et al show that their approach captures the simultaneous modeling of heteroscedasticity and slowly moving unconditional variance well, thus offering a substitute that works better than its conventional, parametric counterparts.
The authors put forward an explainable machine learning model predicting credit default using a real-world data set provided by a Norwegian bank.
The authors identify a regime-switching Fréchet model which can be used to identify the behavior of extreme values in financial series.
Using new measure of systemic fragility, the author ranks euro area banks and sovereigns and according to their systemic risk contribution.
The authors introduce and apply new semiparametric GARCH models with long memory to obtain rolling one-step ahead forecasts for the value-at-risk and expected shortfall (ES) for market risk assets.