PS 2018
Title | Description | |
---|---|---|
PS18001_Evaluation-of-the-most-pragmatic-approach-to-expected-credit-loss | Project goal: Under IFRS9 the concept of lifetime loss requiring PIT estimates for both stage 2 and stage 3 was introduced. Whilst methodologies are in their infancy the current understanding is that a loss rate or direct approach to ECL is allowed for which is far simpler and possibly more accurate than determining each of the parameters PD, EAD and LGD in isolation. The research problem would then be to evaluate the most pragmatic approach given that accuracy of ECL is the primary objective. Problem statement | |
PS18002_The-use-of-artificial-intelligence-in-risk-measurement-and-decision-making | Project goal: Developing models that use artificial intelligence / machine learning approaches towards risk research. This includes comparative analysis with more established / traditional methodologies. Problem statement | |
PS18003_The-use-of-Global-Credit-Data-(GCD)-data-in-LGD-modelling-in-SA | Project goal: To evaluate the appropriateness/representativeness of pooled credit data, specifically the Global Credit Data (GCD) (formerly known as the Pan-European Credit Data Consortium), in the development of new loss given default (LGD) models for the corporate portfolio and potentially other portfolios. Problem statement | |
PS18004_Machine-learning-and-its-applications-in-predictive-modelling | Project goal: The goal of the project is to propose new methodologies (machine learning) in the predictive modelling context, with a specific focus on: • Factorization• Variable selection using knock-offs• Quantile regression Problem statement | |
PS18005_Investigating-the-transparent-representation-of-machine-learning-applications | Project goal: The goal of the project is to investigate machine learning models for use in retail credit scoring, with a specific focus on: • Transparent representation of machine learning models• Stakeholder insights into so called “black-box” modelling Problem statement |