Aggregation and reporting of model risk

The final stage is reporting to inform the senior management about the model.

An aggregation method could be based on the following criteria:

Data quality
Quality of the model validation
Model use
Documentation quality
Quality of the model development
Ongoing performance monitoring
Legal entity / business units
Trading desks

Table 2 is a case study where the table was presented to the board of a primary financial institution. The table represents a straightforward way to present technical information for senior management not familiar with the model risk topic. The categorisation of the model in the 4 classes is essentially driven by three elements:

The materiality of the model (financial and non- financial impact) and the possibility to remediate a wrong outcome of the model,
The complexity of the model represented in turn by the number of people who knows it and the mathematics/statics embedded in it,
The effective importance in driving the business decision.

Table 2: An example of reporting of model risk

Class 1

Class 2

Class 3

Class 4

High impact on revenues and reputation

Managed by a small number of people

No ability to intervene

Complex statistical / mathematical models

Heavily support to business decision

Medium impact on revenues and reputation

Managed by enough people

Human intervention is permitted in a timely manner

Medium complexity

Moderate support to business decision

Low impact on revenues and reputation

Managed by many people

Mainly simple formulas (e.g. excel)

May support some business decision

No impact on revenues and reputation

Managed by a broad range of people

Outside the area of the key business risks

Algotrading

AER/ APR

Regulatory returns

Structured products pricing

Interest rate model

Evaluation models

Strategic tools

MI reporting for Board and Senior Managers

Planning-tools

MI reporting for mid- management

Project evaluation

Internal reporting not used to drive action

For example, according to the above criteria, it is easy to understand why an algo-trading model is in class 1: high impact on the revenues and reputation with limited possibility to prevent any loss; very few people fully understand how it works given its complexity and it is basically designed to take decisions on its own.

On the other side, an internal reporting which does not support any specific action is collocated in class 4.

Definition of model and maintain a model inventory
Effective governance framework, policies, procedures and controls to manager their model risk
Robust model development and implementation process and ensure appropriate use of models
Appropriate model validation and independent review activities to ensure sound model performance and understanding of model uncertainties.