With prepayments in 2001 and 2002 exhibiting different characteristics from previous refinancing waves, Merrill Lynch has just released a model that incorporates knowledge obtained from more recent prepayment experience.
Merrill analyst Akiva Dickstein noted that aside from the higher level of prepayments recently seen, there has also been a significant differentiation across collateral types and vintages. For instance, 2000 vintage collateral has prepaid much faster than 1999 vintage. Added to this, Dickstein said that there has been a segregation of numerous pools that was based on desirable characteristics such as Alt-A and low loan balance. This has led to adverse selection in the TBA market.
The new prepayment model for fixed-rate residential mortgages will help in analyzing and understanding the impact of these developments, according to Dickstein. It will incorporate a wider range of factors including average loan size, dispersion of loan size, cash out refinancing, refinancing into ARMS, media effects, capacity constraints, and loans originated at a premium.
"The broad scope of the model allows us to use a single unified framework to address a wide variety of market sectors: 30-years, 15-years, jumbo and low loan balance," Dickstein wrote in a report on the new model.
The loan size factor
Merrill has factored into its model a new methodology incorporating loan size. This would allow analysts to accurately explain the observed prepayment differences between the different vintages and those between low loan balance and standard agency pools. The report on the new model stated that a thriving low loan balance market has emerged in the past several months, with low loan balance pools commanding considerable pay-ups over TBAs.
Though the concept of loan size is nothing novel, agency loan limits have gone up and the dispersion in loan size in the agency market has increased significantly. This has led to loan size becoming a critical variable not only in making comparisons between agencies and jumbos, but within the agency market as well.
The main difference between each sector is loan size, so if loan size by definition is an integrated part of the model, then it has already accommodated more or less the main difference between the sectors.
Aside from considering loan size, the model clarifies the different prepayment responses seen in the refinancing waves of 1993, 1998 and 2001.
Merrill chose to incorporate loan size into the new model in a "very natural way," Dickstein wrote: by calculating the net present value of savings resulting from refinancing, and utilizing this as a relevant variable.
"We feel that this is a more realistic approach than the oft-used methods of utilizing the number of basis points in-the-money or the ratio of monthly payments," said Dickstein. "After all, it is the real savings that matter to a borrower, not the ratio of interest rates."
He added that this approach allows analysts to create prepayment curves for each loan size that are consistent with observed behavior.
Using the cross-sector approach
Another significant development in Merrill's new model is the ability to use a single prepayment analysis across sectors. This removes the need for a wide array of prepayment models that are built on a disparate framework, rendering relative value decisions across sectors inconsistent.
Thus, not only can this model be utilized for 30-year collateral, but it also works for jumbos and 15-years with effectively no modifications. Through this new model, Merrill attempts to unify a wide range of market sectors by incorporating, as mentioned above, loan size in its analysis.
In terms of different time periods, for instance, Merrill has used general economic variables and loan characteristics to explain prepayments as an alternative to inserting ad hoc adjustments for various time periods. The model also uses a cohesive treatment of loan size to analyze the differences in prepayment rates between various vintages. Loan size not only aids in explaining the behavior of different vintages during a given refinancing cycle, it also helps in explaining the differences observed between successive refinancing cycles.
In analyzing the 15-year and 30-year sectors, Merrill notes that the major difference between the two is that 15-years tend to have smaller loan balances than 30-years.
Merrill has simply applied its 30-year prepayment model to the 15-year sector, since its model incorporates loan size in a unified framework and the major difference between these sectors happens to be loan size.
In terms of the low loan balance sector, analysts can readily accommodate low loan balance pools without having to apply any ad hoc shifts of the prepayment curve since the model is fully based on loan size. Incorporating the loan size variable in a cohesive fashion has also allowed analysts to use the same model for jumbos as they do for low loan balance pools.
The report stated that one of the more striking phenomena in 2001 prepayments was the clear evidence of cash-out refinancing on loans with no economic incentive to refinance.
To capture this event, analysts carefully modeled cash-outs by considering the impact of such variables as home price appreciation and interest rates.
In their analysis, a cash-out happens when there are three major factors that come into play: equity must be available, the borrower must need the cash; and executing a cash-out must make financial sense.
Other features of the model
By modeling loan size dispersion and its natural change over time - and by considering that larger loans prepay faster - analysts have obtained a significantly more accurate picture of prepayments as well as a better knowledge of the burnout phenomenon.
The joint factors of media effect and capacity constraints that have come to fore as rates reach new lows have also been included in the model. Researchers have also incorporated refinancing from fixed-rate mortgages into ARMs - a factor that has played a key role as the curve has steepened in 2001 and 2002.
The new model also considers the decreased callability of securities backed by borrowers who took loans above the prevailing mortgage rate. These borrowers typically are either less interest-rate-sensitive or do not have perfect credit (or documentation).