Note available here.
This paper provides a comment on the Regulatory Technical Standard (RTS) recently issued by the European Banking Authority’s (EBA) on the use of the Purchased Receivables Approach (PuRA) by European banks. The RTS is designed to clarify an important component of the Basel 3 rules for securitisation capital as applied to banks in Europe. The PuRA would be used by banks to calculate inputs to the SEC-IRBA, the most advanced of the hierarchy of approaches that banks are permitted to employ when calculating regulatory capital for securitisation exposures in the banking book.
To illustrate the implementation of the PuRA, we estimate a set of models based on European Data Warehouse (ED) data for residential mortgages. We estimate models for 8 European countries using loan level data for the pools of 504 Residential Mortgage Backed Security deals. We are able to formulate models at country, originator and deal levels. We examine how data from these different models may be combined to estimate PDs for a single notional deal. We discuss how adjustments may be made for originator underwriting standards and for different Margins of Conservatism. The exercises we perform help us to identify ways in which the EBA’s RTS could be helpfully extended or made more precise.
We conclude that there are three areas in which the RTS could be clarified. First, the RTS state that the data hierarchy for banks implementing the PuRA is the inverse of what would normally hold for an IRB model implementation in that external data on exposures resembling those of the securitisation pool in question should be preferred to the bank’s internal data on loans that it has originated itself. This implies that PuRA modelling should be considered as the development of a methodology that can be applied to multiple datasets in a dynamic way as securitisation deals present themselves. It would be helpful for the EBA to acknowledge this as it represents a significant departure from the usual IRB modelling practice and has multiple important implications.
Second, the EBA has provided little guidance in the crucial area of what data is acceptable. A key question is whether aggregate performance data of the sort contained in securitisation market investor reports and used by ratings agencies in risk analysis of securitisations may serve as the basis for analysis or whether loan level data is essential to obtain supervisory approval. This study employs loan level data but we are aware that ratings agencies often analyse securitisations on the basis of cumulative default curves broken down by vintage.
Third, the EBA’s 2017 guidelines on IRB modelling require banks implementing IRB models to develop Margins of Conservatism and to allow for the impact of underwriting standards. While significant in other IRB exercises, these two requirements are central to PuRA modelling involving as it does multiple data sources used in a varied and dynamic way. Making all aspects of these activities fully statistical and data driven (without judgmental input) appears scarcely feasible. Acknowledging this even while stipulating that any judgments be prudent and evidence-based would be helpful.