AD&Co. Update
A
New & Improved HEL Model
by Sanjeeban Chatterjee
AD&Co. is developing a new model for sub-prime loans to replace the existing model. In this month's article, we describe the data we are using to develop the model, as well as some of the descriptive statistics obtained from the data. The model factors and the actual model results will be published in a subsequent paper.
The data we are using to develop the model has been obtained from the Intex database for Home Equity Loans (HEL) and Asset-Backed Securities (ABS). These loans represent first lien fixed and adjustable-rate mortgages.
We plan to develop prepayment models that reflect the performance for the top 10 HEL issuers. While the general features of the model will be the same, there may be some differences by issuer.
AD&Co's existing prepayment model for HELs was developed with data available up until the end of 2000. However, the structure of the HEL market has changed substantially since then, and the profile of borrowers who take these loans has also been evolving. Hence, we decided to develop a new model for HELs that would incorporate the new trends we have observed in the data.
A benefit of using data from Intex is that the model can be easily integrated through systems that use Intex to obtain the cash flows. To avoid compatibility issues, the model will use fields that are available through Intex.
Description of Data
We obtained data from the Intex database for Home Equity Loans (HEL) and Asset-Backed Securities (ABS). In particular, we used all loans that met the following criteria:
AssetBack_Type = 0 (Normal
Whole Loans)
AssetBack_Type = 3 (Home Equity Loans)
As of December 2003, the top 10 HEL issuers were as follows:
|
Rank
|
Deal
Name ID
|
Issuer
|
Current
Balance
|
Original
Balance
|
|
1
|
RFC
|
Residential Funding Mortgage Securities II
|
71,494,232,159
|
242,072,452,603
|
|
2
|
CWF
|
Countrywide
|
53,251,532,489
|
121,935,129,123
|
|
3
|
RES
|
RESI
Finance LP 2003-CBI/RESI Finance DE Corp 2003-CBI
|
48,353,376,295
|
77,933,014,483
|
|
4
|
WMS
|
Washington
Mutual Mortgage Loan Trust
|
48,318,167,783
|
117,637,520,395
|
|
5
|
SAS
|
Structured
Asset Securities Corp.
|
31,736,689,846
|
79,690,769,680
|
|
6
|
CSF
|
CS
First Bostin Mortgage Securities Corp.
|
25,136,599,885
|
51,358,734,896
|
|
7
|
WFM
|
Wells
Fargo Asset Securities Corp.
|
19,626,493,594
|
54,301,040,825
|
|
8
|
AMQ
|
Ameriquest
Morgage Securities
|
19,536,932,462
|
26,361,542,952
|
|
9
|
BMS
|
Bank
of America Mortgage Securities
|
18,245,906,439
|
58,215,862,836
|
|
10
|
MAS
|
MASTR
Asset Securitization Trust
|
16,932,910,886
|
26,611,801,713
|
The model will have
a parameter set for each of the above 10 issuers. While running the model,
if issuer information is not available, then a generic parameter set (an average
across all issuers) will be used.
Developing a model involves access to variables that do not have too many missing values. Even though FICO is an important explanatory variable in describing the variation in prepayments for HELs, we do not have access to a full set of FICO scores across all issuers. Hence, we use a proxy variable to determine the credit rating of a borrower. We use the Spread at Origination (SatO) variable which is calculated as the difference between the loan rate and the rate on a par-priced FNMA 30 year mortgage.
The other variables we looked at include the following:
Refinancing Incentive
Aging
Original Balance
Original Loan-to-Value
Home Price Appreciation
Prepayment
Trends
The overall
prepayment speeds by issuer are shown in Chart 1.
For this study we looked at deals issued since January 2000. Any older deals were not included for modeling purposes.
We see from Chart 1 that most issuer prepayments have followed a similar trend. However, we feel that there are enough differences to warrant a separate model for each issuer.
The charts below show a few interesting trends we observed for some of the issuers. These and other trends will be discussed in detail in a subsequent paper.
For Countrywide, we see that the percentage of loans made to borrowers in the 0 to 2 SatO group kept going up as we moved from 2000 to 2003 vintage. Similarly, the percentage of loans made in the 2 to 4 SatO group kept going down.
Chart 3 shows the aging curves by SatO buckets for RFC.
We see that loans age faster as SatO goes from <0 to about positive 2. For SatO higher than 2, we observe aging curves that are slower and they plateau at lower CPRs.
Chart 4 shows the effect of refinancing incentive by SatO bucket for RFC. The refinancing incentive is obtained by dividing the loan weighted-average coupon by the FNMA 30-year Current Coupon (rate at which mortgages are trading at par). We see some interesting trends here. The first two buckets (SatO<0 and 0<=SatO<1) have similar speeds. However, as SatO goes up, speeds show an initial jump but are slower when the loans are really in the money.
The new prepayment model for HELs will soon be added to AD&Co.'s suite of Prepayment and Valuation models. As the HEL market continues to evolve, we will keep making enhancements to the model so that it reflects the state of the market. And as always, client feedback is very important to us, so feel free to contact us at 212-274-9075 with questions or comments.