spring 2010 volume16,number 1 · pdf fileactually at a coupon that is much higher than average...

8
Spring 2010 Volume 16, Number 1 www.iijsf.com The Voices of Influence | iijournals.com

Upload: duonghanh

Post on 09-Mar-2018

217 views

Category:

Documents


3 download

TRANSCRIPT

Page 1: Spring 2010 Volume16,Number 1 · PDF fileactually at a coupon that is much higher than average mortgage rates, but are unable to refinance because of a number of factors ... 1 illustrates

Spring 2010 Volume 16, Number 1www.iijsf.com

The Voices of Influence | iijournals.com

Page 2: Spring 2010 Volume16,Number 1 · PDF fileactually at a coupon that is much higher than average mortgage rates, but are unable to refinance because of a number of factors ... 1 illustrates

THE JOURNAL OF STRUCTURED FINANCE SPRING 2010

Predicting Agency Prepayments in the Current Market Environment: Why Yesterday’s Predictive Models Won’t Hold Water TodayJONATHON WEINER, WESLEY WINTER, AND KYLE G. LUNDSTEDT

JONATHON WEINER

is a vice president of research and development at LPS Applied Analytics in San Francisco, [email protected]

WESLEY WINTER

is a senior analyst at LPS Applied Analytics in San Francisco, [email protected]

KYLE G. LUNDSTEDT

is a managing director at LPS Applied Analytics in San Francisco, [email protected]

The use of predictive models to evaluate mortgage-backed secu-rities (MBS) is a long-established practice in the investment world.

Historically, these models have sought to pre-dict the fraction of the loan pool making up a security that prepays every month. Prepay-ment, for the purposes of this article, covers any fraction of the pool that terminates in a month, for whatever reason.

Typically, terminations of mortgages in securitized pools come in one of four forms: homeowner refinance; a move; loan default; and a small percentage of curtailment activity (borrowers pay off their loan fractionally more each month and thereby shorten the return timeline). For a great many years, the most important termination to predict—and the largest fraction of terminations—was the refinance component.

In the years during and leading up to the housing bubble, consistently rising home values and relatively stable lending standards created an environment where any borrower who was previously approved and given an agency-backed loan was able

to take advantage of refinancing opportuni-ties that occurred. Since almost every area of the country saw positive home-price growth from the mid-1990s through 2006, all bor-rowers originating a loan with 80% loan-to-value ratio (LTV) saw their current LTV fall and therefore qualified for refinancing.

Because the lending environment (char-acterized by a set of mortgage rate “spreads” for various credit and LTV bands) was stable until mid-2007, borrowers with LTVs above 80%, or who took out second liens on their properties, or even had impaired credit were able to refinance into similar products while paying no more above the prevailing mort-gage rates they had paid on their existing loan at origination. Thus, since mortgage rates had fallen, refinancing was robust and lim-ited only by borrowers’ willingness to seek a new loan.

As a result, prepayment models of this era were able to explain the entirety of prepay-ment observations using the concept of refi-nancing incentive paired with a treatment of the (unobserved) variations among borrowers of the same loan type. A refinancing incentive

JSF-WEINER.indd 21JSF-WEINER.indd 21 4/19/10 6:47:25 PM4/19/10 6:47:25 PM

Page 3: Spring 2010 Volume16,Number 1 · PDF fileactually at a coupon that is much higher than average mortgage rates, but are unable to refinance because of a number of factors ... 1 illustrates

PREDICTING AGENCY PREPAYMENTS IN THE CURRENT MARKET ENVIRONMENT SPRING 2010

For investors still using models from as recently as a couple of years ago, there is significant cause to doubt the validity of their projections. Two years ago, default models for agency securities were all but nonexistent. The paradigm has shifted. Most borrowers today are actually at a coupon that is much higher than average mortgage rates, but are unable to refinance because of a number of factors.

REFINANCING LOCK-OUT

Since mid-2007, borrowers with impaired credit have had more difficulty getting agency loans as a result of tighter underwriting standards and revised pricing matrices. Exhibit 1 illustrates the composition by credit band of recent vintages of FNMA/FHLMC 30-year fixed loans. While loans with FICO scores below 750 made up about half of originations during 2007 and before, their share dropped to about a quarter in 2009. Exhibit 2 shows the average spread that borrowers of various credits were given. Relative mortgage rates for low-credit borrowers rose by as much as about 40 basis points around the onset

was typically formulated as the difference between a borrower’s coupon and prevailing mortgage rates for an equivalent loan product. Adjustments that accounted for a borrower’s spread-at-origination (SatO) were necessary to properly inhibit prepayment for high-coupon, low-credit, or high-LTV borrowers. In addition, pools contained a mixture of borrowers, each having a different willing-ness to refinance. After an initial mortgage rate drop, the “fast” prepayers would refinance quickly, leaving behind a pool dominated by more sluggish borrowers. Subsequent prepayment rates for equal incentive would be reduced, and the pool was said to be “burnt-out.”

Today, the MBS market is a much different land-scape than it was just a few short years ago. Agency securities (those issued by Fannie Mae, Freddie Mac, and Ginnie Mae) now make up 75% of all new-originations activity and account for approximately $10 trillion in outstanding mortgage obligations, dwarfing all other MBS investments. The old rules about what to model for have changed as well, and smart investment managers ensure that their predictive models are recalibrated to ref lect these new realities.

0%

20%

40%

60%

80%

100%

2005

Q1

2005

Q2

2005

Q3

2005

Q4

2006

Q1

2006

Q2

2006

Q3

2006

Q4

2007

Q1

2007

Q2

2007

Q3

2007

Q4

2008

Q1

2008

Q2

2008

Q3

2008

Q4

2009

Q1

2009

Q2

2009

Q3

2009

Q4

Fra

ctio

n of

Orig

inat

ions

<650 650 - 675 675 - 700 700 - 725 725 - 750 750 - 775 >775

E X H I B I T 1FNMA/FHLMC 30-Year Fixed Loan Originations by Credit Band and Quarter

Source: LPS Applied Analytics Loan Servicing Database.

JSF-WEINER.indd 22JSF-WEINER.indd 22 4/19/10 6:47:25 PM4/19/10 6:47:25 PM

Page 4: Spring 2010 Volume16,Number 1 · PDF fileactually at a coupon that is much higher than average mortgage rates, but are unable to refinance because of a number of factors ... 1 illustrates

THE JOURNAL OF STRUCTURED FINANCE SPRING 2010

(calculated from original LTV and HPI change as of Jan-uary 2010) for FNMA/FHLMC 30-year fixed loans origi-nated between 2005 and 2008. Slightly more than half of these borrowers are currently over 80% in loan-to-value, making refinancing much less affordable and effectively locking them out. Furthermore, these vintages comprise the bulk of mortgages with coupons currently over par. In addition, many borrowers took out second liens in the years leading up to 2007. Exhibit 4 depicts the prevalence of second liens through time and across various vintages. The measurement of second lien existence was performed using an anonymous matching process and public record property data so that it included both simultaneous and delayed “silent” second lien originations. Exhibit 4 shows that about half of pre-2007 borrowers had taken out second liens on their properties by the start of 2007. The 2005–2006 vintages were dominated by simultaneous second lien origination, while the 2003–2004 vintages appeared to use second liens to take cash out of their properties in the midst of the housing bubble. As a result, about half of borrowers that are today under 80% LTV actually have combined LTVs over 80%. Since second liens effectively cannot be refinanced in the current environment, these borrowers are also locked out of refinancing.

of the mortgage crisis in early 2008. The improved credit composition of the 2009 vintage can be thought of as a by-product of the low-credit refinancing “lock-out” depicted in Exhibit 2. Of particular interest in Exhibit 2 is the marginal spread increase for even high-credit bor-rowers. Since recent originations have been composed of predominantly higher-credit borrowers, the prevailing mortgages rates that are reported are, in essence, biased.

Modern agency prepayment models must account for these changes. Within a given vintage, higher-coupon pools tend to be composed of poorer-credit borrowers (since those borrowers had to pay a larger spread at origi-nation). In the current environment, the forecasted pre-payment rates of those higher-coupon pools need to be lowered relative to lower coupon pools to a greater extent than in prior epochs. It is also critically important not to associate the measured average mortgage rate with a constant-credit-quality mortgage rate. In particular, the credit band that defined the par coupon from 2000 to 2007 is now subject to an extra 30bp spread.

Home prices have fallen since 2006 in almost every part of the country. As a result, many borrowers owe more on their mortgages than their properties are worth. Exhibit 3 shows the distribution of current LTV

-30

-20

-10

0

10

20

30

40

50

60

2000 2002 2004 2006 2008 2010

Spr

ead

at O

rigin

atio

n (b

p)600-625 625-650 650-675 675-700

700-725 725-750 750-775 775-825

E X H I B I T 2FNMA/FHLMC 30-Year Fixed Spreads by Credit Band and Month of Origination

Source: LPS Applied Analytics Loan Servicing Database.

JSF-WEINER.indd 23JSF-WEINER.indd 23 4/19/10 6:47:25 PM4/19/10 6:47:25 PM

Page 5: Spring 2010 Volume16,Number 1 · PDF fileactually at a coupon that is much higher than average mortgage rates, but are unable to refinance because of a number of factors ... 1 illustrates

PREDICTING AGENCY PREPAYMENTS IN THE CURRENT MARKET ENVIRONMENT SPRING 2010

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

40% 50% 60% 70% 80% 90% 100% 110% 120%

Current Loan-To-Value Ratio

Fra

ctio

n of

Loa

ns B

elow

Cur

rent

LT

V

E X H I B I T 3Cumulative Distribution of Current LTV for FNMA/FHLMC 30-Year Fixed, 2005–2008 Originations

Source: LPS Applied Analytics Loan Servicing Database and LPS Applied Analytics HPI.

0

10

20

30

40

50

60

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Date

Sec

ond

Lien

Pre

vale

nce

(% o

f Act

ive

Bal

ance

)

2000 2001 2002 2003 20042005 2006 2007 2008 2009

E X H I B I T 4Percentage of Outstanding Balance Corresponding to Loans Having a Second Lien

Source: LPS Applied Analytics Loan Servicing Database and LPS Applied Analytics Public Records Database.

JSF-WEINER.indd 24JSF-WEINER.indd 24 4/19/10 6:47:25 PM4/19/10 6:47:25 PM

Page 6: Spring 2010 Volume16,Number 1 · PDF fileactually at a coupon that is much higher than average mortgage rates, but are unable to refinance because of a number of factors ... 1 illustrates

THE JOURNAL OF STRUCTURED FINANCE SPRING 2010

the entire appraisal curve is shifted up by about 6%. This extra “inf lation” is even more pronounced in loans origi-nated for the purposes of taking cash out. An effect of this appraisal bias is to push borrowers to higher current LTVs, causing even further lock-out.

Some have had an urge to attribute the new reality of lower agency refinancing activity, even at historically low mortgage rates, to “stricter underwriting standards” and re-parameterize prepayment models accordingly. While essentially correct, this approach misses an opportunity to expand our understanding of refinancing to all of its drivers and dooms the model to future errors where the mix of factors is not the same. For instance, the 2009 vin-tage does not suffer from low-credit, high-current LTVs or second liens as did its predecessors. Therefore, if mortgage rates were ever to dip enough to provide incentive to this vintage, borrowers would be expected to refinance readily, even in the current environment. Furthermore, refinancing over the next several years will likely be determined by how quickly currently locked-out loans are given the ability to refinance, through changes in either the factors causing the lock-out or the industry’s pricing of loans.

We also uncover some evidence for appraisal bias, which is most significant for the 2006–2008 vintages, when house prices had started to f latten or fall. Using public records data, we can calculate the expected property value for a particular loan at the time of its origination. Exhibit 5 plots a distribution of appraisal values of non-purchase loans relative to this expectation for two different HPI environments. While appraisals were relatively accu-rate in the rising HPI environment, they were extremely inf lated in the falling HPI environment, as evidenced by the rightward shift. By comparison, purchase loans have distributions of property values centered on the expecta-tion, consistent with the repeat-sales methodology of the HPI used. Appraisal bias, measured as the centroid of the distributions shown in Exhibit 5, is plotted against HPI change in the six months prior to origination in Exhibit 6. The downward sloping trend evident in Exhibit 6 repre-sents what we would expect to see if appraisers were out-dated in their appraisal by six months. This phenomenon is fairly intuitive because there are delays in the reporting of property data, and appraisers must look backwards in time to find comparison sales. In addition to the slope,

0

0.005

0.01

0.015

0.02

0.025

0.03

–60% –40% –20% 0% 20% 40% 60%

Appraisal Value of Property Relative to Indexed Prior Sale

Fre

quen

cy (

by H

PI E

nviro

nmen

t)

Down 4% in Prior 6 months Up 6% in Prior 6 months

E X H I B I T 5Distribution of Appraised Property Values Relative to Expectation Based on Indexed Prior Sale for Two Different HPI Environments

Source: LPS Applied Analytics Loan Servicing Database and LPS Applied Analytics Public Records Database.

JSF-WEINER.indd 25JSF-WEINER.indd 25 4/19/10 6:47:26 PM4/19/10 6:47:26 PM

Page 7: Spring 2010 Volume16,Number 1 · PDF fileactually at a coupon that is much higher than average mortgage rates, but are unable to refinance because of a number of factors ... 1 illustrates

PREDICTING AGENCY PREPAYMENTS IN THE CURRENT MARKET ENVIRONMENT SPRING 2010

buyouts for these agencies more difficult. Recent changes to accounting standards suggest that, going forward, most loans delinquent more than three months will be bought out quickly. In addition to a short-term spike in buyouts in early 2010 when the new accounting standards took effect, we expect buyout rates for FNMA/FHLMC to add about 3% conditional prepayment rates (CPRs) well into 2011. Pools backed by 2005–2007 vintage loans, pools from states having strong real estate bubbles, or higher SatO loans could have much higher buyout rates.

THE BOTTOM LINE

For the portfolio manager at an insurance com-pany who owns $5–10 billion in agency securities and is trying to determine the underlying risks associated with the portfolio, the change from interest rate and cash-out refinance prepayment risk modeling to a focus on default-triggered buyouts represents nothing less than a complete paradigm shift.

Today, many professional investors use agency models provided by investment banks and other enti-ties with which they trade to do their portfolio analysis. Some are using stand-alone commercial models provided

DEFAULTS

The same market conditions that result in a signifi-cant percentage of borrowers being locked out of refi-nancing has resulted in drastic increases in loan defaults. Investors in FNMA/FHLMC/GNMA securities are intrinsically protected from losses due to defaults, but are sensitive to prepayments generated by defaults. The decrease in voluntary prepayments due to lock-out, and the increase in defaults (primarily among those borrowers who are locked out), has led to an increased importance being placed on default prediction.

Loans that become delinquent are generally removed from agency pools prior to actual loan liquidation. These removals are referred to as buyouts and appear to inves-tors as prepayments. GNMA buyouts are performed by servicers, at their discretion. Historically, servicers have bought out eligible delinquent loans at a monthly rate of 18.8% on average over the 5-year period from January 2005 to January 2010. Recent increased delinquencies and high buyout rates have contributed more than half of recent GNMA prepayments, as shown in the Exhibit 7.

By contrast, FNMA/FHLMC buyouts are driven by the agencies themselves, making measurement of

–10%

–5%

0%

5%

10%

15%

–10% –5% 0% 5% 10% 15%

HPI Change in Prior 6 Months

App

rais

al B

ias

E X H I B I T 6Appraisal Bias, Measured as the Centroid of Peak in Exhibit 5, as a Function of HPI Environment

Source: LPS Applied Analytics Loan Servicing Database and LPS Applied Analytics Public Records Database.

JSF-WEINER.indd 26JSF-WEINER.indd 26 4/19/10 6:47:26 PM4/19/10 6:47:26 PM

Page 8: Spring 2010 Volume16,Number 1 · PDF fileactually at a coupon that is much higher than average mortgage rates, but are unable to refinance because of a number of factors ... 1 illustrates

THE JOURNAL OF STRUCTURED FINANCE SPRING 2010

simply don’t hold water today. Investors must arm them-selves with updated, more accurate modeling that takes into account the realities of the new environment.

Ultimately, by accurately predicting the value and risk of agency securities with up-to-date modeling, inves-tors will be less likely to overpay, thereby increasing the value of their investments. The only way to make informed investment decisions is to determine how securities are priced and whether it is possible to get better than market in relation to the securities’ intrinsic value. Even for those very liquid securities that are just being originated today, or have been very recently, if inaccurate assumptions are made going forward about prepayment and default, the risk of overpaying for a security remains.

The value of a bond in the agency security space is absolutely determined by its future prepayments. If an investor is wrong on a prediction of what those prepay-ments will be, he or she will be wrong about the value. And if one is wrong about value, it becomes far too easy to pay too much for a less-than-stellar investment, or perhaps worse, to miss out on golden opportunities that one should definitely pursue.

To order reprints of this article, please contact Dewey Palmieri at [email protected] or 212-224-3675.

by analytics firms, while still others continue to use the basic models integrated with whatever portfolio man-agement software they already own.

Regardless of the particular models being used or their sources, investors need to examine the prepayment models currently employed. The main question is whether the model addresses these new issues associated with pre-payments and buyouts. Unfortunately, for the great majority of investors, the answer is no, and the potential impact of these new factors to portfolios cannot be overstated.

Portfolio managers use interest rate derivatives to successfully hedge their exposure to interest rate move-ments. The impact of the factors outlined above to par-ticular securities is either faster or slower prepayments, depending on the collateral type. In the case of loans from the bubble years of 2005 and 2006 and loans with higher coupons, the likelihood is that investors will see much faster prepayments than in the recent past. On the other end of the spectrum, a good number of loans from the 2003–2004 period may be at 5% coupon. Purchased before the bubble peaked, these securities will not be as sensitive to default; however, they are predicted to have depressed prepayment rates, as they are subject to the refinance lock-out constraints discussed earlier.

In short, the predictive models investors have relied upon to gauge the risk and performance of agency securities

0%

1%

2%

3%

2005 2006 2007 2008 2009 2010

Pre

paym

ents

and

Buy

outs

(%

/mon

th)

Total Prepayments Buyouts

E X H I B I T 7Contribution of Buyouts to Total Prepayments for GNMA pools

JSF-WEINER.indd 27JSF-WEINER.indd 27 4/19/10 6:47:26 PM4/19/10 6:47:26 PM