AI For Banks

Day One Morning

AI – Credit Risk – Retail

• Comparison – traditional vs AI credit scoring of, and credit decisioning with respect to, retail customers

• Alternative data sources for retail customers

• AI probability of default (PD) modelling for retail loans

• AI for retail loan pricing and expected credit loss (ECL) provisioning

• AI generation of early warning signals for borrower stress to amend uncommitted credit facilities

• Use of LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations) to explain AI-generated outputs (xAI). Regulatory expectations around explainability

• Case Study 1 – Gradient Boosting Machines vs logistic regression for retail customer PD modelling, risk-based loan pricing, ECL provisioning and credit decisioning. Results of the two methods compared

• Case Study 2 – Random Forests for retail customer credit scoring, fraud detection and early warning signals

• Case Study 3 – Use of LIME to explain why a retail loan application was declined

• Case Study 4 – Neural Networks for retail customer PD modelling

AI – Credit Risk – Corporate

• Challenges in applying AI for corporate loan credit risk modelling

• AI in financial statement analysis – extraction of key credit (and ESG) metrics and generation of credit risk memos

• Case Study 5 – Using OCR (Optical Character Recognition) and computer vision to compile templated financial statements and extract key credit and ESG metrics

• Case Study 6 – Using NLP (Natural Language Processing) to generate credit risk memos

Day One Afternoon

AI – Market Risk

• Case Study 7 – LSTM (Long Short-Term Memory) and GARCH combined with machine learning (ML) to forecast volatility

• Case Study 8 – Deep learning model to forecast asset prices

• Case Study 9 – Using AI to estimated Stressed Expected Shortfall

• Case Study 10 – Market risk stress-testing using: (a) Generative Adversarial Networks (GANs); (b) Variational Autoencoders (VAEs); (c) Reinforcement Learning

AI – Liquidity Risk

• Case Study 11 – Using GANs, VAEs and Reinforcement Learning to generate liquidity stress scenarios

• Case Study 12 – Using LSTM to project stressed cash inflows and outflows and measure liquidity buffer survival horizon