Measuring Retail Lifetime PDs
https://ssrn.com/abstract=2857183
The retail lending industry has generated several seminal approaches for measuring and managing credit risk; scoring, roll rate modeling, vintage analysis are some these methods.
The new IFRS 9 framework raises the issue of mesuring credit risk including dynamic aspects of credit risk over the full lifetime of the loans and there is no consensus, neither in the market nor in the literature, on how to achieve this goal.
My new paper "Lifetime PD Analytics for Credit Portfolios: A Survey" fills this gap and compares the different models to compute PD curves on credit portfolios. We survey how to compute lifetime PDs from scores, from risk class migrations, from roll rates, from vintage analysis or from observed defaults directly, and we process a systematic comparison of all these methods.
Additionally, we detail the dynamics generated by defaulted loans that come back in bonis. We bridge the dynamics of cures with survival analysis and we compute explicitly the proportion of long term survivors as a function of the dynamic parameters of the portfolio.
To all of you that have an interest in retail (or wholesale) PD term structure estimation, go to: https://ssrn.com/abstract=2857183
Making AI accessible to All | AI | Risk Analytics | Certified Machine Learning Engineer l Certified Solution Architect-Professional | Certified AI Engineer l Google Cloud l AWS l Microsoft | CDMP l DAMA | PMP | RMP | SAS
8 年Great review Vivien Brunel!!Too tempted to share!! Very curious on markov chain!!
Junior Partner - BSC Consult
8 年Thank you for your wide review, Vivien! But it seems a little... academic. It is concentrated on PD measurement although more interesting applications are forecasting and macroeconomic modelling the link between macrofactors and PDs. If you want to apply this theory to proactive portfolio management, you should use more sophisticated mathematical methods than considered in Breeden's works. Luckily, they are implemented in professional portfolio management systems. The other question is how to align this math with management. Application of this math can increase bank efficiency by 1.5% of loan portfolio amount. But the ways of its actual implementation are complicated (as one Chineese sage put it, perspectives are bright but path is winding). For interested, I can share some materials.
CEO at BSC
8 年Banks recalibrate score from time to time - that's why better to use risk-classes (not grades). My experience showed that the best way for Pd curves (as well as transition matrixes) estimation for different macroeconomic condition is to use RRAS (Roll Rate Analytic System) technology (https://bsc-consult.com/eng/).
Founder at Padus Consulting
8 年Interesting, like markov chain introduction especially for those coming back from default stage. I would have developed the fact that introducing scoring through Survival theory (which is the most relevant to me) is quite complex to implement. However there is technics to adapt it. Thanks for this article, well done!