Operational Risk Modeling in Financial Services

CHAPTER 9Losses Modelling

Losses modelling can be considered in two ways:

1. Loss Distribution Approach: This consists of modelling the observed distribution of losses, in order to predict the range of potential future losses. This was the most common approach used for calculating the operational risk capital.
2. Loss Regression: This consists of modelling the dependency of losses to other variables, either internal to the bank such as management, or external such as macro-economic conditions. This method is less developed, in particular because of the lack of data, but is relevant for stress testing.

As we discuss in Part II, we believe that the Loss Distribution Approach cannot be considered as a valid risk measurement method, at least not as an internal risk measurement method. Indeed, considering that past losses are representative of risk implicitly means that risk cannot be managed, or, at the very least creates the conditions of a silo approach between risk management and risk measurement.

As people in charge of risk measurement use observed losses, while people in charge of risk management rely on risk and control self-assessment, there is no surprise that the communication between the two teams is limited, if it does exist.

In the Federal Reserve Bulletin of June 30, 2014¹, a very clear statement of the role of the four elements of AMA is made, which, in principle does not gives full precedence to data:

The AMA does not prescribe any specific approach for operational ...