The following article is reprinted from the October/September, 2001 issue
of On the Edge, the Interactive Data Fixed Income Analytics bimonthly newsletter.
Dealer MBS Prepayment Estimates and Median Values
Brig Belvin
Vice President and General Manager, Data
When modeling MBS securities, estimates of future prepayment rates are
vital to the analysis. Buyside and sellside participants rely on proprietary
or third-party prepayment models to determine the cash flows from which
all other risk measures (e.g., duration and convexity) are derived. From
time to time, it is useful to look at the array of prepayment estimates
that comprise the often-cited Dealer Median to see how the individual
estimates are disbursed around the median.
The Dealer Median prepayment estimate summarizes a set of values into
a single number. Single numbers are appealing, as they are easy to comprehend.
One way to derive a single value from the set of dealer estimates is
to take a simple average; however, such a number is vulnerable to being
skewed by outlying values. The median value avoids this drawback, as
it is the value in the middle of the array of inputs. It has an equal
number of values above and below it, and so seems best to represent
a “consensus.” Investors often use the dealer median prepayment
estimates to gauge how quickly or slowly a collateral type is expected
to prepay, for comparing the expected average lives of different MBS,
for regulatory compliance testing, or other selection and monitoring
purposes.
Using the Dealer Median PSA% speeds can be a reasonable thing to do.
As prepayment modeling is an inexact science, there is some perceived
“safety” in a consensus value. We should always bear in mind,
however, that the median summarizes a range of estimates that may be
widely dispersed, and this dispersion is often worth exploring.
Dealer Median Composition
We provide some pictures based on recent data comparing High, Low and
Median PSA’s versus MBS coupon rates. Notice that the median value
does not appear half-way between the high and the low values; rather,
it is skewed toward the value most cited among the contributing dealers.
As we examine coupons from discount to premium, no one dealer’s
model always produces the highest prepay speeds, no one is always the
lowest, and no one is always at the median. Rather, at any given coupon,
the median value rotates among dealers.
About 30% of observed dealers have models that change prepayment rates
quite a bit as a function of coupon (WAC). At low coupons, they have
among the slowest prepay rates and form the low end of the dealer range.
Somewhere in midrange coupons, their prepay rates are in the middle
of the dealer range. At high coupons, they have among the highest prepay
speeds. Another approximate 20% of dealers have prepay speeds that stay
in the lower two-fifths of the dealer range across all coupons. About
20% of dealer models give speeds in the upper range of prepays across
all coupons. Lastly, about 30% of dealer models have output that zigzags
back and forth across the dealer range as we move from low to high coupons.
For example, they may start high, go toward the median and end high,
or they may start low, rise to high and end at low once again.
Thus, the Dealer Median value is not consistent with any one dealer’s
model (or with BondEdge’s model). While the median value tends
to advance smoothly from coupon to coupon, as it does so it more closely
reflects one model, then another, and so on. This pattern occurs in
all recent issue fixed coupon MBS, whether Conventional or GNMA, 30-year
or 15-year amortization.
Prepayment Range Pattern
Both Conventional and GNMA fixed coupon MBS show a similar pattern of
estimated prepayment speeds across dealers, but they differ in particulars.
Both types of issues have the narrowest high-low dealer PSA range at
the lowest coupon and the widest range at the highest coupon.
For Conventionals, the high-low PSA range recently differed by about
55 to 75 points for discount to low premium coupons, and then amazingly
expanded to nearly a 600-point difference at 9% premium coupons priced
at 105 or 106. For GNMA’s, the high-low PSA difference is wider
in the middle range of coupons than wtih Conventionals, and is narrower
than Conventionals for both the lowest discount and highest premium
coupons. The PSA high-low difference was recently at about 50 points
at 6% coupons, 160 points at 7% coupons and about 440 for 9% coupons.
Average Life Varies
The differences in the average life computed from the individual dealer
prepayment estimates are significant. Using BondEdge, we calculated
the average life of new issue MBS using various dealer prepay assumptions
and found that the average life of a given collateral type can shorten
or extend by a year or more relative to the median value, depending
upon which dealer estimate is used. The high-low range of average lives
for a single coupon was as narrow as 0.5 year and as wide as 2.8 years.
The pattern across coupons is complex. All things being equal, the
higher the coupon, the shorter the average life and the narrower the
difference between high and low average lives. However, the prepay range
greatly expands at high coupon levels, widening the average life high-low
difference that high coupons would otherwise have. The net is rather
like a “bow tie” effect on Conventionals. The widest average
life variation recently is 1.3 to 1.9 years for discounts and high premiums.
The narrowest, at about 0.5 year, occurs at bonds priced between 102
and 103, recently represented by a 7.50% coupon.
The high-low pattern for GNMA is even more unusual. The high-low average
life difference is much the same as Conventionals at the 6% coupon,
but then widens to a 2.8-year difference at the 7% coupon! It then decreases
to its narrowest at the 8% coupon before widening again to the 9% coupon.
Except at the highest coupons, GNMA has distinctly greater average life
variation than Conventionals based on the variation in dealer prepay
estimates.
Summary
In summary, MBS are subject to considerable prepayment uncertainty.
The Dealer Median prepayment estimate can be used for various comparisons
and to get a general sense of the market’s prepayment expectations,
but as a single value it conceals important differences across models.
A reasonable, respectable prepayment model is likely to produce PSAs
that differ from the Dealer Median but are still well within the range
of estimates that comprise the Median value.
Real time events prompt participants in MBS markets to frequently adjust
their models. As actual prepays speed up or slow down relative to expectations,
last year’s outlier may well become this year’s new consensus;
almost by definition, some outlier will turn out to have been more accurate
than the previous consensus. We recently published a new, monthly MBS
report that summarizes the results from the BondEdge prepayment model
across collateral types, including a comparison of BondEdge’s PSA
speeds relative to the Dealer Median. We hope this analysis will prove
to be useful in monitoring and understanding both the BondEdge prepayment
model and the range of estimates underlying the Dealer Median.
