Architecture of internal models

Architecture of internal models

137 Views
Added in AFIR / ERM

Thanks! Share it with your friends!

URL

You disliked this video. Thanks for the feedback!

Sorry, only registred users can create playlists.
URL

Description:

Speaker: Tigran Kalberer

Many, especially large and international, insurance companies use internal models for risk measurement, e.g. under Solvency II.

Internal models can reflect the risk profile of insurers more adequately than the standard formula and allow to align the metrics used to manage the company and the metrics used to assess compliance with regulatory requirements.

We have observed that the architecture of internal models, i.e. the main building blocks and their interaction, are quite similar while the building blocks themselves show remarkable differences.
The typical architecture is based on global simulations of the risk, factors, which change is simulated and the approximated change in the own funds is derived for each such simulation. More often than not the result is determined for the different risk types (i.e. market risks, credit risks, insurance risks etc.) and then somehow "aggregated".

The main building blocks are the risk factors, i.e. all parameters (like interest rates, real estate prices, credit spreads, morbidity, expenses etc.) which impact the value of the assets and the liabilities, their joint distribution (especially their dependency structure), the approximations used to determine the change in the own funds given a realization of the risk factors and especially the "aggregation" of the results per risk-type, if the internal model is not based on comprehensive global simulations.

The typical approach to model these components and their popularity will be described as well as the Pros and Cons of the main architectures observed.
Astonishingly enough, some popular approaches exhibit severe weaknesses, which we also will discuss, as we will sketch some solutions to address these.

Post your comment

Sign in or sign up to post comments.
Be the first to comment