Anticipating the downgrade and, eventually, the default of a client is key to controlling a bank normal cost of risk. In addition to the traditional risk framework, some banks currently use a prediction model based on machine learning to address this challenge.

This presentation aims to give an overview of the current state-of-art of the early warning framework.

• The key element is the Data
• Few challenges to overcome
• Model is good and by nature imperfect
• Best practices to implement an early warning framework

Among all models impacted by recent regulations and post-TRIM ‘IRB repair programmes’, LGD models stand out both by the materiality of impacts and by the degree of controversy in some of the requirements. The presentation goes through the main pain points, dwelling not only on capital impacts but also on consequences for the use of models in credit risk management. Finally, Basel IV expectations are discussed.

• Why LGD models have been under fire from supervisors?
• Are TRIM objectives achieved?
• Realised LGD – Is it really realised and what does it actually represent?
• Is conservatism of LGD models excessive?
• What to expect going forward?

The quantitative fundamental method of ratios used to rate banks in a country do not give enough information to be translated into a probability of default, including some qualitative information murk the analysis, so you want to know how much the bank can absorb in terms of net losses before it defaults. You can complement your analysis by using Monte Carlo to perform stress tests.

• How to design a reasonable stress testing for banks of a specific country?
• Choosing probability functions depending on the variables to stress

Your objective is to be able to rate all the banks with or without an external rating for a specific country. Gaining sensibility to the probability of default of any bank by defining the credit profile using Monte Carlo.

How Raiffeisen Russia uses AI for non-retail (corporate) credit decisions.