Articles Uncovering Loan Dissimilarities for Better ReturnsMortgage Technology - June, 2006
There also often exist processes and institutional structures to value the prepayment option for loans in aggregate, e.g. to compare FNMA 5.5 vs. FNMA 6. A "pool-level" prepayment model is used in these valuations. To obtain a measurement of prepayment propensity with any accuracy at the loan level, e.g. to compare two FNMA 5.5 loans to each other, a loan level score that summarizes these loans' attributes (other than type, age and WAC) is needed. That score then corresponds to a modifier to a standard "pool-level" prepayment model. The resulting model forms a functional connection between a projection of interest rates and a projection of prepayments for this individual loan, and one can thus calculate hedge ratios and relative value for individual loans as well as portfolios with MBS positions.Uncovering Differences A mortgage with a greater propensity to default (given borrower and loan characteristics) will be priced lower. A "reasonable" credit quality collateral may have the economic value of the default option between 2 and 10 basis points per year, or approximately 0.1%-0.6% of the value of the loan. To emphasize the point, otherwise similar loans, priced similarly, may have as much as 5% variability in their intrinsic economic value, depending on the differential propensity to refinance. Note that I used the term propensity, not probability, since the probability of current coupon refinancing is zero (no incentive). Probability is a function of interest rates and is calculated by prepayment models that take interest rates projections as input, whereas propensity is a function of borrowing and loan characteristics. Dissimilarity of Similar Loans Market participants routinely price loans with different ages and coupons differentially, but otherwise similar loans (same coupon, age loan type) with other characteristics like loan size, geography, purpose, etc. being different, are generally priced very similarly - these loans are almost always mis-priced. Variability in the intrinsic value of the mortgages because of the differential in propensities to refinance represents the largest arbitrage opportunity in the mortgage market. For example, the variability in the economic values of otherwise similar (and similarly priced) income-only strips can be as high as 30%. For otherwise similar loans, the value of loan characteristics other than age and weighted average coupon (WAC) in their effect on a loan's differential propensity to refinance or to move is summarized in its score. This measurement or score is used as a modifier to a "generic" prepayment model, a model already designed to accurately project aggregate behavior of these loans, though without taking into account the borrower and loan characteristics other than loan type, age and WAC. The algorithm for such a score is developed using historical loan prepayment experience vs. the "generic" model projections. A deal may be backed by hundreds of pools, or thousands of loans. Their detailed characteristics are generally not available through analytical systems, and even if they were, evaluation of the effects of the distributions of these characteristics of individual tranches is extraordinarily difficult. Prepayment propensity scores simplify the process of loan valuation vis a vis prepayment propensity in the same way that FICO credit scores simplify loan valuations via their default propensities. The prepayment scores do not change with movements of interest rates and implied volatilities, whereas the values of loan prepayment options do. Seeing Is Believing We scored all conforming loans originated in 1997 with WAC of 7.5% (+-0.25%). These loans were then bucketed into deciles by score. Note, in the absence of refinancing incentive (until mid 1998 and between mid 1999 and January 2001) these buckets prepaid at similar speeds, but when the interest rates dropped, the higher scored buckets always prepaid faster than the lower scored ones. And the difference was dramatic. It is interesting to note that these loans were priced in 1997 identically to each other, even though their economic values were clearly different. Given the interest rates and volatilities environment in 1999, the calculated economic values of these loans differed by as much as 3%, the value differential that was clearly realized once the interest rates did move. We have scored all agency pools in similar fashion based on additional pool information like geographic distribution, loan size, FICO, LTV, and purpose. The score's ability to discriminate pools' prepayment behavior is equally dramatic. Utilizing a score-modified prepayment model, we calculated differential economic values of a variety of pools and compared them to market pay-ups for low loan balance pools. We have found that there is fairly little relationship between the market pay-ups and the intrinsic economic value of that additional information. Other Applications We also have further extended the analysis to agency and non-agency CMO's. One can calculate the scores of agency CMO's based on the pool scores that back them, and of non-agency CMO's based on the scores of the loans that back them. We maintain databases of pools backing all agency CMO's and loans backing about 80% of non-agency CMO's. Most analytical systems do not provide pool-level or loan level information on the collateral backing the deals. The combination of scoring and the databases allows both access and highly efficient use of this information. There are a number of ways in which market participants have started utilizing this technology to take advantage of the arbitrage. Several proprietary trading desks have re-organized their trading activities around the scoring algorithm. They have been executing trades that involve a long position in a low scored IO and a high scored PO of the same collateral type, age, and coupon, and a short position in TBA's. The trade removes most of the market risk, and allows one to benefit from the desirable score-based prepayment behavior. Some trading desks have been using the algorithm to deliver lower valued (on the basis of score) collateral against TBAs, or cherry picking specified pools when the market does not price the economic value of the score properly. Mortgage bankers have started cherry picking their portfolios by selling off higher scored loans and looking to acquire lower scored ones. The score allows the immediate identification of desirable loans without having to perform a loan-by-loan analysis of the collateral backing them. Michael Bykhovsky is president and CEO of Applied Financial Technology, an analytic software firm providing mortgage banks and MBS traders with mortgage loan prepayment propensity scoring.
|
 |
 |
Prepayment propensity of mortgage loans is now seen as a critical factor in portfolio retention decisions and valuation calculations. This makes sense since there are only two risk factors in a loan: default or prepayment. There already exist institutional structures to control, value and price the option to default, but prepayment risk for many remains difficult to gauge and to manage.