Journal of Operational Risk

Risk.net

Optimal B-robust posterior distributions for operational risk

Ivan Luciano Danesi, Fabio Piacenza, Erlis Ruli and Laura Ventura

  • Computationally efficient method for robust Bayesian estimation of AMA-LDA models
  • Bayesian capital estimates insensitive to extremely high or extremely low losses
  • Bayesian estimates which are more stable than maximum likelihood-based counterparts
  • Bayesian estimates which are reasonably efficient

ABSTRACT

The aim of operational risk modeling is to provide a reasonably accurate, reasonably precise and reasonably robust estimation of capital requirements, including a level of sensitivity that is consistent with the changes of the risk profile. A way to obtain robust capital estimates is through optimal B-robust (OBR) methods. Previous research has shown that OBR methods might mitigate the bias in capital risk quantification when compared with classical maximum likelihood estimation. Motivated by requirements related to operational risk measurement, the aim of this work is to integrate prior information into a robust parameter estimation framework via OBRestimating functions. Unfortunately, the evaluation of OBR-estimating functions for different parameter values is cumbersome, and this rules out the use of many pseudo-likelihood methods. To deal with this issue, we suggest resorting to approximate Bayesian computation (ABC) machinery, using the OBR-estimating function as the summary statistic. Unlike other methods, the proposedABC-OBR algorithm requires the evaluation of the OBR-estimating function at a fixed parameter value but using different data samples, which is computationally trivial. The method is illustrated using a small simulation study and applications to two real operational risk data sets.

To continue reading...

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an indvidual account here: