There is wide variance in the internal models that banks use to calculate risk-weighted assets (RWAs), according to a report published by the Basel Committee on Banking Supervision on Friday. The report suggests best practices for reducing this variability.

The report follows up on earlier research from the Basel Committee, which found considerable variation in average RWAs among banks that use internal models to calculate their credit risk regulatory capital requirements.

Among other things, Friday’s report found wide variation in banks’ practices for estimating their default exposure, which can contribute materially to overall RWA variability. It also found that “some of the estimates used in internal credit risk models lack empirical support.” As well, some data relating to one asset class is being used to support estimates for unrelated asset classes.

The research also found a “reasonable relationship” between estimates of default probabilities and actual default rates, based on a sample from 35 major internationally active banks. However, there appears to be a weaker relationship between loss estimates from credit risk models and actual losses. “At an individual bank level, there is much greater variation between parameters estimated using credit risk models and actual outcomes,” the report notes.

The report also sets out best practices for banks to independently validate their models, including the governance and methodology of these models, suggesting that this could help reduce variability.

“Although there may be legitimate reasons for validation practices to differ from bank to bank, the Committee believes there is a set of sound practices for model validation that are both consistent with the Basel framework and, where properly applied, may reduce some of the variation in practice that otherwise might exist,” the report says.

“The adoption of sound model validation practices can be expected to promote more robust [internal] estimates. Therefore, harmonization in the area of model validation could ultimately lead to reductions in practice-based RWA variation,” it adds.