Prepayment Modeling Challenges

There are five key factors that are considered when refining and maintaining the BondEdge prepayment model. First, we consider the historical performance of the model. Second, we examine whether the risk measures across a wide range of vintages and coupon ranges are reasonable and consistent. Third, we keep a close eye on macro-economic variables that may influence prepayment behavior. Fourth, we compare our fixed rate mortgage (FRM) index duration against the major index providers. Fifth, we compare our model against the SIFMA dealer median prepayment speeds. 1 Prepayment Model Specification Every prepayment model is unique; however three factors lie at the core of most models. First and foremost, a prepayment model consists of a given set of functionalities. For example, a prepayment model might contain functionality to either slow down or speed up prepayment projections based on the age of the loans under consideration. Similarly, models might also be sensitive to original loan size, borrowers with larger loan sizes will save more in terms of absolute dollars than borrowers with smaller loans given an equivalent refinancing rates. Second, the parameterizations for the functionalities mentioned above determine the ultimate behavior of the model. This is best illustrated by considering the financial and housing crisis of 2008. Prior to 2008, a given prepayment model may have forecast high prepayment projections (e.g. in excess of 45% CPR) given a 100 basis point refinancing incentive prior to 2008. This was justified by the economic conditions at the time. Lenders were making loans, even without full documentation of income or assets. In addition, the housing market was strong as measured by home price appreciation and existing home sales. However in 2008 and 2009, the same prepayment model would not likely have produced the same high prepayment projection given the same 100 basis point refinancing incentive. This is due to the fact that many models received a parameter update during this timeframe to more accurately reflect the economic conditions that applied. Specifically, this timeframe was marked by a weakening housing market, rising levels of unemployment, and tougher lending standards creating a liquidity crisis for borrowers. Third, pool and loan level data are helpful in determining the true incentive that borrowers have to refinance. Fannie Mae, Freddie Mac, and Ginnie Mae supply monthly updates on the pools they issue or back. Such updates include current pool factor, weighted average loan age, and weighted average maturity, among others. 1 2 Historical Performance One of the best ways to test the predictive power of a prepayment model using these three factors (i.e. model functionality, model parameterization, and loan characteristics) is to perform an historical analysis. The historical performance of a prepayment model can be measured by comparing projected prepayment rates versus actual prepayment rates. The historical analysis can provide feedback as to whether the prepayment model possesses sufficient functionality to respond to various economic conditions and how a given set of parameter settings of the model project prepay behavior over the passage of time. As seen in the Figure 1 below, the average 5.00% Fannie Mae 30-year fixed rate mortgage issued in 2005, represented by the BondEdge identifier FN050038, had historic prepayment speeds ranging from about 3.0% CPR in April 2008 to just over 30.0% CPR in July of 2010. An almost imperceptible seasoning profile holds from February 2008 through January 2010, as actual historical prepayments rose modestly from less than 1.0% CPR to about 3.0% CPR. This pattern of seasoning is also visible in the model results. It is also interesting to note that as the credit crisis hit in 2008, actual prepayment speeds fell to under 6.00% due to falling housing prices and a lack of funding opportunities in the refinancing arena. The results of an historical comparison can then lead to a parameter update to improve the correlation between the speeds projected by the model and the actual speeds. Typically, the decision to update model parameters will not be based solely on the performance of a narrow set of collateral characteristics, but on a range of coupon rates, and vintage years. Model stability (i.e. stable model functionality and parameters) is desirable, since it leads to a consistency in calculated risk measures such as option-adjusted spread and 1For structured deals with collateral backed by either whole loans (WH) or residential mortgage backed securities (RMBS), current collateral information is more difficult to come by since the loans are not associated with Fannie Mae, Freddie Mac, or Ginnie Mae. BondEdge has recently been enhanced to include collateral updates on WH and RMBS collaterals from a leading third party data provider. ©2010 Interactive Data Fixed Income Analytics May not be reproduced by any means without express permission. All rights reserved. 2010(0930) Prepayment Modeling Challenges 3 FN050038 0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 Feb-08 Aug-08 Feb-09 Aug-09 Feb-10 Aug-10 Historical CPR BondEdge CPR Figure 1: Prepayment comparison of historical speeds versus the BondEdge Prepayment Model for Fannie Mae issued, 30-year amortization, 5.00% Net WAC, 2008 vantage, fixed rate mortgages. effective duration over time. Consequently, prepayment model parameters tend to be retuned once or twice a year, or when major structural changes occur in the economy which influence borrower behavior. An update history of the BondEdge FRM prepayment model is included within Figure 4 below. 3 Risk Measure Reasonability and Consistency A prepayment model should be designed to deliver consistent risk measures for each issuer and loan term across the wide range of possible coupon and vintage combinations. For example, Figure 2 below show the relationship between lifetime prepayment speed (in this case % of PSA), option-adjusted spread (OAS), effective duration, and convexity for newly issued Fannie Mae (FNMA) and Ginnie Mae (GNMA) issues with 30-year and 15-year loan terms. This information is available on a monthly basis in the Fixed Rate MBS Report on the BondEdge Private Client Website. Ensuring that the prepayment model provides consistent results typically means making sure that risk measures produced by the model exhibit reasonably continuous behavior across coupons within the same issuer and vintage year. For example, one should expect the effective duration to decrease as the coupon rate increases (for newly issued collateral), since the higher prepayment speeds associated with the higher coupon rates will shorten the life of the security. 4 Macro-Economic Environment Since the mortgage and credit crisis of 2008, a number of unusual macro-economic events influenced the way that prepayment models have performed. In response, the prepayment models have been updated to more accurately reflect borrower behavior. For example, from the end of 2008 through much of 2009, the relationship between primary and secondary mortgage rates 3 became unstable (refer below to Figure 3). Primary rates reflect the mortgage rate that borrowers can actually obtain from lenders at a specific time. One of the best known measures of the primary mortgage rate is through the weekly Primary Mortgage Market Survey (or PMMS) conducted by Freddie Mac. Secondary mortgage rates are captured by price dy2The inverse relationship between coupon and effective duration may not hold for more seasoned mortgages since they tend exhibit burnout (i.e. a decreased sensitivity to refinancing incentive) as time progresses. 3 Primary and Secondary Mortgage Rates in Today’s Economy ©2010 Interactive Data Fixed Income Analytics May not be reproduced by any means without express permission. All rights reserved. 2010(0930) Prepayment Modeling Challenges 4