Journal of Risk Model Validation

This issue contains the usual four papers along with an addendum to a paper that was previously published in The Journal of Risk Model Validation. This addendum has been submitted by the previous paper’s two authors with the help of a third. As an editor, I very much like receiving addenda and general correspondence about what we have published in the past. Apart from reassuring me that people actually read our papers, it is a sign of good scholarship and creates a community around the journal.

The issue’s first paper, “Asset correlations and procyclical impact” by Kung-Cheng Ho, Jiun-Lin Chen and Shih-Cheng Lee, is an examination of the behavior of asset correlations for companies inTaiwan under Basel’s asymptotic single-risk-factor (ASRF) approach. Using Merton’s model to estimate firm default probability from 1990 to 2013, the authors find that assets are positively correlated with firm size and negatively correlated with default probability. They also confirm the result that asset correlations are asymmetric and have a procyclical impact on the real economy after these effects are controlled for. This suggests that constant correlation modeling can be improved on. The challenge is to find a robust specification that allows for easy estimation and captures this phenomenon.

Jimmy Skoglund and Wei Chen are the authors of our second paper: “Rating momentum in the macroeconomic stress testing and scenario analysis of credit risk”. This looks at the addition of rating momentum to the popular factor model of credit risk. The conceptual feature that causes this augmented structure to differ from the usual setup is its non-Markovian nature. The authors find that models that take the stylized fact of rating momentum into account can accelerate the loss timing significantly when compared with a non-rating-momentum case. The exact effect depends on the scenario time horizon, the severity and the portfolio quality. In general, it takes more time for differences to be realized in good-quality portfolios; the effect on lower-quality portfolios, however, can be almost immediate, with significant loss underestimation. This new structure will also have a different explained versus idiosyncratic decomposition, with changes in exposure to factor risk as a result. This is an issue that regulators need to take into account.

“A model combination approach to developing robust models for credit risk stress testing: an application to a stressed economy” is the third paper in this issue of The Journal of Risk Model Validation. Here, Georgios Papadopoulos investigates the impact of macroeconomic shocks on bank-specific risk factors. The author’s concern is model frailty, and his proposed solution is to use a model combination approach to develop robust macrofinancial models for credit risk stress testing. The empirical part of his research utilizes data from the Greek economy, which experienced a sharp change from normal to distressed conditions; this makes it a particularly challenging case to forecast. The paper constitutes an interesting approach to this topic, and the highly visible problems of the Greek banking sector give it contemporary relevance.

Our fourth paper, “Point-in-time probability of default term structure models for multiperiod scenario loss projection” by Bill Huajian Yang, involves an important technical problem, which will be easier for me to quote than paraphrase:

Rating transition models have been widely used for multiperiod scenario loss projection for Comprehensive Capital Analysis and Review (CCAR) stress testing and International Financial Reporting Standard 9 (IFRS 9) expected credit loss estimation. Although the cumulative probability of default (PD) for a rating can be derived by repeatedly applying the migration matrix at each forward scenario sequentially, divergence between the predicted and realized cumulative default rates can be significant, particularly when the predicting horizon is extended.

The author offers a solution to this problem by directly modeling the forward PDs, validating them using a corporate portfolio. He reports greatly improved results relative to more conventional methods.

The issue concludes with the aforementioned addendum by Torsten Pyttlik, Mark Rubtsov and Alexander Petrov, which refers back to a 2016 paper titled “A point-in-time–through-the-cycle approach to rating assignment and probability of default calibration” by Mark Rubtsov and Alexander Petrov (The Journal of Risk Model Validation 10(2), 83–112). In the addendum the authors offer an analytical solution to a problem that two of the authors had previously solved numerically.

Steve Satchell
Trinity College, University of Cambridge