This special issue of The Journal of Credit Risk on early warning models is dedicated to work presented at the 2018 Interagency Early Warning Model Workshop, which featured papers on many different types of early warning systems (EWSs). The workshop was hosted by the Federal Reserve Bank of Philadelphia and co-organized by the Federal Reserve Bank of Dallas and the Federal Deposit Insurance Corporation. It served as a forum for members of the regulatory community to share their views on the development and implementation of supervisory models, to discuss academic contributions to the literature, and to outline directions for future work in the early warning space. While the forum was motivated by the Great Recession and its aftermath, the EWSs presented here are also relevant for the ongoing Covid-19 crisis.
As many researchers have highlighted, a rise in household debt is associated with ﬁnancial crises. Just recently, the rapid increase in US household debt played a major role in the subprime mortgage crisis of 2007–9. It is therefore important to assess the risk that household debt poses to the economy. Neil Bhutta, Jesse Bricker, Lisa Dettling, Jimmy Kelliher and Steven Laufer tackle this important and timely issue in “Stress testing household debt”, our ﬁrst paper. They develop a household stress test, similar to those the Federal Reserve and regulators use during ﬁnancial crises, to stress test household balance sheets. The authors develop a county-level model of household delinquency and use it to conduct “stress tests” of household debt, using house price and unemployment rate shocks from Comprehensive Capital Analysis Review stress tests. They ﬁnd that household debt currently poses a lower risk to ﬁnancial stability than it did before the ﬁnancial crisis, due to the improvement in consumers’ credit quality. The stress tests for household debt developed in the paper are a useful starting point that can be ﬁne-tuned to provide a valuable tool for regulators. Moreover, while the number of consumer defaults has not increased dramatically during the Covid-19 pandemic (due in part, perhaps, to the ﬁscal and monetary responses), it is possible that this number will start increasing if the economy does not improve. The stress tests developed in this paper for household debt may therefore prove useful for regulatory agencies in the current pandemic-driven crisis as well as during future crises.
The issue’s second paper, “The impact of data aggregation and risk attributes on stress testing models of mortgage default” by Feng Li and Yan Zhang, investigates whether a variety of different stress models based on different degrees of data aggregation or the incorporation of information on risk attributes have any bearing on predicting mortgage defaults. They ﬁnd that it is not always the most granular type of analysis (ie, at the loan level) that leads to better predictive power for projecting residential mortgage loan defaults. They do, however, conﬁrm that including risk attributes, whether at the loan level or at the borrower level, greatly improves the accuracy of models regardless of the data aggregation level. This paper complements our ﬁrst paper by shedding light on the various considerations that go into designing accurate stress-testing models for speciﬁc portfolios, with potential applications for modeling strains in different types of loans, such as credit card loans or business loans. Being able to create accurate stress test models for different loan portfolios may prove especially important during the current pandemic, as well as when vulnerabilities materially differ between businesses and households.
Finally, we include in this special issue of The Journal of Credit Risk a paper by Christopher C.Henderson, Shaohui Jia and Charles Mattioli: “Supervisory bank risk early warning modeling: an examiner’s ﬁrst line of defense”. The paper’s ultimate goal is to introduce a more data-driven and efﬁcient EWS to detect changes in the safety and soundness of small commercial banks. This paper focuses on modeling downgrades in CAMELS ratings, which are supervisory assessments of commercial banks. Its main result, through a comparison of different types of models, is that robust forward-looking statistical models are superior to backward-looking assessments of supervisory compliance. EWSs, as broadly applied to many banks, help detect emerging risks and ﬁnancial stability-related issues. At the same time, powerful EWSs allow for a more efﬁcient allocation of supervisory resources, especially in times of unusual challenges to the entire banking system. Indeed, we are arguably in one of those periods at the moment, and all the papers in this special issue will hopefully contribute to making sense of the fast-moving developments in the current banking system due to Covid-19.
The authors estimate a county-level model of household delinquency and use it to conduct “stress tests” of household debt.
The results of this paper show that robust forward-looking statistical models are superior to backward-looking assessments of supervisory compliance, which could lead to less regulatory burden when integrated into the examination process, particularly at…
In this paper, the authors investigate how data aggregation and risk attributes affect the development and performance of stress testing models by studying residential mortgage loan defaults.