Score

To improve the returns of trading or investing in mortgage assets, FIS Applied Analytics has developed a prepayment score. This score acts as a modifier to our standard model that enables a better understanding of prepayment propensity at the loan level.

Figure 1 depicts historical single monthly mortality (SMM) of a pool of mortgages originated in 1999. SMM is a measure of the reduction in principal experienced by a mortgage pool in any given month. All of these loans have nearly the same coupon of 7.5% and were priced with the same prepayment expectations in August of 1999. 

Figure 1. Historical prepayment of a pool of mortgages (SMM)

Figure 2 charts the prepayment forecast that FIS Applied Analytics' model provided for this pool compared to the actual prepayment; given knowledge of future interest rates. 

Figure 2. FIS Applied Analytics model performance vs. historical actual

Model performance is seen to be excellent. Model users have a forecast of prepayment performance for the pool of loans in any future rate environment.

However, not all loans in the above pool prepaid exactly as the pool prepaid.  Individual loans and smaller groups of loans may prepay more quickly or slowly than the average. Therefore, they will have more or less value than the average, and require different pricing, funding and hedging strategies than the average.

Figure 3 depicts the historical prepayment of three different subsets of the original pool.

Figure 3. Historical prepayment of pool subsets

We can see that a subset of a pool of loans always prepays differently than other subsets of the pool. The opportunity for better than average performance lies in discovering these sub-pools and understanding how they will perform.

The FIS Applied Analytics Prepayment score

To realize the opportunity associated with understanding prepayments at the loan level FIS Applied Analytics has created a modifier to its standard model. The modifier takes the form of a score that can be attached to individual loan records and MBS.

The score was created using loan level data and prepayment histories obtained from various lenders and servicing data aggregators.  The score considers the effects of many variables on a borrower's relative likelihood to prepay. The variables may include the original loan amount, the loan-to-value ratio, the geographic location and the purpose of the loan and borrower credit.  The score expresses the combined effect of all of these variables on the propensity to prepay.

It is important to note that all the loan indicatives used to compute the score are defined at the origination of the loan.

Figure 4. Score Creation

What is the FIS Applied Analytics Prepayment Score

In the simplest sense, the prepayment "score" is a modifier to a standard prepayment model.  Given a current interest rate environment, the standard prepayment model provides a predicted prepayment speed for a loan based on the following characteristics: the loan type (Conforming, Jumbo, etc), the loan origination date, the current age of the loan and the current coupon of the loan.

The meaning of a score is:  in a given "bucket" of loan coupon and loan age, a loan with a higher score is more likely to refinance than a loan with a lower score. 

Obtaining a score

There are several ways to obtain a score:

An FIS Applied Analytics ftp site has been established that accepts files containing thousands of loans and returns scores within an hour of submission.

Via McDash Analytics. McDash (www.mcdash.com) is a servicing data aggregator with a total database of 14 million existing mortgages. Under the terms of our partnership with McDash, FIS Applied Analytics has scored all of the loans in the McDash database. Servicers wishing to obtain scores may do so simply by contacting McDash and asking for scores to be included with the file that they receive from McDash each month. Currently, there is no charge for this service.

In the near future, Settlement Service Vendors will distribute FIS Applied Analytics prepayment scores via loan origination software to the point of sale, in much the same way that FICO scores are distributed. Contact Kevin Williams for further information.

Interpreting a score

The FIS Applied Analytics prepayment score can be understood in two ways:

Via our prepayment model
FIS Applied Analytics prepayment licensees can include our score in their analytic systems. The score will then be called by the analytic system and the score will modify FIS Applied Analytics prepayment vectors used.

Via lookup tables on the FIS Applied Analytics web site
FIS Applied Analytics has created tables that provide relative value, duration and short-term prepayment projections, all as a function of the FIS Applied Analytics prepayment score. These tables are available at the FIS Applied Analytics web site. Users can indicate any coupon, WAC and loan type that is traded in the market then reference the score and view results.

Validating a score

As discussed, the score is a modifier to our standard prepayment model. Recall that the meaning of the score is that for a given coupon, a given age and a given interest rate environment, a loan with a higher score is more likely to refinance than a loan with a lower score.  As evidenced in the chart below, higher score loans will prepay faster, averaged over all coupons and all ages

In this chart we have averaged over everything except score for this portfolio of 300,000 loans.  The red line shows all loans in our lowest score bucket, the yellow loan shows all loans in our highest score bucket.  As predicted, the high score loans prepay faster!

Figure 5. Average prepayments for low to high scores

If we use the score to correct our standard model, the match between the actual and predicted prepayment speeds should improve dramatically.

Below, we show the same chart.  This time, we also show the predicted prepayment speeds for our standard model (dashed lines).

Figure 5a. Average prepayments for low to high scores including standard model predictionsThe final task is to use our calculated score for each loan to improve the model prediction.  This chart shows dramatic improvement in the match of the dashed lines to the solid lines!

Figure 5b. Average prepayments for low to high scores using score calculations and model predictions

A prepayment score will add value if it is able to forecast differentiated behavior in a group of loans that the market would otherwise view as similar. The question that a score needs to answer is which one of these loans, with the same coupon and origination date are the most likely to prepay first?  Since prepayments are driven largely by the difference between a mortgage coupon and the current mortgage rate, a score that tells you that a 9% mortgage is more likely to prepay than a 7% mortgage in a 6% rate environment is not worth much. The score must differentiate between instruments that the market would otherwise view as similar or the same.

Figure 6 presents the actual prepayment experience of 9,015 loans that were placed into five groups of equal size.  Each loan's FIS Applied Analytics prepayment score determined the loan's groupings. The score is based on data available at the date of origination.  The groups are labeled from least likely to most likely to prepay in any future interest rate environment. All of the loans were FN 30, had coupons between 7.5% and 8% and were originated in the second half of 1999.  If the score works then more prepayments should come from the most likely bucket and less from the least likely bucket.  That's just what we observed.

Figure 6. Actual prepayment performance by prepayment score bucket

751 of the loans in the "least likely" bucket were still on the books at the end of March of 2003, only 167 loans in the "most likely" bucket did not prepay.  Clearly, the score helped differentiate by that measure.

Additionally, as the score works it should be predictive over time. A loan with a score in the most likely bucket should always have a higher propensity to prepay regardless of interest rates.  Figure 7 demonstrates exactly this. Here we chart the actual prepayment speed measured by SMM for three of the buckets above.  For clarity we have left out the "less" and "more" likely buckets.  The top line in figure 7 is the actual prepayment speed over time for the loans in the "most likely" bucket. This bucket is always faster than the other two; thus the "least likely" bucket always prepays more slowly.

The implications for this are obvious. For example, a mortgage servicer that would like to retain borrowers that might refinance could prioritize calling efforts based upon each loan's relative propensity to prepay.

Figure 7. Actual prepayments over time by score bucket FN 30, 7.5%, August 1999

 

 

 

Copyright © 2000-2005. Fidelity Information Services, Inc. All Rights Reserved. Privacy Statement.

» Site Support «