Model Wars Episode One: The Prepayment Menace


While market players share differing views on the importance of Bear Stearns' new loan-level non-agency prepayment model - dubbed by some as a "supermodel" capable of revolutionizing the industry - even the most skeptical of MBS pros agree that it raises the bar for prepayment analytics.

One trader at a competing firm said that while he didn't buy into "the notion that this is going to change the way people do business," he could envision Bear's model "moving people in the direction of more technologically advanced loan-level models."

"It's good that they did it," he added, "but I'm really torn on the issue."

What makes the model interesting is that it assesses interest-rate risk on a loan-level basis, and is implemented entirely at the property level despite enormous computational requirements. According to some market sources, the current model surpasses previous aggregate-loan and loan-level models both technically and intuitively.

"I think that the direction Bear Stearns is going is the correct direction for prepayment models," said Andrew Davidson, a longtime MBS veteran and president of Andrew Davidson & Co., Inc., a company that privately creates and analyzes prepayment models.

"Loan-level data is increasingly important in both prepayment and default analysis," he said. "Though I don't think its that different than some models that came before it, this is another major evolutionary step, and all the Wall Street firms have to keep up with each other."

Indeed, as Bear actively marketed its product on the road recently after unveiling it to wide acclaim in April, observers questioned the degree to which the model will drive competing companies to follow suit with the release of similar data-driven tools in the near future.

Though several sources believe that other firms will establish their own models over the course of the next year, some say that on a cost-benefit basis, it's not worth putting all the effort into the creation of such a model.

"Prepayment modeling is as much art as science," said a researcher at one of Bear Stearns' major competitors. "There is so much noise in prepayment modeling that it produces a false sense of security. It makes you think you've got it nailed, when actually that is very hard to do." The researcher added that her firm is not going to follow suit.

Non-Agency vs. Agency Models

According to Dale Westhoff, a senior managing director at Bear Stearns, the new model gives investors a level of precision and accuracy that investors were not able to attain ever before.

"Right now, we view this as a transition point between the agency and non-agency sector," said Westhoff. "Not having basic loan-level information in the agency sector puts evaluating the prepayment risks there at a disadvantage. Not having the LTV data means we don't know how much equity borrowers have in their property. We have that information in the non-agency sector. I think there are going to be models coming out that will leverage on that better information in the non-agency sector."

Sources say that the key to the model is the use of home price indices broken down by zip code, revealing a detailed analysis of the changing monthly LTVs at each home. In fact, several sources note that Wall Street researchers are in the midst of breaking down the information even further, beyond the zip code level, creating a model with an unprecedented level of sophistication.

"With the type of technology that Bear introduced, we'll be talking about the 12 homes on the left or right hand side of a short street by the end of the year," predicted one portfolio manager. "The data is there now."

Indeed, most market players agree that the key to the model is the data involved. Updated LTV ratios from companies such as Mortgage Risk Assessment Corp. add to the value of this type of modeling, sources say.

"If these models are going to drive our market - and they will - then why build a prepayment model with limited data when you have this new universe of data that wasn't available just a few years ago?" said Chuck Ramsey, CEO of MRAC and a former Bear Stearns general partner.

Doubts Persist

However, the fact that this new model is so data-dependent in the first place leads some players to doubt its validity.

According to one MBS veteran, recent Street research has concluded that effective durations haven't been a good predictor of price behavior, so if the new models are based on effective durations, as many of them are, "How reliable can the models be?"