This book focuses on the alternative techniques and data leveraged for credit risk, describing and analysing the array of methodological approaches for the usage of techniques and/or alternative data for regulatory and managerial rating models. During the last decade the increase in computational capacity, the consolidation of new methodologies to elaborate data and the availability of new information related to individuals and organizations, aided by the widespread usage of internet, set the stage for the development and application of artificial intelligence techniques in enterprises in general and financial institutions in particular. In the banking world, its application is even more relevant, thanks to the use of larger and larger data sets for credit risk modelling. The evaluation of credit risk has largely been based on client data modelling; such techniques (linear regression, logistic regression, decision trees, etc.) and data sets (financial, behavioural, sociologic, geographic, sectoral, etc.) are referred to as “traditional” and have been the de facto standards in the banking industry. The incoming challenge for credit risk managers is now to find ways to leverage the new AI toolbox on new (unconventional) data to enhance the models’ predictive power, without neglecting problems due to results’ interpretability while recognizing ethical dilemmas. Contributors are university researchers, risk managers operating in banks and other financial intermediaries and consultants. The topic is a major one for the financial industry, and this is one of the first works offering relevant case studies alongside practical problems and solutions.
Author(s): Rossella Locatelli, Giovanni Pepe, Fabio Salis
Publisher: Palgrave Macmillan
Year: 2022
Language: English
Pages: 114
City: Cham
About This Book
Executive Summary (English)
Contents
About the Authors
List of Figures
List of Tables
1 Introduction
2 How AI Models Are Built
2.1 Processing of Unstructured Data in AI Models
2.1.1 The Main Structuring Techniques for “Unstructured” Data Are Text Analysis and Natural Language Processing
2.1.2 What Does “Alternative Credit Data” Mean?
2.2 Stand-Alone AI Models
2.2.1 Decision Trees
2.2.2 Random Forests
2.2.3 Gradient Boosting
2.2.4 Neural Networks
2.2.5 Autoencoder, a Special Type of Neural Network
References
3 AI Tools in Credit Risk
3.1 Use of Alternative Techniques and Data in Probability of Default Models
3.1.1 The Type of Data Analysed and How They Are Managed
3.1.2 The Interpretability of Results: An Important Factor
3.1.3 A Practical Case: Risk Discrimination for Borrowers with Seasonal Businesses
3.1.4 A Practical Case: Identification of Counterparty Risk During the COVID-19 Crisis
3.1.5 A Practical Case: Early Warnings
3.1.6 A Practical Case: Advanced Analytics in Loan Approval
3.2 How to Improve Traditional Models Using AI Techniques
3.2.1 A Practical Application: The Two-Step Approach
3.2.2 The Estimation Methodology Adopted
3.3 Applying an AI Model to Asset Management
3.4 Use of AI Models for the Validation/Benchmarking of Traditional Models
3.4.1 ML Techniques for Benchmarking Capital Requirements Models
3.4.2 Initial Applications for Management Purposes
References
4 The Validation of AI Techniques
4.1 Possible Comparison Criteria Between Traditional Models and AI Models
4.1.1 Principle 1: Accuracy
4.1.2 Principle 2: Robustness
4.1.3 Principle 3: Fairness
4.1.4 Principle 4: Efficiency
4.1.5 Principle 5: Explainability
4.2 Interpretability and Stability of the Models’ Outcomes
4.2.1 Main Legislation
4.2.2 Interpretability Methodological Notes
4.2.3 Key Methodologies
4.2.4 Focus Points
4.2.5 Stability Methodological Notes
References
5 Possible Evolutions in AI Models
5.1 The Role of AI Models in the Credit Risk of Tomorrow
5.2 Ethics and Transparency of Results
5.2.1 Privacy
5.2.2 Transparency
5.2.3 Discrimination
5.2.4 Inclusion
References
Appendix
Glossary
Bibliography
Index
