Pimco deploys neural net model in agency bonds

Old models for $7 trillion mortgage market overstate risk in some cases, asset manager says

Pimco is deploying machine learning models to forecast prepayment risk in the $7 trillion market for US agency mortgage-backed securities and says conventional models were overstating the risk.

The new models are up to 20% more accurate in distinguishing between faster and slower prepaying pools of mortgages, the firm says.

“When you go into specific pools of loans, machine learning can capture nuances that we do not even take into account in the traditional models,” says Stefano Risa, head of