Journal of Risk Model Validation

Steve Satchell

University of Cambridge

We live in an environment in I am writing this editorial in early September having endured another quarter of extreme volatility and I can comment that the one bright aspect is an increased demand for risk model validation. I hope some of this increased demand will lead to benefits for our readers and contributors as well as benefitting the journal directly.

Recently, I have been asked to explain in detail to nonspecialists what exactly risk model validation entails, and I can report to you my tentative definition. Risk model validation involves the comparison of a risk model’s forecast of risk with the ex post risk experience of the portfolio whose risk is being forecast, while keeping in mind that both are estimates of the true but unknown value of risk. The “true but unknown” concept will be known to the designer of a Monte Carlo experiment or to a deity. Pragmatically, risk validation is much closer to what regulators define it to be and this is the subject of the first paper in this issue: Carlos Branco’s “Addressing the issue of conservatism in probability of default estimates: a validation tool”, which contains research motivated by the recommendations of the Basel Committee. The approach taken in this paper is to recognize the sample-dependent nature of such measures and construct a benchmark for comparative purposes.

The second paper in the issue is by David Maher and is titled “On the use of t copulas for economic capital calculations”. It addresses the issue of which copula should be used in stress testing and economic cost calculations. The paper is very clear and the conclusions are surprising.

The third paper, by Jimmy Skoglund andWei Chen, is titled “On the choice of liquidity horizon for incremental risk charges: are the incentives of banks and regulators aligned?” Given recent news reports, one may be tempted to think that the answer is “clearly not”, but the authors focus on a particular aspect, namely liquidity horizons, and they show that there is more incentive compatibility than one might think. The link with model validation is the fact that such considerations follow from the Basel Committee’s recommendations.

Finally, we have “Value-at-risk forecasts: a comparison analysis of extreme-value versus classical approaches” by Gözde Ünal. Extreme-value theory is a topic that often passes “practical men” by – it is much more an academic topic. It is pleasing, therefore, to me at least, that the author finds evidence of good performance in terms of value-at-risk forecasting for extreme-value based models. In conclusion, the emphasis in this issue is again primarily on credit model risk, and this reflects where most of the interest in model validation lies. As always, I would be delighted to receive papers on other versions of model validation: namely, models of risk for strategies, equities or managers, or novel applications of validation techniques.