|
|
A number of UK insurers have introduced or are planning to introduce separate risk calibrations for a broad range of non-standard credit assets, such as asset-backed securities, according to research by Towers Watson.
Residential mortgage-backed securities, commercial mortgage-backed securities, infrastructure debt and covered bonds are the asset classes that insurers most commonly modelled separately, the survey of insurers found. Some insurers were also modelling collateralised debt obligations and secured loans.
Tim Wilkins, a senior consultant at Towers Watson, in Reigate, who led the survey, says the results are consistent with the trend to diversify asset portfolios as a result of the continued low interest rate environment.
The survey identified credit risk modelling as a key area of focus and change for insurers, but it also underlined its challenges. Credit risk is difficult to model and there are a number of ways of doing it, with little consensus on the best approach.
Some firms might want to review their choice of model, particularly if a more basic approach is used, suggests Wilkins. "It is worth bearing in mind that greater complexity and granularity are not ends in themselves. The choice of model should be linked to the nature of the portfolio and the intended use of the model", he says.
A growing trend towards greater granularity in credit risk models was revealed by the study, with most firms now allowing for firm-dependent risk distributions and more models distinguishing between different sectors of bond issuer.
Financial-issuer bonds were particularly badly hit during the credit crunch and continue to exhibit higher spreads than non-financial issuer bonds of equivalent term and rating, says Wilkins, suggesting their risk profile is very different from non-financial bonds.
A few insurers were implementing or considering implementing portfolio models that would potentially allow an even higher degree of granularity. Such models require large amounts of data. "Calibrating and validating such models is likely to be highly challenging", says Wilkins.
Full article (Risk.net subscription required)