The analysis focuses on loans granted by 5 large French banking groups to large corporates operating in France and rated by several banks under the Advanced Internal Rating Based approach (the so-called AIRB approach).

The analysis differs from the existing studies since it is based on a detailed dataset of common counterparties for the five banks. Since the comparison is done on identical counterparties, the differences in RW or in risk parameters are not related to the composition of loan portfolios. This article uses a unique dataset that has been collected by the APCR in 2012 through an ad hoc survey sent to banks regarding a sample of common counterparties among the five banks.

The analysis shows that banks have similar RWA rates (Risk-Weighted Assets/Exposures at Default), except one bank which is more conservative than others. Regarding Probabilities of Default (PDs), the banks exhibit broadly similar levels of average PDs. But, for Loss Given Defaults (LGDs), there is a wider dispersion. The analysis also shows that the dispersion on the RWA rates is mainly due to differences in LGDs more than the other parameters. Part of the dispersion in LGDs may be related to differences across banks in their collateral policy as well as the inclusion of collateral in LGD calculation, and in the effectiveness of the recovery process in case of default. In addition, the regulatory provision to add margins of conservatism to cover the expected range of estimation errors, may also be an explanatory factor of this dispersion, as well as the calculation of the downturn LGD.

If some differences observed in LGD estimates would appear unwarranted, it could be considered to improve harmonization by focusing supervision of internal models on LGDs and by providing more rules for their computation. Therefore, in the debate around the role of the AIRB approach, this article suggests that, instead of replacing this approach, the current framework for large corporates portfolios could rather be adapted to restore confidence in internal models

Download the Analysis and synthesis N° 42

Updated on the 25th of February 2025