Journal of Risk Model Validation

Our journal's subject matter continues to evolve in response to current issues. This is clearly a very healthy process and is of obvious benefit to all associated with The Journal of Risk Model Validation.

An instance of this process of evolution is seen in the first paper in the issue: "Economic capital model validation: a comparative study" by Zhenya Hu, Amnon Levy and Jing Zhang. The authors use a long history of public firm defaults to illustrate a validation approach for jointly testing the impact of probability of default and correlation upon economic capital model performance. This extension of analysis to economic capital models seems, to me at least, to be an interesting extension to the journal's subject matter. I hope I do not receive angry letters from hard-line risk model validators.

In our second paper, "Loss given default modeling: a comparative analysis", Olga Yashkir and Yuri Yashkir compare the performance of several frequently used loss given default models: the least-squares method, the Tobit model, the three-tiered Tobit model, the beta regression model, the inflated beta regression model and the censored gamma regression model. This paper contains a great deal of useful advice for risk managers, and academics, who wish to apply these models empirically.

I have worked with Gini coefficients at various points during my career and I thought there was nothing new to know on the subject, but they have enjoyed a rebirth due to their use in early papers on the receiver operating characteristic and related analysis. Marius-Cristian Frunza makes a new contribution to this topic in a paper titled "Computing a standard error for the Gini coefficient: an application to credit risk model validation". The essence of the paper is to use regression techniques to derive analytic formulas for measures associated with the Gini coefficient such as standard error.

Finally, in a paper titled "Credit portfolio models in the presence of forward-looking stress events", Alexander Denev describes a method for improving credit portfolio models based on the Merton model by adding forward looking tails deduced through use of a Bayesian networks technology to the underlying distributions. This is an original contribution to the stress testing literature and indirectly illustrates the great durability and power of the Merton model. It further provides evidence of the importance of having an intuitively plausible model at the heart of more complex model structures. In my experience, this will always dominate pure black box modeling as it usually allows us to see model failure more easily.