Modelling operational risk for Basel pillar 2 (internal model for OpRisk VaR)

Question

I am somewhat familiar with OpRisk for pillar 1. As far as I know OpRisk for pillar 1 will be replaced by standard approaches soon. So what is left is proper modelling in pillar 2.

What are best practice and/or regulatory references on how an internal model for OpRisk (with the result of a value-at-risk or similar) for pillar 2 should look like?

From a talk on the web I found the usual topics from a loss distribution approach (LDA):

• Loss Distribution Approach (LDA): frequency/severity, models, heterogeneity
• Modelling large losses: EVT, heavy tailed distributions, single loss approximation
• Estimation: maximum likelihood approach, method of moments; Bayesian, MCMC, EM.
• Combining data and scenario analysis: Bayesian approach, p-boxes, credibility
• Numerical methods for Aggregation of risks: MC, FFT, Panjer recursion
• Capital allocation: Euler allocation, marginal allocations

Would the regulatory expect all of this for pillar 2?

Answer 1

Not a complete answer put perhaps partial help.

1. What alternatives do we have for pillar 2?

Hard to say. There doesn't seem to be any specific alternatives, apart adjusting your old procedures to whatever the institutions you are working with require.

2: Does it have to be a loss distribution approach (LDA) or could it be something between LDA and a standard approach (SA)? Something less challenging as an LDA but more sensitive than an SA?

From the first § of Basel 4: the way ahead:

The BCBS has introduced a single non‑model based method for the calculation of operational risk (OpRisk) capital, the Standardised Approach (SA). This will replace all three existing approaches for OpRisk under Pillar 1: the Basic Indicator Approach (BIA), the (Alternative) Standardised Approach (TSA/ASA) and the Advanced Measurement Approach (AMA). The SA will apply from 1 January 2022.

So, yes, it seem that it will be an updated SA.

Also from here:

The simplicity of current approaches to loss event dependency warrants considerable improvement, and with machine learning techniques coming to the forefront of model construction, investing in enrichment of internal data – from systems to management surveys, employee satisfaction surveys and cultural assessments – may, in time, provide insights into loss drivers and the dependency of losses across an organisation. This area, however, is under development and some way off inclusion in Pillar 2 capital calculation methods.