By Terry Osterweil, Director, Scott Mason, Director, Michael Stock, Director, and Leslie Albergo, Director, of Standard & Poor's Ratings Services.
The dynamics of the U.S. RMBS market are such that continual refinements and innovations to existing technologies, in search of more efficient ways to perform the mortgage lending process, pose an ongoing challenge when attempting to capture this data for risk analysis. One refinement has been the increasing use of automated underwriting (AU) systems and the resulting allowances for the borrower's level of documentation and type of appraisal needed. This phenomenon led Standard & Poor's Ratings Services to increase the amount of required data needed to accurately perform its loan level analysis. That, in addition to updates to current criteria, has resulted in the release of an updated version of Standard & Poor's LEVELS(r), its proprietary mortgage credit risk analysis software.
LEVELS v5.6, which will be utilized by Standard & Poor's for all transactions closing on or after Dec. 1, 2003, reflects extensive research by Standard & Poor's residential mortgage group. This research and analysis has led to: expanded data requirements for mortgage loan analysis; a new Standard & Poor's Housing Volatility Index(tm) (HVI), incorporating new methodology for measuring housing volatility at the metropolitan statistical area (MSA) level; incorporation of a revised loss severity scoring algorithm; and a review of additional Automated Valuation Models (AVMs).
New data input requirements
The new submission requirements, in conjunction with LEVELS v5.6, have led to a revised file format, originally announced Feb. 7, 2003. The revised file format will include new required fields that indicate:
* Whether the borrower's assets were verified as part of the origination process (for purchase loans only);
* The type of appraisal obtained when originating the loan;
* Whether an automated valuation model was used in the appraisal process and, if so, which system was utilized;
* Whether the borrower is self-employed;
* Whether a NextGen FICO score is being supplied; and
* Whether the loan is a high-cost loan or covered loan governed by any anti-predatory lending laws.
Beginning with December 2003 transactions, the inclusion of the above data fields in the loan level file are mandatory for Standard & Poor's to analyze and rate the transaction. In addition, in conjunction with a revised loss severity scoring algorithm (see below) that is part of LEVELS v5.6, the product description field will be used to indicate the type of servicing that is performed on the loan. The loss severity calculated on a loan or pool of loans may change depending upon the product description code selected. While the product description field is not mandatory, failure to code this field correctly could result in an incorrect and/or different answer from that calculated by Standard & Poor's analysts.
Revisions have also been made to some of the codes included in the property type, documentation type, loan type, and primary mortgage insurer fields. These changes included in a new property-type code indicate a manufactured housing property, along with new criteria established to analyze such loans within the LEVELS model.
To facilitate compliance with data and format requirements, Standard & Poor's has developed a Data Quality Control Manual to serve as a reference and training resource. The manual includes Standard & Poor's loan file format, collateral field definitions and codes, and a quality control program that can be used by issuers to preview files before they are submitted for ratings.
Advances in alternative valuation methods
To keep up with the recent record-breaking origination volume, lenders have sought out more efficient ways to perform the mortgage lending process. As a result, a number of new models emerged and were implemented for the RMBS market. One such model was applied to streamline the property evaluation process, a process that began taking hold in 2000 when 10% of all new originations in the residential-mortgage market had an automated valuation attached to them in some way. Since 1998, Standard & Poor's has continued its evaluation of AVMs. "Our ongoing evaluation process is designed to address the credit risk associated with the potential use of this valuation method in the private-label residential mortgage market," said Leslie Albergo, a director in Standard & Poor's Structured Finance group. Standard & Poor's has already seen a significant increase in second-lien product, according to Ms. Albergo, and a slight increase in prime-quality first-lien product over the past year.
Mortgage originators are increasingly using AVMs as a basis for determining the value of real estate used as collateral for mortgage loans. Investors should be aware that any mortgage pool featuring properties assessed by a system that has not been reviewed by Standard & Poor's might feature appraisal variances that exceed Standard & Poor's standards, Ms. Albergo explained.
Any AVM used within a pool of securitized mortgages to be rated by Standard & Poor's must be reviewed and its results incorporated into Standard & Poor's ratings criteria. Standard & Poor's reviews a new electronic home valuation tool solely at the request of the vendor. To begin the testing process, Standard & Poor's provides the vendor with a test portfolio of properties with known sales prices for valuation. The ability of the system's user to manipulate the results is also determined and incorporated in the analysis.
A significant feature of Standard & Poor's testing process is the creation of system-specific acceptance parameters so that the property values accepted from a system are within an acceptable tolerance range. These parameters are then embedded within both LEVELS and DACSS(r) (Documentation and Automated Collateral Scoring System), another Standard & Poor's model used in the industry to determine appropriate income documentation and appraisal requirements. DACSS evaluates the attributes of a loan application and the credit quality of a borrower, and then assigns alternative documentation and collateral assessment requirements in accordance with Standard & Poor's rating criteria, without affecting credit enhancement requirements. The DACSS algorithm is embedded within the LEVELS model to calibrate appropriate loan level credit enhancement.
Results of the most recent review indicate that performance and geographic coverage are converging among AVM models. The findings show that systems can be extremely accurate for some property characteristics or geographic locales, and inaccurate for others. Before testing, hedonic and house price index-based models were most prominent. A current trend is the emergence of hybrid models. Standard & Poor's has created and embedded system-specific criteria in the LEVELS and DACSS models for: Basis100 (PASS); C&S Marketing (AVM Select); Countrywide Home Loans (CAPES); First American Real Estate Solutions (ValuePoint 4); FISERV CSW, Inc. (CASA); FNIS (Value Sure); Freddie Mac (Home Value Explorer); Lender's Service Inc. (LSI Indicator); and MRAC (HPA 2000).
Standard & Poor's recently began accepting vendor requests for its next AVM testing period, currently scheduled for first-quarter 2004. AVM vendors provide values for each property in the database. The results are stratified and measured in numerous permutations, using inputs such as variance (a measure of the actual sale price compared with the estimated value given by the system), sale price, state, county and Zip code.
In addition, each AVM vendor requesting a Standard & Poor's review must submit a technical paper on its system, along with the development and testing of that system. An operational overview of user aspects is also required. The written requirements should include, but are not limited to, a general description; system development; user factors; database management; and planned enhancements. Based on the review results, system-specific acceptance parameters are created and incorporated into Standard & Poor's ratings criteria, which are then embedded into both LEVELS and DACSS.
The quality of data supporting an AVM model determines the quality of its performance. The data, which is the core of any model's results, must be accurate, current and complete. These are ongoing challenges that AVM vendors continually address. Accordingly, Standard & Poor's system review process has become integral to the overall credit risk analysis of residential-mortgage portfolios.
All of the improvements in LEVELS v5.6 are designed to achieve more accurate calculations of a loan's foreclosure frequency, loss severity and loss coverage requirements, thus augmenting the comfort level of investors when they purchase a mortgage-backed security that Standard & Poor's has rated.
New HVI is a pure volatility index
In LEVELS v5.6, Standard & Poor's HVI, using innovative technology, calculates the volatility of the home prices in a given MSA and determines the probability of house-price declines. The output of the HVI is used to adjust assumed Market Value Decline (MVD) for predicted increases and decreases in property values by rating category (see chart 1 for standard MVD assumptions). Most critically, 331 MSAs will now be captured and used in the analysis, compared with 199 MSAs captured in the earlier index. The data in v5.6 is based on Office of Federal Housing Enterprise Oversight's (OFHEO) first-quarter 2003 house price index (HPI).
Keeping pace with a fast-moving market
As the U.S. RMBS market continues to grow in issuance and complexity, the use and precision of mortgage risk assessment models takes on greater importance. In such an environment, refinements and innovations to these models are an ongoing challenge. By regularly updating its mortgage risk assessment models, Standard & Poor's adjusts to the many intricacies that characterize this evolving market.
The above piece was abridged from a larger article by S&P.