1. I
T
he vast majority of mortgage loans in the United States
are securitized in the form of agency mortgage-
backed securities (MBS). Principal and interest payments
on these securities are passed through to investors and
are guaranteed by the government-sponsored enterprises
(GSEs) Fannie Mae or Freddie Mac or by the government
organization Ginnie Mae.
1
us, investors in these securities
are not subject to loan-specic credit risk; they face only
interest rate and prepayment risk—the risk that borrowers
may renance the loan when rates are low.
2
In the primary mortgage market, lenders make loans to
borrowers at a certain interest rate, whereas in the secondary
market, lenders securitize these loans into MBS and sell them
to investors. When thinking about the relationship between
these two markets, policymakers and market commentators
usually pay close attention to the “primary-secondary spread.”
is spread is calculated as the dierence between an average
1
Fannie Mae is the Federal National Mortgage Association (or FNMA);
Freddie Mac is the Federal Home Loan Mortgage Corporation (FHLMC;
also FGLMC); Ginnie Mae is the Government National Mortgage
Association (GNMA).
2
ey also face the risk that borrowers prepay at lower-than-expected speeds
when interest rates rise.
• Whiletheprimary-secondarymortgage
ratespreadisacloselytrackedseries,itis
animperfectmeasureofthepass-through
betweensecondary-marketvaluationsand
primary-marketborrowingcosts.
• Thisstudytrackscashowsduringandafter
themortgageoriginationandsecuritization
processtodeterminehowmanydollars
(per$100loan)areabsorbedbyoriginators,
eithertocovercostsorasoriginatorprots.
• Theauthorscalculateaseriesoforiginator
protsandunmeasuredcosts(OPUCs)for
theperiod1994-2012,andshowthatthese
OPUCsincreasedsignicantlybetween
2008and2012.
• Althoughsomemortgageoriginationcosts
mayhaverisen,alargecomponentofthe
riseinOPUCsremainsunexplainedby
costincreasesalone,pointingtoincreased
protabilityoforiginators.
FRBNY Economic Policy Review / December 2013 17
AndreasFuster,LaurieGoodman,DavidLucca,LaurelMadar,LinseyMolloy,
andPaulWillen
T R G
P S
M R
Andreas Fuster and David Lucca are senior economists in the Federal Reserve
Bank of New York’s Research and Statistics Group; Laurie Goodman is the
center director of the Housing Finance Policy Center at the Urban Institute;
Laurel Madar and Linsey Molloy are associates in the Bank’s Markets Group;
Paul Willen is a senior economist and policy advisor in the Federal Reserve
Bank of Boston’s Research Department.
Corresponding authors: andreas.fuster@ny.frb.org; david.lucca@ny.frb.org
is article is a revised version of a white paper originally prepared as back-
ground material for the workshop “e Spread between Primary and Secondary
Mortgage Rates: Recent Trends and Prospects,” held at the Federal Reserve
Bank of New York on December 3, 2012. e authors thank Adam Ashcra,
Alan Boyce, James Egelhof, David Finkelstein, Kenneth Garbade, Brian Landy,
Jamie McAndrews, Joseph Tracy, and Nate Wuerel for helpful comments,
and Shumin Li for help with the data. e views expressed are those of the
authors and do not necessarily reect the position of the Federal Reserve Bank of
New York, the Federal Reserve Bank of Boston, or the Federal Reserve System.
18 The Rising Gap
mortgage interest rate (usually coming from the Freddie Mac
Primary Mortgage Market Survey) and a representative yield
on newly issued agency MBS—the “current-coupon rate.”
Chart 1 shows a time series of the primary-secondary
spread through the end of 2012. e spread was relatively
stable from 1995 to 2000, at about 30 basis points; it
subsequently widened to about 50 basis points through early
2008, but then reached more than 100 basis points in early
2009 and during 2012. Following the September 2012 Federal
Open Market Committee announcement of additional MBS
purchases, the spread temporarily rose to more than 150 basis
points—a historical high that attracted much attention from
policymakers and commentators at the time.
While the primary-secondary spread is a closely watched
series, it is an imperfect proxy for the degree to which secondary-
market movements are reected in mortgage borrowing costs
(the “pass-through”) since, among other things, the secondary
yield is not directly observed, but model-determined, and thus
subject to model misspecication. Furthermore, mortgage
market pass-through depends on the evolution of the GSEs’
guarantee fees (or “g-fees,” the price the GSEs charge for insuring
the loan) as well as on mortgage originators’ margins. To
understand changes in the extent of pass-through over time, it
is useful to track the two components separately. While g-fee
changes are easily observable, we argue that originator margins
are best studied by tracking the dierent cash ows during
and aer the origination process, rather than by looking at the
primary-secondary spread (even aer netting out g-fees). Indeed,
since originators are selling the loans, their margin depends on
the price at which they can sell them, rather than the interest rate
on the security into which they sell the loans.
To get a sense of what lenders earn from selling loans, we
rst consider a simple “back-of-the-envelope” calculation.
We track the secondary-market value of the typical oered
mortgage loan (according to the Freddie Mac survey) over
time, assuming that the lender securitizes and sells the loan
as an agency MBS. To do so, we rst deduct the g-fee from
the loan’s interest stream. We then compute the value of
the remaining interest stream by interpolating MBS prices
across coupons and subtracting the loan amount of $100.
3
Chart 2 shows that the approximate net market value of a
mortgage grew from less than 100 basis points (or $1 per
$100 loan) before 2009 to more than 350 basis points in
the second half of 2012. Taken literally, the chart implies
that lender costs (other than the g-fee), lender prots, or a
combination of the two must have increased by 300 basis
points, or a factor of four, in ve years.
In this article, we rst present a more detailed calculation
of originator prots and costs, and then attempt to explain
their rise by considering a number of possible factors
3
For instance, assume that the mortgage note rate is 3.75 percent and
the g-fee is 50 basis points, such that the remaining interest stream
is 3.25 percent. Assuming that the 3.0 percent MBS trades at 102 and
the 3.5 percent MBS trades at 104.5, the approximate market value of
this mortgage in an MBS pool would then be simply the average of the
two prices, 103.25, or 3.25 net of the loan principal.
Chart 1
The Primary-Secondary Spread
Basis points
Sources: Bloomberg L.P.; Freddie Mac.
0
25
50
75
100
125
150
175
1210080604020098961995
Eight-week
rolling window
Weekly
0
1
2
3
4
5
1211100908072006
C
Back-of-the-Envelope Calculation of the
Net Market Value of a Thirty-Year Fixed-Rate
Mortgage Securitized in an Agency MBS
Sources: JPMorgan Chase; Freddie Mac; Fannie Mae; authors’
calculations.
Notes: e chart shows the interpolated value of a mortgage-backed
security (MBS) with coupon (r
prim ary
– g-fee) minus 100. e line
reects an eight-week rolling window average; the calculation uses
back-month MBS prices.
Dollars per $100 loan
FRBNY Economic Policy Review / December 2013 19
aecting them. In section 2, we begin with a general
discussion of the mortgage origination and securitization
process, and how originator prots are determined.
Here, we include a detailed discussion of the valuation of
revenues from servicing and points as well as costs from
g-fees, based on standard industry methods. Next, in
section 3 we use these methods to derive a time series of
average originator prots and unmeasured costs (OPUCs)
for the period 1994-2012, which largely reects the time-
series pattern of Chart 2. We then compare OPUCs and
the primary-secondary spread as measures of mortgage
market pass-through. Finally, in section 4 we turn to
possible explanations for the increase in OPUCs, including
putback risk, changes in the valuation of mortgage servicing
rights, pipeline hedging costs, capacity constraints, market
concentration, and streamline renancing programs. While
some of the costs faced by originators may have risen over
the period 2008-12, we conclude that a large component of
the rise in OPUCs remains unexplained by cost increases
alone, suggesting that originators’ prots likely increased
over this period. We then discuss possible sources of the
rise in protability. Capacity constraints likely played a
signicant role in enabling originator prots, especially
during the early stages of renancing waves. Pricing power
coming from renancing borrowers’ switching costs could
have been another factor sustaining originator prots.
4
2. M P
M O
2.1 e Origination and Securitization
Process
e mortgage origination process begins when a borrower
seeks a quote for a loan, either to purchase a home or to
renance an existing mortgage. Based on the borrower’s
credit score, stated income, loan amount, and expected
loan-to-value (LTV) ratio, an originator oers the borrower
a combination of an interest rate and an estimate of the
amount of money the borrower will need to provide up front
4
Importantly, this article focuses on longer-term changes in the level of
originator prots and costs, rather than on the high-frequency pass-through
of changes in MBS valuations to the primary mortgage market.
to close the loan.
5
For example, for a borrower who wants
a $300,000, thirty-year xed-rate mortgage, the originator
may oer a 3.75 interest rate, known as the “note rate,” with
the borrower paying $3,000 (or 1.0 percent) in closing costs.
If the borrower and originator agree on the terms, then the
originator will typically guarantee these terms for a “lock-in
period” of between thirty and ninety days, and the borrower
will ocially apply for the loan.
During the lock-in period, the originator processes the
loan application, performing such steps as verifying the
borrower’s income and the home appraisal. Based on the
results of this process, borrowers may ultimately not qualify
for the loan, or for the rate that the originator initially
oered. In addition, borrowers have the option to turn
down the loan oer, for example, because another originator
may have oered better loan terms. As a result, many loan
applications do not result in closed loans. ese “fall-outs”
uctuate over time and present a risk for originators, as we
discuss in more detail in section 4.
Originators have a variety of alternatives to fund loans:
they can securitize them in the private-label MBS market or
in an agency MBS, sell them as whole loans, or keep them on
their balance sheets. In the following discussion, we focus on
loans that are “conforming” (meaning that they fulll criteria
based on loan amount and credit quality, so that they are eligi-
ble for securitization by the GSEs), and assume securitization
in an agency MBS, meaning that this option either dominates
or is equally protable to the originator’s alternatives.
6,7
5
roughout this article, we use the terms “lender” or “originator” somewhat
imprecisely, as they lump together dierent origination channels that in
practice operate quite dierently. Currently, the most popular origination
channel is the “retail channel” (for example, large commercial banks that lend
directly), which accounts for about 60 percent of loan originations, up from
around 40 percent over the period 2000-06 (source: Inside Mortgage Finance).
e alternative “wholesale” channel consists of brokers and “correspondent”
lenders. Brokers have relationships with dierent lenders that fund their
loans, and account for about 10 percent of originations. Correspondent
lenders account for 30 percent of originations, and are typically small
independent mortgage banks that have credit lines from and sell loans
(usually including servicing rights) to larger “aggregator” or “sponsor” banks.
Our discussion in this section applies most directly to retail loans.
6
e fraction of mortgages that are not securitized into agency MBS has
steadily decreased in recent years, according to Inside Mortgage Finance:
while the estimated securitization rate for conforming loans ranged
from 74 to 82 percent over the period 2003-06, it has varied between
87 and 98 percent since then (the 2011 value was 93 percent). e private-
label MBS market has eectively been shut down since mid-2007, with the
exception of a few deals involving loans with amounts exceeding the agency
conforming loan limits (“jumbo” loans).
7
Our discussion throughout this article applies directly to conventional
mortgages securitized by the GSEs Fannie Mae and Freddie Mac; the process
of originating Federal Housing Administration (FHA) loans and securitizing
them through Ginnie Mae is similar, but with some dierences (such as
insurance premia) that we do not cover here.
20 The Rising Gap
A key feature of an agency MBS is that principal and interest
payments for these securities are guaranteed by the GSEs.
8
e
GSEs charge a monthly ow payment, the g-fee, which is a xed
fraction of the loan balance. Flow g-fees do not depend on loan
characteristics but may dier across loan originators. Until 2012,
ow g-fees averaged approximately 20 basis points per year,
but during 2012 they rose to about 40 basis points, reecting a
Congressionally mandated 10-basis-point increase to fund the
2012 payroll tax reduction and another 10-basis-point increase
mandated by the Federal Housing Finance Agency (FHFA). As
we discuss below, originators can convert all or part of the ow
g-fee into an up-front premium by “buying down” the g-fee.
Alternatively, they can increase the ow g-fee and receive an up-
front transfer from the GSE by “buying up” the g-fee.
Since 2007, the GSEs have also been charging a separate
up-front premium due upon delivery of the loan, known as
the loan-level price adjustment (LLPA).
9
e LLPA contains a
xed charge for all loans (currently 25 basis points) known as
an adverse-market delivery charge, as well as additional loan-
specic charges that depend on loan characteristics such as
the term of the loan, the LTV, and the borrower’s FICO score.
For instance, as of early 2013, the LLPA for a borrower with a
FICO score of 730 and an LTV of 80 was 50 basis points (for a
thirty-year xed-rate loan; the charge is waived for loans with
a term of een or fewer years). Together with the 25-basis-
point adverse-market delivery charge, this implies that the loan
originator pays an up-front fee equal to 0.75 percent of the loan
amount. us, the total up-front transfer between the originator
and GSE consists of the LLPA plus or minus potential g-fee
buy-ups or buy-downs, which can be either positive or negative.
For simplicity, our discussion assumes that the transfer from the
originator to the GSE is positive and refers to it as an “up-front
insurance premium” (UIP).
Once an originator chooses to securitize the loan in an
agency MBS pool, it can select from dierent coupon rates,
which typically vary by 50-basis-point increments. e note rate
on the mortgage, for example, 3.75 percent, is always higher than
the coupon rate on an agency MBS, for example, 3.0 percent.
Who receives the residual 75-basis-point interest ow?
Assuming the originator does not buy up or down the g-fee,
approximately 40 basis points go to the GSEs (as of early 2013),
leaving 35 basis points of “servicing income.” e GSEs require
the servicer to collect at least 25 basis points in servicing income,
known as “base servicing.” Base servicing is tied to the right
8
If the loan is found to violate the representations and warranties made by the
seller to the GSEs, the GSEs may put the loan back to the seller.
9
LLPA is the ocial term used by Fannie Mae; Freddie Mac calls the
corresponding premium “postsettlement delivery fee.” e respective fee grids
can be found at www.fanniemae.com/content/pricing/llpa-matrix.pdf and
www.freddiemac.com/singlefamily/pdf/ex19.pdf.
and obligation to service the loan (which involves, for instance,
collecting payments from the borrower) and can be seized by the
guaranteeing GSE if the servicer becomes insolvent. Servicing
income in excess of 25 basis points—10 basis points in this
example—is known as “excess servicing,” and is a pure interest
ow. One might surmise here that a loan in a 3.0 percent pool
must have a rate of 3.65 percent or higher (3.0 plus 40 basis
points for the g-fee plus 25 basis points for base servicing),
but recall from above that the originator can buy down the
g-fee so, in fact, the minimum note rate in a 3.0 percent pool
is 3.25 percent. In practice, for a mortgage of a given note rate,
originators compare the protability of pooling it in dierent
coupons, as described below.
Originators typically sell agency loans in the so-called TBA
(to-be-announced) market. e TBA market is a forward market
in which investors trade promises to deliver agency MBS at
xed dates one, two, or three calendar months in the future. For
concreteness, Exhibit 1 displays TBA prices from Bloomberg at
11:45 a.m. on January 30, 2013. At this time, investors will pay
102 14+/32≈102.45 for a 3.0 percent Fannie Mae (here denoted
FNCL) MBS for April settlement. To understand the role of the
TBA market, suppose that Bank A expects to have $100 million
of 3.5 percent note rate mortgages available for delivery in
April. In order to hedge its interest rate risk, Bank A will then
sell $100 million par of 3.0 percent pools “forward” in the TBA
market at a price of $102.45 per $100 par, to be delivered on
the standard settlement day in April. Over the following weeks,
E 1
Example of a TBA Price Screen
Source: Bloomberg L.P.
Notes: Prices are quoted in ticks, which represent 1/32
nd
of a dollar; for
instance, 103-01 means 103 plus 1/32 = $103.03125 per $100 par value.
e “+” sign represents half a tick (or 1/64). Quotes to the le of the
“/” are bids, while those to the right are asks (or oers).
FRBNY Economic Policy Review / December 2013 21
Bank A assembles a pool of loans to be put in the security and
delivers the loans to Fannie Mae, which then exchanges the
loans for an MBS. is MBS is then delivered by Bank A on
the contractual settlement day to the investor who currently
owns the TBA forward contract in exchange for the promised
$102.45 million. A key feature of a TBA trade is that at the time of
trade, the seller does not specify which pools of loans it will deliver
to the buyer—this information is “announced” only shortly before
the trade settles. As a consequence, market participants generally
price TBA contracts under the assumption that sellers will deliver
the least valuable—or “cheapest-to-deliver”—pools at settlement.
10
2.2 How Does an Originator Make Money on
the Transaction?
A mortgage loan involves an initial cash ow at origination
from investors to the borrower, and subsequent cash ows
from the borrower to investors as the borrower repays the
loan principal and interest.
Exhibit 2 maps these cash ows for
a mortgage loan securitized in a Fannie Mae MBS and sold
in the TBA market. e top panel shows the origination cash
ow, which involves the investor paying price TBA(r
coupon
) to
the originator in exchange for an MBS with coupon rate r
coupon
.
10
See Vickery and Wright (2013) for an overview of the TBA market.
From the investor’s payment, an originator funds the loan and
pays any UIP to Fannie Mae.
11
Together with points received
from the borrower, the cash ow to the originator when the
loan is made equals:
Ω
_
_
_
Origination cash ow (1)
= TBA(r
coupon
) + points − 100 − UIP.
rough the life of the loan (middle panel of Exhibit 2),
a borrower pays the note rate, r
note
, from which Fannie Mae
deducts the g-fee and the investor gets r
coupon
, leaving servicing
cash ow to the originator equal to:
σ
t
_
_
_
servicing cash ow
t
= r
note
− g-fee − r
coupon
. (2)
Originator prots per loan are the sum of prots at
origination (equation 1) and the present value (PV) of the
servicing cash ow (equation 2) less all marginal costs (other
than the g-fee) of originating and servicing the loan, which we
call “unmeasured costs.” us,
originator prots = Ω + PV(σ
1
, σ
2
,…) (3)
− unmeasured costs.
11
Here and below, “originator” refers to all actors in the origination and
servicing process, that is, if a loan is originated through a third-party
mortgage broker, for instance, the broker will earn part of the value.
Exhibit 2
Mortgage Loan Securitized in an Agency MBS and Sold in TBA Market: The Money Trail
Cash ow
from investor
to borrower
(at time of
origination)
Cash ow
from borrower
to investor
(during life of
loan; expressed in
annual terms)
Net benet
•
Receives $100 for loan
•
Pays points to originator
for closing costs
100 - points
- PV(r
note
)
- PV(principal repayment)
Origination Cash Flow:
Ω = TBA(r
coupon
) +
points – 100 – UIP
Servicing Cash Flow:
σ
t
=
r
note
- g-fee - r
coupon
OPUCs = Ω + PV(σ
1, ...
)
= TBA(r
coupon
) - UIP
- 100 + points +
PV(r
note
- g-fee - r
coupon
)
UIP + PV(g-fee)
Borrower Originator
Government-Sponsored
Enterprise Investor
•
Receives UIP
•
Receives g-fee
PV(r
coupon
)
+ PV(principal repayment)
- TBA(r
coupon
)
•
Pays TBA(r
coupon
)
for loan
•
Pays r
note
•
Pays principal repayment
•
Receives r
coupon
•
Receives principal
repayment
Note: TBA(r
coupon
) is the price of a mortgage-backed security (MBS) with coupon rate r
coupon
in the “to-be-announced” market; UIP is up-front insurance
premium (consisting of loan-level price adjustments plus or minus potential g-fee buy-ups or buy-downs); PV is present value.
22 The Rising Gap
In our empirical exercise below, we study the sum of prots
and unmeasured costs, which is what we can observe:
originator prots and (4)
unmeasured costs (OPUCs) = Ω + PV(σ
1
, σ
2
,…).
In later sections of the article, we attempt to assess to what
extent changes in unmeasured costs can explain uctuations
in OPUCs.
We next consider a specic transaction to illustrate how
the computations in Exhibit 2 are done in practice. Consider
a loan of size $100 with a note rate of 3.75 percent locked in
on January 30 for sixty days by a borrower with a FICO score
of 730 and an LTV ratio of 80. e borrower agrees to pay
1 point to the originator for the closing, and the originator
sells the loan into a TBA security with a 3.0 percent coupon
for April settlement to allow sixty days for closing. Assuming
the loan closes, how high are the OPUCs?
Computing the net revenue at origination, Ω, is relatively
straightforward. According to Exhibit 1, investors pay
$102.45 for every $100 of principal in a TBA security with
a 3.0 percent coupon. As discussed earlier, the up-front
insurance premium from the LLPA (and assuming no g-fee
buy-up/-down) at the time was 0.75 percent of the loan
(or 0.75 points). e originator collects 1 point from the
borrower, remitting $100 for the loan, yielding Ω = 2.7 points.
Valuing the stream of servicing income aer origination,
(σ
1
, σ
2
, …), is more complicated. For now, we assume that the
originator does not buy up or down the g-fee—a decision that
we will revisit below. is means that from the borrower’s
interest ow of 3.75 percent, the GSEs collect 40 basis points,
while the investors get 3.0 percent, leaving 35 basis points in
ow servicing income, σ
t
, decomposed into 25 basis points of
base servicing and 10 basis points of excess servicing. ere
are a number of alternative ways to determine the present
value of these ow payments:
IO Strip Prices or Coupon Swaps
Servicing income can be thought of as an interest-only (IO)
strip, which is a security that pays a ow of interest payments,
but no principal payments, to investors as long as a loan is
active.
12
e main driver of the valuation of an IO strip is
the duration of the loan—an IO strip is far more valuable if
one expects the borrower to prepay in ve years as opposed
to one year; as in the latter case, interest payments accrue
for a much shorter time period. One simple way to value
IO strips is to construct them from TBA securities through
coupon swaps. For example, going long on a 3.5 percent
MBS and short on a 3.0 percent MBS generates interest cash
ows of 50 basis points with prepayment properties that
correspond roughly to loans in 3.0 and 3.5 pools. According
to Exhibit 1, that 50-basis-point IO strip for April settlement
would cost 2 11/32 (104 25+/32 minus 102 14+/32) ≈ 2.34.
Since our originator has only 35 basis points of servicing,
the coupon swap method would value servicing
rights at 35/50 × 2.34 ≈ 1.6, resulting in OPUCs of
2.7 + 1.6 = 4.3 points.
13
is method ignores the fact that base servicing
generates other revenues, such as oat income, in addition
to the IO strip. To account for this additional value, it
is oen assumed that the base servicing is worth more
than the present value of the IO strip. Assuming that base
servicing is worth, for example, 25 percent more than
excess servicing would yield a PV of servicing income of
(25 × 1.25 + 10)/50 × 2.34 ≈ 1.9, so that OPUCs would equal
2.7 + 1.9 = 4.6 points.
Another shortcoming of the coupon swap method is that the
coupon swap reects dierences in assumed loan characteristics
across coupons. For example, TBA prices may reect the
fact that higher coupons are older securities having dierent
prepayment characteristics. ese dierences will distort the
valuations of interest streams from the coupon swaps.
14
Constant Servicing Multiples
An alternative method for valuing servicing ows is to use
xed accounting multiples that reect historical valuations of
12
Another way to describe an IO strip is as an annuity with duration equal to
the life of the loan.
13
is is the method implicitly used in the back-of-the-envelope calculation
in Chart 2, except that there we ignored points paid by the borrower.
14
As an illustration, a 50-basis-point IO strip from a new 4.0 percent loan
may not be worth as much as the price dierence between the 3.5 and the
4.0 TBAs suggests, because the 4.0 TBAs may consist of loans that are older
or credit impaired and thus prepay more slowly.
Computing the net revenue at
origination, Ω, is relatively
straightforward.… Valuing the stream
of servicing income after origination,
(σ
1
, σ
2
, …), is more complicated.
FRBNY Economic Policy Review / December 2013 23
servicing. In the industry, the base servicing multiple is oen
assumed to be 5x, meaning that the present value of 25 basis
points equals 1.25, while excess servicing is assumed to be
valued at 4x, so that the value of the excess servicing in our
example is 0.40. Using these servicing multiples, we see that
the servicing income in our example is worth 1.65, meaning
that OPUCs for this loan would be 2.7 + 1.65 = 4.35 points.
Buy-ups
As mentioned above, originators can convert the g-fee into an
up-front premium, or vice versa, using buy-ups and buy-downs.
A buy-up means that the ow g-fee increases, but to compensate,
the GSE will reduce the UIP (or, in case it is negative, transfer
money to the originator upon delivery of the loan). us, buying
up the g-fee is a way to reduce the ow servicing income and
increase income at the time of origination.
e GSEs oer a buy-up multiple, which is communicated
to originators (but not otherwise publicly known), and varies
over time, presumably with the level of the coupon swap. If,
for example, the buy-up multiple is 3x, then a 10-basis-point
increase in the g-fee reduces UIP by 30 basis points, lowering
σ
t
by 0.1 and raising Ω by 0.3. Note that only excess servicing,
σ
t
, -0.25, can be “monetized” this way, while 25 basis points
of base servicing still need to be retained and valued by the
originator. If we assume a base servicing multiple of 5x, as
above, then buying up the g-fee by 10 basis points would lead
to OPUCs of 3.0 + 1.25 = 4.25.
e buy-up multiple provides a lower bound on the
valuation of excess servicing—the originator (or some other
servicer) may value it at a higher multiple; but if it does not,
it can sell its excess servicing to the GSEs. To what extent
originators want to take advantage of this option depends on
a number of factors. For example, as we discuss in section 4.1,
the upcoming implementation of Basel III rules may require
banks to hold additional capital against mortgage servicing
assets, which may lower their eective valuation of servicing
income. By buying up the g-fee, these banks can turn servicing
cash ows that are subject to additional regulatory capital
charges into cash. Another potential factor is the originator’s
beliefs about the prepayment properties of a pool of loans. For
example, if a lender believes that the expected lifetime of a pool
is shorter than average, it may choose to buy up the g-fee.
Market Prices of Servicing Rights
Finally, there is an active market for trading servicing rights,
which can be sold by originators at origination or well aer-
ward. One can use market prices to value servicing rights, but
since not all servicing rights change hands, it is dicult to
know whether the ones that trade are systematically more or
less valuable than the ones that originators hold.
2.3 Best Execution
Lenders can decide to securitize a loan into securities having
dierent coupons, which involves dierent origination and
servicing cash ows. e strategy that maximizes OPUCs is
known in the industry as “best (or optimal) execution.”
15
us far, we have assumed that the originator securitizes
the loan in a 3.0 coupon. However, since the note rate is 3.75,
the originator could alternatively sell it in a 3.5 coupon.
16
Given that the originator must retain 25-basis-point base
servicing, such a choice would require buying down the
entire 40-basis-point g-fee, meaning that instead of any ow
payment to the GSE, the originator pays the full insurance
premium up front. Exactly like the buy-up multiple discussed
above, the GSEs also oer a (higher) buy-down multiple,
which determines the cost of this up-front payment.
Using the prices in Exhibit 1, we note that the price
of a 3.5 TBA coupon is 104 24+/32=104.77, meaning
that changing coupons would increase loan sale revenues
by 2.32 points. If we assume the buy-down multiple
equals 7, then UIP would increase by 2.8 points relative to
the 3.0 coupon case. Ω is thus equal to 2.22, or 0.48 less than
it would be for the 3.0 coupon case. Meanwhile, servicing
income is now simply σ
t
= 0.25, as the ow g-fee has been
bought down to zero, and with an assumed base servicing
multiple of 5x, OPUCs for this execution would equal
2.22 + 1.25 = 3.47.
Comparing this OPUC value with the “constant
servicing multiples” case above, we see that pooling into
the 3.0 coupon would generate higher OPUCs than the
3.5 coupon and thus would be best execution for a mortgage
with the 3.75 percent note rate.
However, this conclusion is sensitive to a number of
assumptions—in particular, the valuation of excess servicing
and the buy-down multiple.
17
As shown in Table 1, pooling in
the higher coupon becomes more attractive as the buy-down
multiple decreases or the excess servicing multiple decreases.
15
See Bhattacharya, Berliner, and Fabozzi (2008) for an extensive discussion
of pooling economics and mortgage pricing that also includes nonagency
securitizations.
16
e originator could also place the loan in a 2.5 percent or lower coupon—
the only restriction is that the note rate cannot be more than 250 basis points
above the coupon.
17
As base servicing always needs to be retained, its valuation does not aect
best execution—it shis OPUCs up or down equally for all coupons.
24 The Rising Gap
2.4 Rate Sheets and Borrower Choice
Until now, we have taken the borrower choice as given—the
borrower pays 1 point at origination and is oered a note rate of
3.75. However, from our OPUC calculations, it is clear that there
are other combinations of note rate and points that would be
equally protable for the originator. For example, if the borrower
paid a note rate of 4.0 instead, and the originator still pooled
the loan into a 3.0 coupon, then excess servicing would increase
by 25 basis points, leading to 1 point higher revenue under
an excess servicing multiple of 4x. erefore, the originator
could maintain its prot margin by oering the borrower a
combination of 0 points at closing and a note rate of 4.0.
18
Indeed, originators oer borrowers precisely these sorts
of alternatives between closing costs and rates. Table 2 shows
part of a rate sheet provided by a bank to a loan ocer on
January 30, 2013.
19
e entries in the table are “discount
points,” which are points paid by the borrower at closing to
lower the note rate on the loan. For example, assume that the
total closing fees the originator would charge the borrower
without any discount points would equal 1.58 points—
sometimes referred to as “origination points.” ese fees
include application processing costs, compensation for the
loan ocer, and also the LLPA (0.75 points in our example),
which is usually charged directly to the borrower.
Our baseline borrower has a sixty-day lock-in period and
a note rate of 3.75 percent; accordingly, based on the rate
sheet, the borrower is contributing -0.581 discount points.
is means that the bank is actually paying the borrower
cash up front (oen referred to as a “rebate”), which reduces
closing costs from 1.58 points to the 1 point assumed
18
In fact, the 4.0 note rate might increase the prot margin, because it would
potentially alter the best-execution coupon.
19
Actual sample rate sheets can be found, for instance, at www.53.com/
wholesale-mortgage/wholesale-rate-sheets.html. Most lenders do not make
their rate sheets available to the public.
throughout the example. If the borrower wanted a lower note
rate, for example, 3.5 percent, then the closing costs would
rise by 1.044 − (-0.581) = 1.625, or from 1 to 2.625 points.
Alternatively, by choosing a rate of 4.125 percent, the
borrower could get a rebate of 1.581 points and would pay
nothing at closing.
As shown in the rate sheet, there is no single “mortgage
rate.” Rather, a large number of dierent note rates are
available to borrowers on any given day, typically in
increments of 0.125.
20
Originators simply change the number
of discount points oered for the dierent note rates one or
more times a day, reecting secondary-market valuations
(TBA prices), servicing valuations, and GSE buy-up/
buy-down multiples.
21
20
at said, banks will oen quote a headline mortgage rate, which is
generally the lowest rate such that the number of discount points required
from the borrower is “reasonable” (this rate is sometimes referred to as
the “best-execution” rate for the borrower, not to be confused with the
originator’s best execution). In the example rate sheet, this rate would likely be
3.75 or 3.625, as going below 3.625 requires signicant additional points from
the borrower.
21
e set of available note rates on a given day generally depends on which
MBS coupons are actively traded in the secondary market.
T
ExampleofaMortgageRateSheet
Lock-in Period
Note Rate Fieen Days irty Days Sixty Days
4.750 (3.956) (3.831) (3.706)
4.625 (3.831) (3.706) (3.581)
4.500 (3.706) (3.581) (3.456)
4.375 (3.331) (3.206) (3.081)
4.250 (3.081) (2.956) (2.831)
4.125 (1.831) (1.706) (1.581)
4.000 (1.456) (1.331) (1.206)
3.875 (1.081) (0.956) (0.831)
3.750 (0.831) (0.706) (0.581)
3.625 (0.081) 0.044 0.169
3.500 0.794 0.919 1.044
3.375 1.669 1.794 1.919
3.250 2.544 2.669 2.794
3.125 3.919 4.044 4.169
Source: www.53.com/wholesale-mortgage/wholesale-rate-sheets.html on
January 30, 2013.
Notes: Figures are in percentage points of the loan amount. Loan type is a
thirty-year xed-rate loan. Column 1 shows the annual interest rate to be
paid by the borrower over the life of the loan. Columns 2-4 show the points
the borrower needs to pay up front to obtain the interest rate in column 1,
for dierent lock-in periods. Parentheses denote negative gures.
T
DependenceofBestExecutiononExcessServicing
andBuy-DownMultiples
Excess Servicing Buy-Down OPUCs(3.0) OPUCs(3.5)
Multiple Multiple (Points)
4x 7x 4.35 3.47
4x 5x 4.35 4.27
3x 5x 4.25 4.27
Sources: Bloomberg L.P.; authors’ calculations.
Note: OPUCs are originator prots and unmeasured costs.
FRBNY Economic Policy Review / December 2013 25
2.5 Summary: Trade-os, Trade-os
Everywhere
As shown in the preceding discussion, the dierent actors in
the origination and securitization process have a number of
trade-os available to them. Borrowers can decide between
paying more points up front and paying a higher interest rate
later. Originators can choose between dierent coupons into
which to pool a loan, which imply dierent origination and
servicing cash ows; in addition, as part of this decision, origi-
nators can choose whether to pay the GSE insurance premium
up front or as a ow. Finally, investors can choose to invest
in securities with dierent coupons, with higher coupons
requiring a larger initial outlay, but subsequently generating
higher ow payments. Investor demand for dierent coupons,
which reects their prepayment and interest rate projections,
ultimately aects originators’ best-execution strategies and
thus the point-rate grid oered to borrowers.
3. M OPUC T
Our goal in this section is to derive an empirical measure
of average OPUCs (equation 4) for thirty-year xed-rate
mortgages for the period 1994 to 2012. To do so, we need to
make a number of assumptions.
First, rather than valuing each possible loan note rate, we
value a hypothetical mortgage having a note rate equal to the
survey rate from Freddie Mac’s Primary Mortgage Market
Survey, at weekly frequency. We also use the weekly time
series of average points paid from the same survey.
Second, rather than accounting sepa-
rately for LLPAs and the ow g-fee, we use an “eective” g-fee,
which assumes that LLPAs are paid over the life of the loan,
as reported in Fannie Mae’s Securities and Exchange Com-
mission Form10-Q lings. e average size of the eective
g-fee is shown in Chart 3. In our calculations, we incorporate
anticipated changes in g-fees. In particular, the 10-basis-point
increases that came into eect on April 1, 2012, and Decem-
ber 1, 2012, are assumed in our calculations to apply to loans
originated January 1 and September 1, respectively, which is
right aer the increases were announced.
ird, as explained above, we need to value the servicing
income ow. e coupon swap method has the advantage of
being based on current market prices that reect changes in
the duration of the cash ows. But, as mentioned earlier, the
coupon swap may also reect dierences in assumed loan
characteristics across coupons; therefore, it may be a poor
proxy for the value of an interest strip from a new loan.
To circumvent this issue, and also for the sake of
simplicity, our baseline calculations use xed multiples of
5x for base servicing, 4x for excess servicing, and 7x for
buy-downs.
22
ese are commonly assumed values in
industry publications. Later, we explore the sensitivity of
OPUCs to alternative assumptions.
Finally, we do a best-execution calculation, considering
three dierent TBA coupons (using back-month prices)
into which the mortgage could potentially be pooled.
23
e
highest coupon is set such that it requires the originator to
buy down some or all of the g-fee up front, while instead,
for the other two possible coupon options, the originator
retains positive excess servicing because the loan’s interest
payment is more than sucient to cover the g-fee and base
servicing.
24
e best execution among the three options
determines our OPUC value for the week in question.
Before turning to the weekly OPUC time series, we report
in Table 3 a detailed OPUC calculation on a given day. We
can infer, from the bottom of the table, that the mid-coupon
execution is optimal in this example.
22
We assume the buy-up multiple to be smaller than 4x, such that, in our
calculations, buy-ups are never used.
23
e use of back- rather than front-month TBA price contracts reects the
originators’ desire to hedge price movements during the lock-in period, as
discussed in more detail in section 4.
24
Depending on the mortgage rate, pooling into the highest candidate
coupon may not actually be a possibility—as explained, the mortgage rate
needs to exceed the coupon rate by at least 25 basis points.
15
20
25
30
35
40
45
50
55
12100806042002
Chart 3
Average Effective Guarantee Fee
Basis points
Source: Fannie Mae SEC Forms 10-K and 10-Q, various issues
through 2012:Q4.
26 The Rising Gap
3.1 Results
e weekly OPUC series over the period 1994 to 2012 is
shown in Chart 4. e series averaged about $1.50 between
1994 and 2001, then temporarily increased to the
$2.00-$3.00 range over 2002-03, before declining again
and remaining below $2.00 for most of the period 2005-08.
e OPUC measure jumped dramatically to more than
$3.50 in early 2009 and then again in mid-2010. Most notably,
however, it increased further over 2012, and reached highs
of more than $5 per $100 loan in the second half of the year,
before declining again toward the end of 2012.
As shown in the back-of-the-envelope calculation in
Chart 2, the higher valuation of loans in the MBS market is
the main driver of the increase in OPUCs toward the end
of our sample period. Relative to that gure, the increase
in OPUCs over 2009-12 in Chart 4 is less dramatic; this is
because the earlier calculation implicitly valued servicing
through coupon swaps, which were very low in early 2009
but relatively high since 2010. In contrast, in Chart 4 we have
0
1
2
3
4
5
6
121008
06
04
020098961994
Dollars per $100 loan
Sources: JPMorgan Chase; Freddie Mac; Fannie Mae; authors’ calculations.
Chart 4
Originator Profits and Unmeasured Costs,
1994-2012
Eight-week
rolling window
Weekly
T
ExampleofOPUCsBest-ExecutionCalculation
TBA Coupon (Percent) 3.5 4.0 4.5 (1)
Coupon-independent inputs (percent)
Mortgage rate 4.78 4.78 4.78 (2)
Points 0.7 0.7 0.7 (3)
Eective g-fee 0.261 0.261 0.261 (4)
Base servicing 0.25 0.25 0.25 (5)
Excess servicing 0.769 0.269 -0.231 (6) = (2) − (1) − (4) − (5)
Coupon-specic inputs (dollars per par value)
TBA price (back-month) 97.55 99.95 101.67 (7)
Value of base servicing 1.25 1.25 1.25 (10) = 5 × (5)
Value of excess servicing 3.08 1.08 (11) = 4 × (6) if (6) > 0
G-fee buy-down -1.62 (12) = 7 × (6) if (6) < 0
Revenues from TBA sale less payout to borrower -1.75 0.65 2.37 (13) = (7) − (100 − (3))
Value of servicing net of g-fee 4.33 2.33 -0.37 (14) = (10) + (11) + (12)
OPUCs
By coupon 2.58 2.98 2.00 (15) = (13) + (14)
Best-execution 2.98 (16) = max(15) if (2) − (1) > .25
Source: Authors’ calculations.
Note: Calculation is for April 30, 2009. OPUCs are originator prots and unmeasured costs; TBA is “to-be-announced.”
FRBNY Economic Policy Review / December 2013 27
assumed constant multiples.
25
As we discuss in more detail
below, servicing right valuations appear to have declined,
rather than increased, over the past few years, supporting the
use of xed multiples rather than coupon swaps.
When interpreting the OPUC series, it is important to keep
in mind a few notes. First, the measure uses data on thirty-year
conventional xed-rate mortgage loans only and therefore
bears no direct information on other common types of loans,
such as een-year xed-rate mortgages, adjustable-rate
mortgages, Federal Housing Administration loans, or jumbos.
Second, since the measure uses survey rates/points and
average g-fees, our OPUC series is an average industry
measure rather than an originator-specic one. In addition,
rates and points may be subject to measurement error that
could distort the OPUC measure at high frequency, although
this should not have much eect on low-frequency trends.
ird, the measure is a lower bound to the actual industry
OPUCs, as it uses TBA prices to value loans, while originators
may have more protable options available. Indeed, as
noted in section 2, about 10 percent of conforming loans
are held on balance sheet, implying that originators nd it
more (or equally) protable not to securitize these loans.
In addition, a signicant fraction of agency loans is securitized
in specied MBS pools that trade at a premium, or pay-up,
to TBAs. In fact, the fraction of mortgages sold into the
non-TBA market appears to have increased substantially in
2012, relative to earlier years. Table 4 shows an estimate of
pools that are being issued as specied (“spec”) pools, rather
than TBA pools.
26
Over the rst ten months of 2012, only
about 60 percent (value-weighted) of all pools were issued to
be traded in the TBA market, while the rest were issued as
spec pools. e increase in spec-pool issuance is due in part
to Making House Aordable (MHA) loans originated under
the Home Aordable Renance Program (HARP), which
account for about 20 percent of all issuance and typically trade
25
Another dierence is that we take changes in points paid by borrowers into
account, but this matters relatively little (the average amount of points paid by
borrowers was relatively stable, between 0.4 and 0.8 over the period 2006-12).
26
We do not know with certainty whether a pool is ultimately traded in the
TBA market or as a specied pool; we simply assume that pools that strictly
adhere to certain specied pool criteria are also subsequently traded as such.
at signicant pay-ups to TBAs, owing to their lower expected
prepayment speeds. For example, over the second half of 2012,
Fannie 3.5 and 4 MHA pools with LTVs above 100 traded
on average about 1 1/2 and 3 1/2 points higher than
corresponding TBAs. Low-loan-balance pools, the second
largest spec-pool type, received similarly high pay-ups.
3.2 OPUCs, the Primary-Secondary Spread,
and Pass-rough
In assessing the extent to which secondary-market
movements pass through to mortgage loan rates, most
commentators focus on the primary-secondary spread—the
dierence between primary mortgage rates and the yield on
MBS securities implied by TBA prices. As shown in Chart 1,
the spread reached record-high levels over the course of
2012, suggesting that declines in primary mortgage rates
did not keep pace with those on secondary rates. For
example, while the primary-secondary spread averaged
73 basis points in 2011, the corresponding number was
113 basis points in 2012.
While the primary-secondary spread is a closely tracked
series, it is an imperfect measure of the pass-through between
secondary-market valuations and primary-market borrowing
costs for several reasons.
T
IssuanceofVariousGSEThirty-YearFixed-RatePool
Types,January–October2012
Pool Type
Balance
(Millions of Dollars)
Loan
Count
Balance
(Percent)
Count
(Percent)
TBA 379,763 1,347,516 59 46
MHA
a
124,779 559,180 20 19
Loan balance
b
97,161 867,628 15 30
Other specied
c
36,588 138,735 6 5
Tota l 638,292 2,913,059 100 100
Sources: Fannie Mae; Freddie Mac; 1010data; Amherst Securities.
Note: GSE is government-sponsored enterprise. TBA is “to-be-
announced.” MHA is the Making Home Aordable program.
a
Includes pools that are 100 percent re with 80<Orig LTV≤105, and
pools with loans >105 LTV.
b
Includes pools that contain only loans with balances less than or equal to
$175,000.
c
Includes 100 percent investor, NY, TX, PR, low FICO pools, and “mutt”
pools (variety of specied loan types). Excludes GSE pool types that are
jumbo, FH reinstated, co-op, FHA/VA, IO, relo, and assumable.
The higher valuation of loans in the
MBS market is the main driver of the
increase in OPUCs toward the end of
our sample period.
28 The Rising Gap
First, the yield on any MBS is not directly observable,
because the timing of cash ows depends on prepayments.
erefore, the calculation of the yield is based on the MBS
price and cash ow projections from a prepayment model,
which itself uses as inputs projections of conditioning
variables (for example, interest rates and house prices). In
addition, for TBA contracts, the projected cash ows and
the yield also depend on the characteristics of the assumed
cheapest-to-deliver pool. e resulting yield is thus subject
to errors due to model misspecication.
Second, the primary-secondary spread typically relies on
the theoretical construct of a “current coupon MBS.” The
current coupon is a hypothetical TBA security that trades
at par and has a yield meant to be representative of those
on newly issued securities.
27
Historically, this par contract
has usually fallen between two other actively traded TBA
coupons; however, in recent times, even the lowest coupon
with nontrivial issuance has generally traded significantly
above par (Chart 5). As a result, the current coupon rate
is obtained as an extrapolation from market prices, rather
than a less error-prone interpolation between two traded
27
An alternative is to calculate the yield on a particular security, which
may trade at a pay-up to the cheapest-to-deliver security. However, such a
calculation is still subject to other model misspecication and would not be
representative of the broad array of newly issued securities.
points.
28
Importantly, the impact of potential prepayment
model misspecification on yields is amplified when the
security trades significantly above (or below) par because
the yield on the security depends on the timing of the
amortization of the bond premium.
A better way to think about pass-through is to look
directly at what happens with the money paid by an
investor in the secondary market—does it go to borrowers,
originators, or the GSEs (either up front, or through
equivalent flow payments)? The purpose of the OPUC
measure is to track how many dollars (per $100 loan) get
absorbed by originators, either to cover costs other than
the g-fee, or as originator profits.
29
G-fees also contribute
to the overall cost of mortgage credit intermediation—
increasing these fees means that less money goes to
borrowers (or equivalently, that they need to pay a
higher rate). So, full pass-through of secondary-market
movements to borrowers would require OPUCs and g-fees
to remain constant (or, alternatively, a rise in g-fees would
need to be offset by a decrease in OPUCs).
In panel A of Chart 6, we conduct a counterfactual
exercise in which we compute a hypothetical survey note
rate during 2012, assuming that either the OPUCs only
(dark blue line), or both the OPUCs and the g-fee (light
blue line), had stayed at their average levels in 2011:Q4.
30
The comparison of the light blue line with the black line,
the actual realized mortgage rate, shows that had the cost
of mortgage intermediation stayed constant relative to
2011:Q4, mortgage rates during 2012 would at times have
been substantially lower, with a maximum gap between the
two rates of 55 basis points in early October 2012.
Comparing the black line with the dark blue line (holding
only OPUCs xed but letting g-fees increase), we note that
over most of 2012, much of the gap between the actual
and counterfactual rate derives from the rise in OPUCs.
28
Additionally, the current coupon is typically based on front-month contract
prices, while a more accurate measure would use back-month contracts,
because loans that rate-lock today are typically packaged into TBAs at least
two months forward.
29
It is important to keep in mind that changes in the secondary yield, even
if correctly measured, do not necessarily translate one-to-one into changes
in originator margins, which are determined by the TBA prices of dierent
coupons (which in turn determine optimal execution), and also by points
paid by the borrower. e primary-secondary spread, even net of g-fees, is
thus at best an imprecise measure of originator margins and protability.
30
e eective g-fee in our calculation for 2011:Q4 is 28.8 basis points,
which then increases to 38.9 basis points for the period January-March 2012
(as the announced increase eective April 1, 2012, is assumed to already be
relevant for loans originated at that point), 40.3 basis points for the period
April-June 2012, 41.8 basis points for July and August, and then increases
by another 10 basis points, to 51.8 basis points, for the rest of 2012 as the
December 1 g-fee increase becomes relevant to pricing.
Source: eMBS; JPMorgan Chase.
Notes: TBA is “to-be-announced.” “Sizable issuance” means that
the coupon accounts for at least 10 percent of total issuance in
that month.
C
Price of Lowest Fannie Mae TBA Thirty-Year
Coupon with Sizable Issuance
Monthly average price (in dollars)
92
94
96
98
100
102
104
106
1210080604022000
FRBNY Economic Policy Review / December 2013 29
Additionally, it is apparent that in times when rates are stable
or increasing, the counterfactual rate with constant OPUCs
tends to be close to the actual rate, and most of the gap
between the black and the light blue lines comes from the
higher g-fees (this is the case, for instance, toward the end
of the year). It is during times when rates fall (secondary-
market prices increase) that actual rates do not fall as much
as they would with constant OPUCs. As we discuss later, this
is consistent with originators having limited capacity, which
means they can keep rates relatively high and make extra
prots. at said, one should not necessarily interpret the
counterfactual rate series as indicating “where rates should
have been,” as this would require a judgment regarding
the “right” level of OPUCs. Here, we took the average over
2011:Q4 as our baseline, but if instead we took a lower value,
such as the average OPUCs over all of 2011, the dark blue and
light blue lines would be signicantly lower.
In panel B of Chart 6, we conduct a similar counterfactual
rate analysis, but using the primary-secondary spread as the
measure of the cost of mortgage intermediation. Holding
this spread (measured as the Freddie Mac survey rate minus
the Bloomberg current coupon yield) constant, we again get
a hypothetical mortgage rate under full pass-through. As
shown in panel B, while the overall pattern is similar to the
counterfactual rate with constant OPUCs and g-fees in panel A,
the series in panel B is more volatile, with the gap between the
counterfactual and actual rate spiking at 75 basis points in late
September 2012. is volatility of the counterfactual rate and
the presence of such large spikes illustrate the imperfect nature
of the primary-secondary spread as a pass-through measure.
4. P E
R C P
e rest of the article explores in more detail factors that may
have driven the observed increase in OPUCs over the period
2008-12. On the cost side, we focus on changes in pipeline
hedging costs, putback risk, and possible declines in the
valuation of mortgage servicing rights. We also briey discuss
changes in loan production expenses. On the prot side, we
focus on potential increases in originators’ pricing power due
to capacity constraints, industry concentration, or switching
costs for renancers.
4.1 Costs
Loan Putbacks
Originators pay g-fees to the GSEs as an insurance premium;
in exchange, the GSEs pay the principal and interest of the
loan in full to investors when the borrower is delinquent.
2.6
2.8
3.0
3.2
3.4
3.6
3.8
4.0
4.2
Fixed-rate mortgage (FRM)
rate with constant OPUCs
FRM rate with
constant OPUCs
and constant g-fee
C
Counterfactual Paths of Mortgage Rates over 2012
Actual
FRM rate
Percent
Panel A: Holding OPUCs and g-fees constant at 2011:Q4 averages
Panel B: Holding primary-secondary spread constant
at 2011:Q4 average
2.6
2.8
3.0
3.2
3.4
3.6
3.8
4.0
4.2
Q4Q3Q2Q1
FRM rate with constant
primary-secondary spread
Sources: Bloomberg L.P.; Freddie Mac; authors’ calculations.
Note: OPUCs are originator prots and unmeasured costs.
Actual
FRM rate
Over most of 2012, much of the gap
between the actual and counterfactual
rate derives from the rise in OPUCs.
30 The Rising Gap
However, mortgage originators or servicers are obligated to
repurchase nonperforming or defaulted loans under certain
conditions, for example, when the GSEs establish that the loan
did not meet their original underwriting or eligibility require-
ments, that is, if the loan representations and warranties are
awed.
31
e repurchase requests have increased rapidly since
the 2008 nancial crisis and have been the source of disputes
between originators and GSEs. e increased risk to origina-
tors that the loan may ultimately be put back to them has been
cited as a source of higher costs and thus OPUCs.
How can we assess the magnitude of the contribution of
putback costs to OPUCs? To do so, one needs to imagine
a stress scenario—not a modal one—with a corresponding
default rate, and then assume fractions of putback attempts
by the GSEs, putback success, and loss-given-defaults for
servicers/lenders forced to repurchase the delinquent loan.
To construct a ballpark estimate of the possible putback
cost on new loans, we start from the experience of agency
loans originated during the period 2005-08. Based on a
random 20 percent sample of conventional rst-lien xed-rate
loans originated during that period in the servicing data set
of LPS Applied Analytics, we nd that about 16.5 percent of
GSE-securitized mortgages (value-weighted) have become
sixty-or-more days delinquent at least once, and 11.5 percent
of them have ended in foreclosure.
32
Importantly, these
vintages include a substantial population of borrowers with
relatively low FICO scores, undocumented income or assets,
or a combination of these factors. For instance, the median
FICO score was around 735, while the 25th percentile
was at 690. In 2012, however, the corresponding values on
non-HARP loans were around 770 and 735, respectively.
33
31
It is also possible that originators need to repurchase incorrectly
underwritten loans prior to a loan becoming delinquent. However, the
repurchase of nondelinquent loans is likely less costly to originators. e
rest of this section therefore focuses on repurchases of delinquent loans.
32
ese statistics are as of November 2012.
33
Origination LTVs have not changed as dramatically: in 2012, approximately
16 percent of non-HARP loans had an LTV at origination above 80; this is only
slightly lower than during the period 2005-08. However, the fraction of loans
with second liens was likely higher during the boom period. Also, in 2012 there
are no non-HARP Freddie Mac loans with incomplete documentation (this is
not disclosed in the Fannie Mae data, but is likely similar).
To account for the tighter underwriting standards on new
loans, we focus on the performance of GSE-securitized loans
from the 2005-08 vintages with origination FICO of at least
720 and full documentation. Among those, “only” about
8.8 percent have become sixty-or-more days delinquent, and
5.5 percent have ended in foreclosure. us, because of today’s
more stringent underwriting guidelines for agency loans, our
expectation in a stress scenario would be for delinquencies,
and hence potential putbacks, to be roughly half as large,
relative to those experienced by the 2005-08 vintages. Further-
more, we would expect the frequency of putback attempts to
be roughly half as large for loans with full documentation as
for the overall population of delinquent loans.
We obtain an estimate of the fraction of loans that the
GSEs could attempt to force the lender to repurchase from
Fannie Mae’s 2012:Q3 Form 10-Q, which states (on page 72)
that as of 2012:Q3, about 3 percent of loans from the 2005-08
vintages have been subject to repurchase requests (compared
with only 0.25 percent of loans originated aer 2008). us,
given that repurchase requests are issued primarily conditional
on a delinquency, we would anticipate repurchase requests
in a stress scenario to be about one-quarter (0.5 delinquency
rate × 0.5 putback rate) as high as those recorded on the
2005-08 vintage, or about 0.75 percent.
34
Based on repurchase disclosure data collected from the
GSEs,
35
it appears that about 50 percent of requests ultimately
lead to buybacks of the loan. Furthermore, if we assume a
50 percent loss-given-default (which seems on the high side),
this would generate an expected loss to the lender/servicer of:
0.75 percent × 0.5 × 0.5 = 19 basis points
is estimate, which we think of as being conservative
(given the unlikely repetition at this point of large house
price declines experienced by the 2005-08 vintages), would
imply a putback cost of 19 cents per $100 loan. is cost is
modest relative to the widening in OPUCs experienced over
the period 2008-12.
36
at said, perhaps the “true” cost of
putback risk comes from originators trying to avoid putbacks
in the rst place by spending signicantly more resources
on underwriting new loans or on defending against putback
34
Without the assumption that full-documentation loans are less likely to
be put back, the expected putback rate would be 1.5 percent, resulting in an
expected loss of 37.5 basis points.
35
Source: Inside Mortgage Finance.
36
Furthermore, the FHFA introduced a new representation and warrant
framework for loans delivered to the GSEs aer January 2013 that relieves
lenders of repurchase exposure under certain conditions (for example, if the
loan was current for three years). is policy change should further reduce
the expected putback cost going forward.
The increased risk to originators that the
loan may ultimately be put back to them
has been cited as a source of higher
costs and thus OPUCs.
FRBNY Economic Policy Review / December 2013 31
claims. Furthermore, the remaining risk on older vintages is
larger than on new loans, and many active lenders are also
still subject to lawsuits on nonagency loans made during the
boom. It is unclear, however, why these claims on vintage
loans should aect the cost of new originations.
Mortgage Servicing Rights Values
e baseline OPUC calculation assumes constant servicing
multiples throughout the sample of 5x for base servicing
and 4x for excess servicing ows. While these are commonly
assumed levels, according to market reports, mortgage
servicing right (MSR) valuations have declined over the past
few years. In this section, we study the sensitivity of OPUCs
to alternative multiple assumptions.
We obtain a time series of normal (or base) servicing
multiples for production agency MBS coupons from the
company Mortgage Industry Advisory Corporation (MIAC).
37
ese multiples declined from about 5x in early 2008 to about
3.25x in November 2012.
38
To evaluate the impact on OPUCs,
we repeat our earlier calculation using the MIAC base multi-
ples.
39
e results are shown in Chart 7. Comparing the black
(baseline) and dark blue (MIAC) lines, we see that the lower
multiple values reduce OPUCs by about sixty cents at the end
of 2012, a somewhat signicant impact.
Some commentators have attributed the decline in
multiples to a new regulatory treatment of MSRs under the
2010 Basel III accord. While the three U.S. federal banking
regulatory agencies released notices of proposed rulemaking
to implement the accord on June 12, 2012, the introduction
of the new rules, originally set for January 2013, has been
postponed. Under the June 2012 proposal, concentrated
MSR investment will be penalized and will generally receive
a higher risk weighting.
40
e long phase-in period for
37
ese multiples come from MIAC’s “Generic Servicing Assets” portfolio
and are based on transaction values of brokered bulk MSR deals, surveys of
market participants, and a pricing model.
38
Key drivers of servicing right valuations are expected mortgage
prepayments—lower interest rates mean a higher likelihood that the servicing
ow will stop due to an early principal payment—and, in the case of base
servicing, varying operating costs in servicing the loan, for example, when
loans become delinquent. Another important component is the magnitude of
the oat interest income earned, for instance, on escrow accounts.
39
We assume a 20 percent discount for excess servicing and keep the g-fee
buy-down multiple unchanged at 7x. Also, as our MIAC series ends in
November 2012, we assume that the multiple in December is identical to
that in November.
40
MSRs will be computed toward Tier 1 equity only up to 10 percent of their
value, and risk-weighted at 250 percent, with the rest being deducted from
Tier 1 equity. is treatment is signicantly more stringent than the status
quo that risk-weights the MSRs at 100 percent and limits MSRs to 50 percent
of Tier 1 capital of banks (100 percent for savings and loans).
these rules makes it unclear how much the expected tighter
regulatory treatment is already aecting MSR multiples.
Nonetheless, in order to assess an upper-bound impact
on OPUCs, we consider here a more stressed scenario
than implied by the MIAC multiples. In this scenario, our
baseline multiples are halved starting (for simplicity) with
the disclosure by the Basel Committee of the capital rules in
July 2010.
41
e resulting eight-week-rolling OPUC series is
also depicted in Chart 7. As shown in the chart, following a
halving of the MSR multiples, the implied OPUC declines are
signicant, but still not sucient to explain the historically
high OPUC levels in 2012.
We conclude that lower multiples, while having a sizable
impact on OPUCs, can only partially oset their increase
over the past few years.
41
In this alternative scenario, base servicing is now valued at 2.5x, while
excess servicing is valued at 2x. (e GSE buy-down multiple is assumed to
stay at 7x.) e optimal execution in this exercise again takes into account the
lower levels of the multiples.
1
2
3
4
5
20122011201020092008
Chart 7
Sensitivity of OPUCs to Alternative Assumptions
about Mortgage Servicing Right Multiples
Dollars per $100 loan
Sources: JPMorgan Chase; Freddie Mac; Fannie Mae; MIAC; authors’
calculations.
Notes: The data reflect an eight-week rolling window. MIAC is the
Mortgage Industry Advisory Corporation.
1
/2 multiples
MIAC
multiples
Baseline
multiples
32 The Rising Gap
Pipeline Hedging Costs
For loans that are securitized in MBS, the “mortgage pipeline”
is the channel through which an originator’s loan commit-
ment, or rate-lock, is ultimately delivered into a security or
terminated with a denial or withdrawal of the application. e
originators’ commitment starts with a rate-lock that typically
ranges between thirty and ninety days. is time window
appears to have increased signicantly in recent years. For
example, the time from application to funding for renancing
applications increased from about thirty days in late 2008
to more than y days in late 2012 (as shown graphically in
section 4.2 below).
Originators face two sources of risk while the loan is in
the pipeline: changes in the prospective value of the loan due
to interest rate uctuations and movements in the fraction
of rate-locks that do not ultimately lead to loan originations,
referred to as “fallouts.”
e rst risk—potential changes in the value of the loan
due to interest rate movements—can be hedged by selling
TBA contracts: at the time of the loan commitment, origina-
tors who are long a mortgage loan at the time of the rate-lock
can oset the position by selling the yet-to-be-originated
loan forward in the TBA market. e calculation in section 3
already takes into account these hedging costs: when comput-
ing the OPUC measure, we use the back-month TBA contract
price that settles on average about forty-ve days following
the transaction. To the extent that originators may have been
able to sell into the front-month TBA market when the length
of the pipeline was shorter, our calculations may understate
OPUCs for earlier years by the price dierence, or “drop,”
between the two contract prices. Yet, this drop is typically
only about 20 basis points in price space. We conclude that
the lengthening of the pipeline does not appear to have had a
signicant economic impact on the cost of price hedging, and
thus the rise in OPUCs experienced over the period 2008-12.
e second risk is due to movements in the fallout rate.
As discussed in section 2, borrowers’ terminations may occur
involuntarily (if they do not ultimately qualify for the loan or rate
oer) or voluntarily. Except for changes in lending standards and
house prices, uctuations in involuntary terminations are largely
driven by idiosyncratic factors that are diversied for originators
with large-enough portfolios. Movements in voluntary
terminations, on the other hand, are mostly due to primary rate
dynamics: following the initial rate-lock, mortgage rates may
fall, prompting borrowers to pursue a lower rate loan with either
the same or a dierent lender. Common ways to hedge this risk
are to dynamically delta-hedge the position using TBAs, using
mortgage options or swap options, or a combination of these
(or other) strategies.
42
To illustrate, we now consider a hedging
example using at-the-money swaptions to gauge the magnitude
and time-series pattern of the interest rate hedging cost.
Based on market reports and data from the Mortgage
Bankers Association (MBA), normal fallout rates average
about 30 percent, and we assume that an originator hedges
as much using swaptions. Chart 8 shows the price premium
in basis points for swaptions on a ve-year swap rate with
expirations of one and three months. Conditional on a
30 percent hedging strategy, the cost of protection, when
using a three-month expiration, would be about 0.3 x 40 basis
points = 12 basis points, or a 12 cent impact on OPUCs. e
extension in the length of the pipeline, which may have led
originators to go from one-month to three-month expiration,
also had a rather small impact on OPUCs.
42
Correspondent lenders, or small lenders that sell whole loans to the GSEs,
can manage the fallout risk by entering into “best-eort” locks with the buyer
of the loan. Under this arrangement, the originator does not need to pay a ne
for not delivering a mortgage that does not close, unlike under “mandatory
delivery.” To compensate, the price oered by the buyer of the loan is lower.
us, in a sense, “best-eort” commitments allow (small) originators to
“outsource” the hedging of fallout risk.
0
25
50
75
100
125
150
175
200
225
12100806042003
Chart 8
Swaption Price Premia
Basis points
Source: JPMorgan Chase.
Three-month,
five-year
One-month,
five-year
Originators face two sources of risk while
the loan is in the pipeline: changes in
the prospective value of the loan due to
interest rate uctuations and movements
in the fraction of rate-locks that do not
ultimately lead to loan originations,
referred to as “fallouts.”
FRBNY Economic Policy Review / December 2013 33
More generally and beyond our specic example, implied
volatility and option price premia have declined signicantly
since the fall of 2008, reecting the lower rate volatility
environment. While we do not explicitly consider other, more
complex hedging strategies, the lower volatility environment has
likely also lowered the cost of these strategies. is is in contrast
with the rise in OPUCs over this period. In sum, changing
hedging costs does not appear to account for a signicant
portion of the rise in OPUCs, and at least the cost of hedging
fallout risk may in fact have declined during the period 2009-12.
Other Loan Production Expenses
A nal possible cost-side explanation for the increase in
OPUCs is that other loan production expenses, including
costs related to the underwriting of loans and to nding
borrowers (sales commissions, advertising, and so on) have
increased substantially over the past few years. While it
is dicult to obtain a variable loan cost series that can be
easily mapped into the OPUC measure, the MBA collects
in its Quarterly Mortgage Bankers Performance Report
survey information on total loan production expenses that
include both xed and variable costs, such as commissions,
compensation, occupancy and equipment, and other
production expenses and corporate allocations. With the
caveat that the sample of respondents is composed of small-
and medium-sized independent mortgage companies, the
data indicate a modest increase in loan production expenses
over the past few years and a fairly stable pattern of these
expenses. For example, total loan production expenses
averaged $4,717 per loan in 2008, and $5,163 per loan in
2012:Q3.
43
is modest increase appears unlikely to explain
the more than doubling in OPUCs over the period 2008-12.
4.2 Industry Dynamics and Originators’
Prots
e discussion in the previous subsection appears to indicate
that the higher OPUCs on regular agency-securitized loans
over the period 2008-12 were not likely driven exclusively, or
even mostly, by increases in costs. As a result, the rise in OPUCs
during this time could reect an increase in prots. If so, what
are the potential driving forces behind such an increase?
43
Source: Mortgage Bankers Association, Press Release Performance Report,
various issues. e numbers cited are gross expenses, not including any
revenue such as loan origination fees or other underwriting, processing, or
administrative fees.
Capacity Constraints
An oen-made argument is that capacity constraints in the
mortgage origination business have been particularly tight in
recent years, and that these constraints become binding when
the application volume increases signicantly, usually due to
a renancing wave. As a result, originators do not lower rates
as much as they would without these constraints, in order to
curb the excess ow of applications.
Chart 9 provides some long-horizon evidence on the
potential importance of capacity constraints for prots, by
plotting our OPUC measure against the MBA application
index (including both purchase and renancing applications).
e chart shows that the two series correlate quite strongly:
Whenever the MBA application index increases, OPUCs tend
to increase, and vice-versa.
44
is correlation suggests that capacity constraints play an
important role in generating the higher OPUCs. at said,
mortgage applications (and other measures of demand and
origination activity, such as MBS issuance) were at higher levels
in the past, without OPUCs being as high as they were in 2012.
Chart 10 shows some more direct evidence on the potential
importance of capacity constraints, by depicting the number
of days it takes from the initiation of a renancing application
to the funding of the loan. e chart is based on data from the
44
Over the period 2004-08, the relationship between the two series appears
weaker than elsewhere—OPUCs appear to be on a downward trend over
much of that time, even when applications increase.
C
Originator Prots and Unmeasured Costs (OPUCs)
and MBA Application Index
Dollars per $100 loan
Index level
1
2
3
4
5
0
500
1000
1500
2000
1210080604020098961994
OPUCs
Left scale
MBA market volume
index (all applications)
Right scale
Sources: JPMorgan Chase; Freddie Mac; Fannie Mae; Mortgage
Bankers Association (MBA); authors’ calculations.
Note: e lines reect eight-week rolling window averages.
34 The Rising Gap
Home Mortgage Disclosure Act (HMDA), which was available
only through 2011 at the time of this writing, and from the
Ellie Mae Origination Insight Report, which is only available
since August 2011.
45
It shows that the median (HMDA) or
average (Ellie Mae) number of days it takes for an application
to be processed and funded has been substantially higher since
2009 than it was in prior years.
46
e processing time moves
in response to the MBA application volume shown earlier; for
instance, it reached its maximum aer the renancing wave of
early 2009 and increased from less than forty days in mid-2011
to more than y-ve days by October 2012, as renancing
accelerated over this period. However, to the extent that the
HMDA and Ellie Mae data are comparable, it does not appear
that it took substantially longer to originate a renancing loan
in 2012 than it did in early 2009, making it dicult to explain
the full rise in OPUCs through capacity constraints.
47
A nal interesting question is how rigid capacity
constraints may be. Current originators can add sta, but it
45
See www.elliemae.com/origination-insight-reports/
EMOriginationInsightReportDecember2012.pdf.
46
e average for HMDA would be higher than the median, but would show
similar patterns.
47
It is interesting to note that the time from renancing application to funding
was signicantly lower in 2003, even though application volume was much
higher than it was over 2008-12. is is likely driven by tighter underwriting in
the recent period compared with during the 2003 renancing boom.
takes time to train new hires. New originators can enter the
market, but entry requires federal and/or state licensing and
approval from Fannie Mae, Freddie Mac, and Ginnie Mae to
fully participate in the origination process. To the extent that
training may take longer than in the past, or that approval
delays for new entrants are longer (as anecdotally reported),
the speed of capacity expansion may have declined compared
with earlier episodes.
48
Another potentially important factor
is that the share of third-party originations (by brokers or
correspondent lenders) has decreased signicantly in recent
years (as discussed in footnote 5). ird-party originators
may, in the past, have acted as a rapid way to adjust capacity,
especially during renancing waves. In sum, while capacity
constraints likely contributed to the rise in OPUCs in recent
years, it is unlikely that they were the only source of this rise.
Market Concentration
A second popular explanation for the higher prots in the
mortgage origination business is that the market is highly
concentrated. It is well known that the mortgage market in
the United States is dominated by a relatively small number
of large banks that originate the majority of loans. However,
as shown in Chart 11, a simple measure of market concentra-
tion given by the share of loans made by the largest ve or ten
originators actually decreased over the period 2011-12, as a
number of the large players reduced their market share. us,
overall market concentration alone seems unlikely to explain
high prots in the mortgage business. is would make sense
from a theoretical point of view: ere is no particular reason
why a concentrated market (but with a large number of fringe
players, and price competition) should incur large prots.
Recent work by Scharfstein and Sunderam (2013) comes
to a dierent conclusion. e authors argue that looking at
national market concentration may mask dierential trends in
local market concentration, which matters if borrowers shop
locally for their mortgages. Using data from 1994 to 2011, the
authors nd that higher concentration at the county level is
48
Additionally, existing capacity may have been diverted to defending against
putbacks instead of new loan origination.
Overall market concentration alone
seems unlikely to explain high prots
in the mortgage business.
20
30
40
50
60
70
1211100908070605042003
Chart 10
Time from Refinancing Application to Funding
(by Month in Which a Loan Is Funded)
Number of days
Sources: HMDA (January 2003 to December 2011); Ellie Mae
(August 2011 to December 2012).
Notes: HMDA is the Home Mortgage Disclosure Act. HMDA data
are restricted to first-lien mortgages for owner-occupants of
one-to-four-unit houses or condos.
Ellie Mae
(average)
HMDA
(median)
FRBNY Economic Policy Review / December 2013 35
correlated with a lower sensitivity of renancing and mortgage
rates to MBS yields. It would be interesting to extend their
analysis to 2012 to see whether their ndings can help explain
the increase in OPUCs in that year.
We next turn to an alternative explanation for why origi-
nators could make larger prots than in the past, namely that
they may enjoy more pricing power on some of their borrow-
ers for reasons unrelated to concentration.
HARP Renance Loans
A market segment where such pricing power may have
been particularly important is the high-LTV segment,
which over the past years has been dominated by
refinancings through HARP, originally introduced in
March 2009. The introduction of revised HARP rules
in late 2011, often referred to as “HARP 2.0,” led to a
significant increase in HARP activity during 2012; the
FHFA estimates that in the second and third quarters of
2012, HARP refinancings accounted for about 26 percent
of total refinance volume.
49
HARP 2.0 provides significant
incentives for same-servicer refinancing (namely, relief
from representations and warranties) that are not present
to the same extent for different-servicer refinancings.
Furthermore, even under identical representation and
warranty conditions, a new servicer may be less willing
to add high-LTV borrowers to its servicing book, because
such borrowers have a higher likelihood of delinquency,
49
See http://www.fa.gov/webles/24967/Nov2012ReReport.pdf.
which makes servicing high-LTV loans more expensive.
For these two reasons, many servicers do not offer HARP
refinancing for loans that they are not currently servicing,
or only at much worse terms. The result is that the current
servicer has significant pricing power over its own high-
LTV borrowers looking to refinance.
Is there evidence that lenders can exploit this higher pricing
power? e observed note rates for HARP-renanced loans are
at least consistent with this idea. As shown in Chart 12, during
2012 the weighted average coupons (WACs; that is, the loan
note rates) on HARP loans with LTVs above 105 tended to be
40-50 basis points higher than those of regular renancing or
purchase loans.
50
Banks earn higher revenues on these HARP
loans than on regular loans for two reasons: given the higher
note rate, they will typically sell these loans into a pool with
a 50-basis-point higher coupon, which usually commands a
price premium of around 1.5-2.0 points. Furthermore, thanks
to the prepayment protection oered by these pools (as a
borrower can only renance through HARP once), investors
are willing to pay a higher price (in the spec-pool market) than
for TBA pools; this can add another 1-3 points (depending on
the coupon) to the originator’s revenue.
50
We can also compare WACs on renancings with LTV between 80 and 95 that
are likely to be HARP loans (based on mortgage insurance information) with
other loans in the same LTV range that are likely non-HARP loans. On average,
the WAC on HARP loans was about 15-20 basis points higher in that range.
3.0
3.5
4.0
4.5
5.0
Refi LTV > 125 (HARP)
Refi 105 < LTV
≤ 125 (HARP)
Refi LTV ≤ 80
Purchase LTV ≤ 80
Nov. Sept. Jul.May Mar. Jan.
Chart 12
Weighted Average Coupons of Different
Loan Types
Percent
2012
Sources: Fannie Mae; Freddie Mac; eMBS.
Note: The data include thirty-year fixed-rate mortgages with loan
amounts less than or equal to $417,000, made to borrowers with a
FICO score of at least 720, on owner-occupied one-unit properites.
0
20
40
60
80
100
Top five
12111009080706052004
Chart 11
Origination Market Concentration
Market share
Source: Inside Mortgage Finance.
Top one
Top ten lenders
36 The Rising Gap
Are these higher revenues compensation for higher
origination costs for HARP loans? is seems unlikely, as
the documentation requirements for HARP loans are in fact
signicantly lighter than for regular loans. us, it is likely
that origination costs are lower, not higher, for HARP loans
relative to regular renancings.
51
Another possibility is that high-LTV borrowers are more cash
constrained than regular renancers and thus require higher
rebates (negative points) at origination to help cover their closing
costs. While this is a possibility, it is unlikely that the dierence
can oset a signicant portion of the additional revenues,
especially since closing costs are likely lower than they are for
regular loans (thanks, for example, to appraisal waivers).
52
Finally, for reasons discussed above, the value of base
servicing on HARP loans may be signicantly lower than that
for non-HARP loans with lower LTVs. Even if we assume that
the multiple on base servicing drops from 5x to 0x, however,
this would only account for 1.25 points, while, as noted above,
revenues are 2.5-5.0 points higher. Furthermore, because
HARP borrowers are expected to prepay slowly, the cash ow
stream from servicing is in fact more valuable than for regular
loans, osetting part of the higher servicing cost. Also, the
expected servicing cost for current servicers declines when
loans are renanced under HARP, as borrowers are less likely
to default aer the note rate declines (see Tracy and Wright
[2012] and Zhu [2012]).
us, the evidence strongly suggests that originators have
been making larger prots on HARP loans than on regular
loans, by being able to exploit their pricing power.
Non-HARP Mortgages
e next question is whether similar pricing power could have
contributed to the rise in our OPUCs on regular (non-HARP)
loans that seems not fully explained by capacity constraints,
as discussed above. While lenders may have pricing power
over their HARP borrowers, it is much less clear whether
such pricing power may also exist for “regular” loans. Pricing
power could arise, for instance, from customers’ impediments
(actual or perceived) to shop around, an unwillingness of
other rms to compete, barriers to entry for new competitors,
or a combination of these. Directly measuring originators’
51
Also, the loans with FICO scores of 720 or above that we include in the
chart are not subject to loan-level price adjustments under HARP.
52
Related to this point, it is not the case that HARP note rates are higher
because principal amounts are lower than for regular renancings (as the same
xed closing cost being rolled into the rate will require a larger rate increase for
lower principal amounts); controlling for loan amount in a regression basically
leaves the estimated dierences across loan categories unchanged.
pricing power is not a trivial task, and we do not attempt a
full analysis here. However, looking at some cross-sectional
patterns may suggest some insights.
Chart 12 shows that over 2012, the WAC on non-HARP
renancing loans tended to be slightly larger than it was on
purchase loans. is is somewhat surprising if one thinks that
the costs of originating a renance loan are likely lower than
those of a purchase loan. In addition, comparing WACs over a
longer time period (not shown), it is the case that the positive
gap in WACs between purchase and renancing loans only
started emerging in 2010 (and has remained there since); over
the period 2005-09, average monthly WACs on renancing
loans were mostly either equal to or below those on purchase
loans.
53
However, the WAC divergence could potentially be
explained by purchase borrowers paying more points than
renancers; this could be, for instance, because they expect to
stay in the mortgage longer or because of tax incentives.
54
One would expect this explanation, if true, to hold across
all lenders. However, looking at lender-specic dierences in
WACs reveals a large variation across lenders. e two panels
of Chart 13 show the monthly average WAC for the sixteen
largest lenders over 2012 (in terms of number of loans sold
to the GSEs), for purchase and renancing loans separately.
We also plot separately the average for all other (smaller)
sellers (the thicker lines). We include only thirty-year xed-
rate loans with FICO scores of 720 and higher, and LTVs of
80 or lower, made to single-unit owner-occupiers in order to
reduce potential disparities due to dierential LLPAs.
55
53
is statement is based on loan-level data from Freddie Mac only, as the
Fannie Mae data only became available in 2012.
54
Points paid in cash are fully tax deductible for purchase mortgages in the
year the loan is closed. For renancing mortgages, the deduction is instead
spread evenly over the term of the mortgage (for example, thirty years),
except if the loan is paid o early, in which case all unused deductions can
be taken in the year the loan is paid o. See, for example, www.irs.gov/
publications/p936/ar02.html#en_US_2011_publink1000229936.
55
ese calculations are based on the complete set of loan-level disclosures
for pools issued in 2012 by Fannie Mae and Freddie Mac.
The evidence strongly suggests that
originators have been making larger
prots on HARP loans than on regular
loans, by being able to exploit their
pricing power.
FRBNY Economic Policy Review / December 2013 37
Panel A of the chart shows that purchase WACs across
sellers were quite homogeneous—with the exception of a
couple of outliers, most lender WACs lie within a range of
approximately 10 basis points. is is consistent with the idea
that the purchase mortgage market is quite competitive, as
presumably many borrowers shop around (perhaps with the
help of their realtor).
Panel B reveals a much larger dispersion for renancing
loans. In particular, while a number of sellers remain con-
centrated around the thicker line representing the average of
smaller players, eight of these large lenders sold loans with
WACs that are 15 basis points or more above the thick line
in at least one month, and, for six of them, that is the case for
at least six out of twelve months.
56
In principle, this observed
price dispersion is certainly not inconsistent with the market
being competitive; however, under this null hypothesis, it is
surprising that the dispersion is so much larger for renancing
loans than for purchases.
As discussed above, during 2012 the HARP program
gained signicant momentum for high-LTV renances. A
perhaps lesser-known fact is that there exist GSE streamline
renancing programs also for non-HARP loans (with LTV
less than 80), with the same cuto date for eligible mortgages
(which must have been delivered to one of the GSEs prior to
May 31, 2009). Streamlined renancing, when done through
the institution that currently services the loan, relieves the
lender from representation and warranties relating to the
borrower’s creditworthiness and home value, while a dierent-
servicer renancing requires more extensive underwriting
of the new loan. As a consequence, for borrowers eligible for
a streamlined renancing, there is an advantage to staying
with the same servicer/lender, as doing so will reduce the
documentation the borrower is required to submit. is,
in turn, again creates some pricing power for the current
servicer (although likely less so than for high-LTV loans).
e population of loans in xed-rate GSE pools originated
prior to June 2009 is large: As of December 2012, about
$1.1 trillion of loans were in such pools, relative to an
overall Fannie Mae/Freddie Mac xed-rate universe of about
$3.8 trillion. During 2012, about 52 percent of all prepayments
came from pools issued prior to June 2009.
57
erefore, if
lenders have pricing power over the renancings of these
loans, this could be a nontrivial contributor to OPUCs.
Is there evidence that such pricing power could explain
the dispersion in renancing WACs? Unfortunately, unlike
for HARP loans, there is no way for us to observe in the data
whether a renancing was streamlined or not. However, we
can look at variation across lenders in the fraction of their
servicing portfolio that could potentially be renanced in
a streamlined manner (that is, loans in pools issued prior
to June 2009) and correlate this gure with the average
WAC of the lenders’ non-HARP renance loans over 2012.
Chart 14 shows that there is indeed a positive correlation
between the two: e lenders that had a large fraction
of potentially streamline-eligible loans in their servicing
56
With the exception of one of these six lenders, the monthly number of sales
of renancing loans always exceeds 500 loans, meaning that these averages
are unlikely to be driven by small-sample noise. Additionally, as above, the
result of large WAC dispersion across lenders for renance loans remains
basically unchanged if loan characteristics such as loan amount are added as
explanatory variables in a regression framework.
57
ese prepayments include renancings as well as the loan simply getting
paid o (for instance, due to the borrower moving).
3.25
3.50
3.75
4.00
4.25
4.50
4.75
Chart 13
Dispersion in Weighted Average Coupons across
Sellers to Fannie Mae and Freddie Mac, 2012
Percent
Panel A: Purchase loans
Sources: Fannie Mae; Freddie Mac; eMBS.
Note: The data include loans with a FICO score of 720 or higher,
an LTV of 80 or lower, an amount less than or equal to $417,000,
on owner-occupied single-unit properties, and only for months in
which a seller made at least 100 sales.
Panel B: Refinancing loans
3.25
3.50
3.75
4.00
4.25
4.50
4.75
Nov.Aug.JulyMayMar.Jan.
Average of
smaller sellers
Average of
smaller sellers
38 The Rising Gap
portfolio at the end of 2011 tend to be those that originated
renance loans with the highest WACs on average over
2012 (that is, those that are above the thick line in panel B
of Chart 13). is result is consistent with (though certainly
not proof of) originators taking advantage of their pricing
power over streamline-eligible borrowers.
5. C
e widening gap between primary and secondary mortgage
rates over the period 2008 to 2012 was due to a rise in orig-
inators’ prots and unmeasured costs, or OPUCs, as well as
increases in g-fees. e magnitude of the OPUCs is inuenced
by MBS prices, the valuation of servicing rights, points paid
by borrowers, and costs such as those from loan putbacks and
pipeline hedging.
e rise in OPUCs was mainly driven by higher MBS
prices, which were not oset by corresponding increases
in measurable costs. Conversely, a decline in the value of
mortgage servicing rights may have reduced OPUCs to
some extent, and thus contributed to the widening primary-
secondary spread. Among harder-to-measure costs, we nd
that expected putback costs and pipeline hedging likely did
not cause a signicant portion of the rise in OPUCs. Absent
increases in other costs that we cannot measure well, such as
operating costs, the rise in OPUCs reected an increase in
originator prots. While market concentration alone does not
seem to explain the rise in these prots, capacity constraints
do appear to have played a signicant role. We also provide
evidence suggesting that originators have enjoyed pricing
power on some of their borrowers looking to renance, due to
borrowers’ switching costs.
Going forward, it will be interesting to study the extent to
which interest rate dynamics, capacity expansions, new entry,
changes in regulations, and (in the longer term) housing nance
reform will aect the pass-through from secondary to primary
markets. As illustrated in this article, a number of factors deter-
mine this pass-through, and it will therefore be important for
policymakers and market participants alike to further improve
the measurement and understanding of these factors.
Share of November 2011 servicing portfolio
in HARP- or streamline-eligible pools
Chart 14
Weighted Average Coupons on Regular (Low LTV)
Refinance Loans Against Fraction of Servicers’
Portfolio Eligible for Streamline Refinancing
Weighted average coupons of non-HARP refis relative to smaller
sellers over 2012 (averaged across months, in percent)
0.30.4 0.50.6 0.7
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
Sources: Fannie Mae; Freddie Mac; eMBS.
Notes: HARP- or streamline-eligible pools are pools issued prior to
June 2009. The data include only sellers/services with servicing
portfolios with more than $1 billion of HARP- or streamline-eligible
pools in November 2011. Non-HARP weighted-average coupons are
calculated on loans with a FICO score of 720 or higher, an LTV of
80 or lower, an amount less than or equal to $417,000, on owner-
occupied single-unit properties.
FRBNY Economic Policy Review / December 2013 39
R
e views expressed are those of the author and do not necessarily reect the position of the Federal Reserve Bank of New York, the
Federal Reserve Bank of Boston, or the Federal Reserve System. e Federal Reserve Bank of New York provides no warranty, express
or implied, as to the accuracy, timeliness, completeness, merchantability, or tness for any particular purpose of any information
contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.
Bhattacharya, A. K., W. S. Berliner, and F. J. Fabozzi. 2008. “e
Interaction of MBS Markets and Primary Mortgage Rates.”
J S F 14, no. 3 (fall): 16-36.
Scharfstein, D. S., and A. Sunderam. 2013. “Concentration in
Mortgage Lending, Renancing Activity, and Mortgage Rates.”
NBER Working Paper no. 19156, June.
Tracy, J., and J. Wright. 2012. “Payment Changes and Default Risk:
e Impact of Renancing on Expected Credit Losses.” Federal
Reserve Bank of New York S R, no. 562, June.
Vickery, J., and J. Wright. 2013. “TBA Trading and Liquidity in
the Agency MBS Market.” Federal Reserve Bank of New York
E P R 19, no. 1 (May): 1-19.
Zhu, J. 2012. “Renance and Mortgage Default: An Empirical
Analysis of the HARP’s Impact on Default Rates.” Unpublished
paper, Federal Home Loan Mortgage Corporation. Available at
www.ssrn.com/abstract=2184514.
