The net present value (NPV) modeling component of the Home Affordable Modification Program (HAMP) – that is, the test that determines whether or not a loan modification is favorable to a foreclosure, from an investor's standpoint – requires consistent use across servicing operations.
That was the message delivered last week by Dane D'Alessandro, director of program management for Freddie Mac's Making Home Affordable-Compliance (MHA-C) division. While speaking at the Mortgage Bankers Association's National Mortgage Servicing Conference & Expo last week, D'Alessandro outlined some of the most common errors his group sees when reviewing servicers' modeling practices. He also delved into the Treasury Department's parameters governing servicers' customization of NPV models.
A servicer's decision to modify a loan under HAMP is largely shaped by an NPV output. Therefore, "there's a need to make sure the usage and also the management of that NPV model is correct," D'Alessandro said, adding that there is an "overriding desire for consistency" in the way the model is used.
Servicers have three basic options for how to use HAMP's NPV model: the Treasury Department's Web portal, a recoded version embedded into a loss mitigation system, or a proprietary model.
The Treasury portal model is the most widely used option, as it requires servicers to plug data into the portal and await an output.
An offshoot of that model allows servicers to essentially recode the Treasury's model and integrate the recoded version into their loss mitigation systems. More than a dozen servicers have chosen this route, D'Alessandro said, and some of the shops are using the embedded NPV model to pre-screen borrowers for HAMP while they're on the phone, resulting in efficiency gains.
The recoded model is a "carbon copy" of the Treasury's version, D'Alessandro said, complete with identical cashflow algorithms. While servicers are allowed to use these recoded versions, they must first gain Treasury approval through a certification process. Under this process, MHA-C will run a random sample of loans through both the servicer's recoded version and the Treasury portal. If the NPV outputs between the models are within a certain threshold, the recoded version is approved.
"We have to be sure that your model provides the same answers for a wide range of scenarios so that we know it will be consistent," D'Alessandro said.
Recoded models – regardless of if they are to be used in an internally developed loss mitigation system or commercial loss mit software – must pass certification. The Treasury currently does not permit the direct certification of vendors. Instead, servicers, as direct participants in HAMP, must request certification of the commercial product they use. The Treasury is actively discussing direct certification of software providers, D'Alessandro added.
"Honestly, I think Treasury wants a little more experience with the third-party providers and how that works, and maybe they'll change their mind after that," he explained. "It's only been three or four months during which we've had software providers going through the certification process."
The third NPV option available to servicers is reserved for large operations. Shops with servicing books larger than $40 billion are allowed to use proprietary models, whose customization begins and ends with the default rate employed in a modification scenario. MHA-C performs additional diligence in regard to proprietary models, D'Alessandro said.
"We want to know it's a better predictor of loss than the Treasury model – you have to make an argument that there's a valid rationale for it," he commented.
D'Alessandro additionally highlighted the point within NPV modeling when servicers most often err: during reevaluations. If a servicer reevaluates a loan using the NPV model based on a borrower's verified income documentation, the servicer is supposed to keep all values except for income and income-driven variables the same as they were during the original calculation.
Many loss mitigation systems will automatically update file information, including a borrower's FICO score or the mark-to-market loan-to-value ratio (LTV), D'Alessandro noted. Despite the natural tendency to want to use only the most recent data, servicers must keep in mind that, for example, a borrower's FICO score or an LTV will change over the course of a three- to five-month-long trial period.
"Those values by themselves can flip a positive decision to a negative decision, discounting any changes in income," D'Alessandro said. "The key principle with reruns is you're not trying to create a new NPV decision with verified information. You're simply trying to verify the original NPV decision, just with better income information."