Skip to content
All apps

Margin of Conservatism (EBA/GL/2017/16)

About this app

Stack and aggregate the add-ons of categories A, B and C interactively — per rating grade.

The idea

Banks that estimate credit risk with internal models under the IRB approach must cover uncertainty in their estimates with a margin of conservatism (MoC). The EBA guidelines on PD and LGD estimation (EBA/GL/2017/16, paragraphs 41–47 according to the app) distinguish three categories: (A) data and methodological deficiencies, (B) relevant changes, for example in underwriting, workout or the definition of default, and (C) general estimation uncertainty. The add-ons must be quantified separately, documented and reviewed regularly; they must not replace fixing structural model weaknesses.

What the app shows

Each bar is a rating grade from AAA to CCC. The base is the long-run average PD (LRA PD), with the three MoC add-ons stacked on top in their own colours. Three sliders set the add-ons relative to the LRA PD (0 to 60 % each). Another slider sets the sample size N per grade: in the app, category C scales with 1/√N, so smaller samples raise the add-on. Aggregation can be “linear additive” or “root of sum of squares”. The effective add-on and the final PD per grade are displayed.

Try it yourself

Reduce N from 2,000 to a few hundred and watch the C share grow in every grade. Then switch between additive and root-based aggregation: the root variant gives a smaller total add-on, but it is only justified if the independence of the components is plausible. The app is a learning model with simplified assumptions and does not replace an institution-specific quantification or a supervisory assessment.

Frequently asked questions

What is a margin of conservatism?
A margin of conservatism is a prudential add-on to estimated risk parameters such as PD or LGD. It accounts for uncertainty in the data, the methods or the estimate itself. Its purpose is to keep capital requirements from being understated.
Which MoC categories does EBA/GL/2017/16 define?
The guidelines distinguish category A for identified data and methodological deficiencies, category B for relevant changes, for example in underwriting standards, recovery processes or the definition of default, and category C for general estimation error. The add-ons are quantified per category and combined into an overall MoC.
Why does estimation error depend on sample size?
The statistical uncertainty of an estimated default rate falls roughly with 1/√N, where N is the number of observations. A rating grade with few obligors therefore has a wider confidence interval and needs a larger add-on for general estimation uncertainty.
What can I adjust in the MoC app?
You set the add-ons for categories A, B and C, the sample size N per grade and the aggregation method. The stacked bars show the LRA PD, the add-ons and the final PD for seven rating grades, and a tooltip gives the values per grade.

Subject: IRBA / banking supervision

More from IRBA / banking supervision

A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps