Model Risk Management:
Models are used across the whole financial sector. Practically anything we use to assist in decision-making, from a simplest excel file to the complex AI algorithm, is a model. In more formal terms, a model is an algorithm which is based on mathematical, statistical, or economic assumptions, is being used for business purposes and is implemented in an appropriate IT tooling.
Nowadays, financial institutions rely on models more than ever before, using them from pricing of new financial products, calculating capital, to marketing analytics using AI models. Most of the financial institution’s decisions are based on a models’ output.
As any widely used tool, models are error-prone and require management. Famous errors like J.P. Morgan’s unnoticed 2 billion trading loss , or Apple’s delay of entering into financial services due to gender biases in the AI algorithm for credit approvals  are only two examples. Naturally, institutions aim to minimize such errors. In this paper, by using the term “institution”, we also refer to non-financial firms which are using models for their operations. For the purpose of this paper, we would like to cover both financial institutions such as banks and companies in a broader sense of the term, for example technology firms.
In this whitepaper, we describe the model risk management cycle and crucial elements which are needed for successful model management. After reading this whitepaper, you will develop an understanding of the key model development steps and key model risks for these steps, as well as possible practical solutions to address the risks.
To effectively manage models and model risk, we propose that institutions establish clear roles, accountability, and responsibility, as well as process around model management. A Model life-cycle, which we explore in this whitepaper, can be a helpful tool for institutions to implement successful model risk management.
Through the text we use some terms. The process of managing model(s) is called simply – “model management”. The process of managing risks which models pose also has a straight-forward name: “model risk management”.