Banks’ Risk Models Understated Loan Defaults, ECB Paper Shows

(Bloomberg) -- A system for measuring the riskiness of bank loans that relies on lenders' own estimates may significantly underestimate the likelihood of default, according to a research paper from the European Central Bank.

After the so-called internal ratings-based approach to risk measurement began to be implemented in German banks after 2007, lenders experienced higher default rates than portfolios assessed using a standard “flat-rate” practice, according to the paper written by Markus Behn, Rainer Haselmann and Vikrant Vig. The IRB models under-predicted actual defaults by 0.5 to 1 percentage point, the researchers wrote in the paper published on Monday.

The paper adds to the debate among global regulators over the best way to account for risk in banks' books, as they search for a balance between the accuracy and risk-sensitivity of assessment tools. At the same time, lenders are pushing back against plans by the Basel Committee on Banking Supervision to curb the use of internal models, arguing tighter rules will hamper their ability to lend to the economy.

“Banks that opted for the introduction of the model-based approach experienced a reduction in capital charges and consequently increased their lending by about 9 percent relative to banks that remained under the traditional approach,” the paper states. “However, a regulation that is based on banks' internal risk models may suffer from both informational and incentive problems.”

The authors also found that even though the use of internal modeling allowed banks to enjoy lower capital charges set by supervisors, on account of supposedly better risk measurement, the same institutions actually charged customers more for credit.

This “suggests that banks were aware of the inherent riskiness of these loan portfolios, even though reported default-probabilities and risk weights did not reflect this,” according to the paper.

The Frankfurt-based ECB's supervisory arm, led by Daniele Nouy, is conducting a multi-year review of the approximately 7,000 risk models in use in the euro area. Nouy has said there's too much diversity in the outcomes of modeled risk assessments.

To contact the reporter on this story: Jeff Black in Frankfurt at jblack25@bloomberg.net. To contact the editors responsible for this story: Paul Gordon at pgordon6@bloomberg.net, Patrick Henry, John Glover