Академический Документы
Профессиональный Документы
Культура Документы
Two parameters determine a mortgage’s credit risk: probability of default and loss severity given
default. While there is a growing body of research relevant to the modeling and estimation of mortgage
default, there are few studies on loss severity (the percentage lost in the event of default) because of
limited data. New data released by Freddie Mac, however, offer analysts the ability to assess mortgage
loans made over 13 years for their loss severity. This robust database of more than 17 million loans
offers a range of new and useful insights, by state, into the ultimate financial losses associated with a
loan after it experiences a credit event. The addition of loan performance information beyond the credit
event is a new and welcome addition to the single-family loan-level dataset.
In this brief, we review loans experiencing four distinct credit events, and for the first time track
what happens to the loans. For loans that liquidate, we determine the loss severity measured by the
percentage of unpaid principal balance lost at the time of default. This analysis allows us to assess both
the value of mortgage insurance (MI) and the accuracy of the preset severities used in Freddie Mac’s
risk-sharing deals. We find MI significantly lowers the severities. We also find that the preset severity
schedule is reasonable for loans with a loan-to-value (LTV) ratio of 60–80 but too high for deals backed
by higher-LTV loans.
We also analyze severity by loan size, state, and type of liquidation. We find that small loans have
higher severity than larger ones, that real-estate-owned (REO) sales have higher severity than short
sales, and that there is no stable relationship between the state of origination and severity. Finally, we
review the components of loss—liquidation value and expenses—and find that the latter contributes
significantly to the ultimate loss.
What Do the New Data Include?
Historically, data allowed market participants to track loans only through the advent of a credit event.
Freddie Mac and Fannie Mae (the GSEs) define a credit event as a loan going 180 days delinquent or
being liquidated through a deed-in-lieu, short sale, foreclosure sale, or REO before the 180-day
delinquency point. Once a loan experienced a credit event, it was removed from the dataset and the loss
was calculated by multiplying the balance affected by a predetermined severity.
In support of their risk-sharing deals, in 2013 the GSEs began releasing quarterly performance data
on their 30-year full-documentation amortizing book of business. In November 2014, Freddie Mac’s
quarterly release contained a broader dataset that included loan dispositions. This increased
transparency may indicate that Freddie anticipates issuing a deal in which risk-sharing payments will be
triggered by actual severities rather than by credit events.
Freddie’s new data allow analysts to calculate actual severities by various credit event types and
timing of dispositions, and to compare actual severities to the predetermined severities on risk-sharing
deals. In addition, the new data allow market participants to break down loss severity into its various
components: net sale proceeds, expenses, MI recoveries, and non-MI recoveries. This is also the first
time the GSEs have released loan-level loss data for mortgages. Before this, loss data had been available
only on mortgages that were collateral for private-label securities.
Among loans originated between 1999 and 2004, 2.3 percent have experienced a credit event (table 1).
For 2007, the single worst issue year, 12.6 percent of originated loans have experienced a credit event.
1
Note that the FICO/LTV distribution for 1999–2004 is very similar to that of 2007. The difference in
default rates reflects how hard the 2007 vintage was hit by the Great Recession and the accompanying
home price depreciation. The earlier vintages had some home price appreciation and, in some cases, had
prepaid by the time the Great Recession hit.
The current book of business has a much higher percentage of high-FICO loans than either the
1999–2004 or the 2007 book of business; it would hence have fewer credit events if one applied these
experiences bucket by bucket. If we applied the composition of the 2013 book of the business to the
1999–2004 experience, defaults would be 1.2 percent. If we applied the composition of the 2013 book
of business to the 2007 experience, defaults would be 8.4 percent.
TABLE 1
Percent of Credit Event Experiences by Origination Year, FICO Score, and LTV
Table 2 shows the current status of 13 years of loans that have encountered credit events, sorted by
vintage and LTV. Not all loans that experience a credit event will be liquidated; some will be
rehabilitated. Because of long timelines, some others will still be in process. We identify eight paths for
2
loans that experience credit events: (1) current without a modification, (2) modified and current on the
modification, (3) prepay without modification, (4) prepay after modification, (5) modified and not
current on the modification, (6) in pipeline with no modification, (7) liquidated via a foreclosure
alternative (deed-in-lieu, short sales, foreclosure sale) and (8) liquidated through REO. Under outcomes
Twenty-two percent of the loans that experienced credit events from 1999 to 2013 have been
rehabilitated in one of four ways and, thus, not likely to experience an eventual loss (table 2, total of
columns 1–4, shown in column 5). The most common route to rehabilitation—constituting 11 percent of
all loans that experienced a credit event—are those that have been modified and are now current. The
next largest category are loans that later prepaid with no modification. Lower-LTV loans are more likely
to prepay without a modification than higher-LTV loans. Fifteen percent of loans with LTVs of 60 or
under became current again, versus 5–6 percent of loans with higher LTVs.
Among loans with credit events from 1999 to 2013, 22 percent have been rehabilitated; 78
percent have not and are likely to experience losses.
Table 2 also shows that 78 percent of loans that experienced credit events from 1999 to 2013 are
likely to experience losses. Of this 78 percent, more than two-thirds (54 percent) have already been
liquidated though REO or foreclosure alternatives. The remaining 24 percent are either modified and
not current or not modified but in the pipeline—that is, they will eventually be liquidated.
These liquidation numbers are, not surprisingly, highly dependent on LTV and vintage year. For
loans with an original LTV of 60 or under, 63 percent of the loans are expected to eventually be
liquidated. In contrast, 77 percent of the loans with LTVs between 60 and 80 are expected to eventually
liquidate, rising to 81 percent for loans with LTVs over 80. The higher original LTV makes a prepayment
less likely (if LTV is measured correctly, a low-LTV borrower who has experienced modest home price
depreciation and is behind on his payments can sell the home without a loss). It also makes
rehabilitation less attractive for the borrower, as the house will likely have negative equity. Similarly,
the percentage of loans that will eventually liquidate is higher during the crisis years, as home price
deterioration put many mortgages into a negative equity situation. Table 3 shows the percent of loans
we expect to liquidate, by LTV and FICO buckets, in a manner consistent with table 1.
23.2 cents for every dollar remaining at default was lost for 1999–2004 liquidated loans;
36–40 cents for every dollar was lost for 2005–08 loans.
Table 4 shows loss severities by FICO/LTV bucket and issue years, including all loans that have already
gone through REO or a foreclosure alternative. The table includes loans that have liquidated without a
loss, but it does not include loans that have cured after their credit event and then prepaid. The
severities at liquidation are 23.2 percent for the 1999–2004 period, rising to 36–40 percent for 2005–
08, then dropping sharply thereafter.
The relationship between loss severities and LTV categories is particularly interesting. Severities for
loans with LTVs over 80 are much lower than for loans with LTVs between 60 and 80. In fact, the
severities for the over-80-LTV loans are even lower than severities for the 60-or-under-LTV loans. The
reason is simple. Loans with LTVs over 80 are required to have mortgage insurance, which covers the
first loss; this coverage is usually deep enough that Freddie is not exposed unless the market value of
the home drops far more than 20 percent. For example, standard practice is to bring down an 85 LTV
mortgage to 73 LTV, a 90 or 95 LTV mortgage to 65 LTV, and a 97 LTV mortgage to 63 LTV. These
results would indicate that mortgage insurance is more effective at protecting the GSEs against losses
than is commonly assumed.
TABLE 4
Severity at Liquidation by Origination Year, FICO, and LTV (percent)
Year FICO ≤ 60 60–80 > 80 Total
≤ 700 20.4 31.0 14.9 22.3
700–750 17.1 29.3 17.8 24.3
1999–2004
> 750 19.0 29.8 20.8 26.3
Total 19.3 30.4 15.9 23.2
≤ 700 28.8 40.6 25.2 35.8
700–750 26.6 39.4 26.8 36.2
2005
> 750 26.2 39.0 27.9 36.2
Total 27.7 39.9 26.0 36.0
≤ 700 35.2 45.3 27.0 39.7
700–750 32.3 43.6 30.1 40.7
2006
> 750 30.6 42.2 29.1 39.6
Total 33.5 44.1 28.0 40.0
≤ 700 38.8 46.1 28.6 38.7
700–750 33.6 43.4 29.3 38.9
2007
> 750 31.7 41.2 27.8 37.3
Total 35.9 44.2 28.7 38.5
≤ 700 31.9 43.6 27.5 36.5
700–750 27.2 40.1 25.3 34.1
2008
> 750 23.1 37.5 23.3 31.9
Total 28.6 40.9 25.8 34.6
≤ 700 21.9 34.4 16.1 30.2
700–750 20.5 30.7 13.6 26.3
2009–10
> 750 18.5 28.4 13.8 24.5
Total 20.5 30.7 14.1 26.5
≤ 700 0.0 23.7 6.5 17.0
700–750 10.4 26.1 8.1 15.5
2011–13
> 750 0.0 22.1 9.2 15.6
Total 8.2 23.8 8.3 15.9%
Total 29.2 39.9 23.1 33.9
To assess the expected severity for loans that have experienced credit events, we multiply the
percentage of loans that we expect to liquidate by the severity on the loans that have already
liquidated. For the 1999–2004 book of business, for loans with 60–80 LTV, we expect 72 percent of the
loans to liquidate, at an average severity of 30.4 percent. Thus, the expected severity from all loans of
this vintage that have experienced a credit event is 22 percent (0.72 x 0.304).
The STACR deals apply a preset severity schedule to the percent of balances affected by credit
events. This schedule is a weighted average of a step function and differs for the 60-to-80-LTV deals and
the over-80-LTV deals. For the 60-to-80-LTV deals, the schedule is 15 percent severity for the first 1
percent of credit events, 25 percent for credit events between 1 and 2 percent, and 40 percent for credit
events over 2 percent. If we look at the 1999–2004 book of business for loans with 60–80 LTV (table 1),
the cumulative credit events are 2.0 percent. Thus, the loss severity would be 20 percent ([1 x 0.15 + 1 x
0.25]/2). If we compare this number to the actual loss severity, 22 percent, the two are very similar.
Overall, for vintages through 2008, Freddie’s preset severity schedule produces results very close to the
actual severities (table 5). The comparison for 2009 and later vintages is less valid. We would expect the
preset numbers to be lower than the actual severities, as the credit events are still ramping up.
The preset severity schedule is reasonable for loans with LTVs of 60–80 but too high for
deals backed by higher-LTV loans.
To calculate losses on over-80-LTV loans, the preset severity schedule is 10 percent for the first 1
percent of credit events, 20 percent for credit events between 1 and 3 percent, 25 percent for credit
events between 3 and 5 percent, and 40 percent for credit events over 5 percent. For example, 4.3
percent of the 1999–2004 book of business experienced a credit event. Thus, the overall loss severity
for that book would be 19 percent ([1 x 0.10 + 2 x 0.20 + 1.3 x 0.25]/4.3).The actual severity was 12
percent. In general, when using the preset severity schedule on the over-80-LTV mortgages the derived
severities are very close to the derived severities on the 60-to-80-LTV mortgages (table 5). However,
the actual severities on the over-80-LTV mortgages are much lower than the preset severity schedule.
TABLE 5
Actual and Calculated Severities by Origination Year, FICO, and LTV (percent)
60–80 > 80
Year FICO Actual severity Calculated severity Actual severity Calculated severity
≤ 700 22 22 11 24
700–750 22 18 14 16
1999–2004
> 750 22 15 17 13
Total 22 20 12 19
≤ 700 32 34 21 35
700–750 32 30 23 32
2005
> 750 33 23 25 26
Total 32 31 22 33
≤ 700 36 36 23 36
700–750 36 33 26 33
2006
> 750 36 24 26 28
Total 36 33 24 34
≤ 700 36 36 24 36
700–750 35 33 25 34
2007
> 750 35 25 24 30
Total 35 33 24 35
≤ 700 33 35 22 35
700–750 31 29 20 31
2008
> 750 30 21 19 24
Total 31 28 21 32
≤ 700 27 22 14 18
700–750 25 16 12 14
2009–10
> 750 24 15 12 10
Total 25 15 12 12
≤ 700 19 15 6 10
700–750 22 15 7 10
2011–13
> 750 18 15 8 10
Total 19 15 7 10
Total 31 22 19 25
Loss severities are also affected by loan size (table 6). For all vintage years, the smallest loans exhibit the
highest severity, and the severities decline monotonically with loan size. For example, for 1999–2004,
loans with a balance of $60K or less had a loss severity of 47 percent, while those with a balance of
$60–100K had a severity of 31.3 percent and those with a balance over $100K had a severity of 18
percent. One possible reason is that loan size may influence liquidation costs. A smaller loan is more
likely to trade at a higher foreclosure discount. In addition, liquidation costs are likely larger on smaller
loans as a percentage of loan unpaid principal balance (UPB).
TABLE 6
Severity by Origination Year, Loan Size, and LTV (percent)
The data also afford us the opportunity to look at loss severity by state by vintages (table 7). This
relationship is not constant through time. The sand states (California, Florida, Arizona, and Nevada) had
very high severities for loans originated in 2005–07, but their severities were around or below the
national average both pre-crisis and post-crisis. By contrast, some of the Rust Belt states (Michigan, Ohio,
Illinois, and Indiana) were above the national average for both pre-crisis and crisis origination years.
TABLE 7
Severity by States and Origination Year (percent)
Severity may also vary by liquidation types, whether REO or foreclosure alternative (table 8). REO
liquidation produces much higher severities than foreclosure alternatives. For example, for the 2007
book of business, REO liquidations had a severity of 42 percent, versus other foreclosure alternatives at
35 percent. This suggests that though the GSEs have put into place incentives to encourage foreclosure
alternatives, the use of stronger incentives merits further study.
TABLE 8
Severities by Liquidation Types
Foreclosure
Year alternative REO All
1999–2004 17% 26% 23%
2005 34% 38% 36%
2006 38% 42% 40%
2007 35% 42% 39%
2008 30% 39% 35%
2009–10 21% 32% 26%
2011–13 11% 22% 16%
Total 32% 36% 34%
It is important to realize that severities are based on the loans that actually liquidate. The last
column of table 7 shows this percentage state by state. As noted earlier, the share of loans that actually
Loss on the sale of the properties, calculated as [1 - (net sale proceeds/defaulted UPB)], averages 40
percent and is higher with over-80 LTV loans. For example, for 2007, the sale loss for over-80 LTV loans
is 49 (100-51) percent, while the sale proceeds per defaulted UPB for 60–80 LTV loans is 40 (100-60)
percent. Mortgage insurance will offset some of these losses. Thus, for 2007, the total severity for over-
80 LTV loans is 36 percent, lower than 44 percent for loans with 60–80 LTV. For the entire universe of
loans liquidated at a loss, mortgage insurance adds 6.1 percent to total recoveries; it adds 19–22
percent for above-80 LTV loans. Other non-MI recoveries average less than 2 percent. Note that
expenses can be sizeable, adding 8.6 percent to severity. These are direct expenses only; they do not
include any expenses from lost interest.
Notes
1. This issue was discussed in Laurie Goodman and Jun Zhu, “The GSE Reform Debate: How Much Capital Is
Enough?” Journal of Structured Finance Spring, 2014.
2. Current is defined as current for the past three months.
3. We calculate severity as the unpaid principal balance of the loan plus expenses, less the net sales proceeds,
less recoveries from MI and other sources as a percentage of the unpaid principal balance. Many researchers
would argue that we should have included lost interest expense, calculated as [the note rate less the servicing
fee] times the amount of time the loan is delinquent or in foreclosure. In fact, the mortgage insurers do
reimburse the GSEs for lost interest expense up to a certain maximum. We deliberately elected not to do so.
The GSEs’ real cost of carrying these loans is not [the note rate less the servicing fee], but rather their short-
term funding cost, which is near zero. We ignored these funding costs, which would very marginally raise
severities, for the purposes of our calculations, as we figured they would be offset by two items that marginally
lower severities: compensatory fees paid by the lenders to Freddie are not included in the recovery amounts,
and recoveries include pool policies that Freddie no longer uses.
4. On the 60- to-80-LTV deals, the Fannie and Freddie preset severity schedules are very similar: the only
difference is that Fannie uses a 10 percent loss rate (rather than Freddie’s 15 percent loss rate) for the first 1
percent of credit events. For the over-80-LTV deals, Fannie uses a 25 percent loss rate (rather than Freddie’s
40 percent) where credit events exceed 5 percent.
5. For a fuller discussion of this point, see Laurie Goodman, “Servicing Is an Underappreciated Constraint on
Credit Access” (Washington, DC: Urban Institute, 2014).
6. This occurrence was most common in the 60-and-under LTV bucket and for loans with mortgage insurance.
Before joining Urban in 2013, Goodman spent 30 years as an analyst and research department manager
at a number of Wall Street firms. From 2008 to 2013, she was a senior managing director at Amherst
Securities Group, LP, a boutique broker/dealer specializing in securitized products, where her strategy
effort became known for its analysis of housing policy issues. From 1993 to 2008, Goodman was head of
Global Fixed Income Research and Manager of US Securitized Products Research at UBS and
predecessor firms, which was ranked first by Institutional Investor for 11 straight years. She has also
held positions as a senior fixed income analyst, a mortgage portfolio manager, and a senior economist at
the Federal Reserve Bank of New York.
Goodman was inducted into the Fixed Income Analysts Hall of Fame in 2009. She serves on the board of
directors of MFA Financial and is a member of the Bipartisan Policy Center’s Housing Commission, the
Federal Reserve Bank of New York’s Financial Advisory Roundtable, and the New York State Mortgage
Relief Incentive Fund Advisory Committee. She has published more than 200 articles in professional
and academic journals, and has coauthored and coedited five books. Goodman has a BA in mathematics
from the University of Pennsylvania and a MA and PhD in economics from Stanford University.
Jun Zhu is a senior financial methodologist at The Urban Institute. She designs and conducts
quantitative studies of housing finance trends, challenges, and policy issues. Previously she s as a senior
economist in the Office of the Chief Economist at Freddie Mac where she conducted research on the
mortgage and housing markets, including default and prepayment modeling. While at Freddie Mac, she
also served as a consultant to the US Treasury on housing and mortgage modification issues. She
obtained her PhD in real estate from the University of Wisconsin–Madison in 2011.
Ongoing support for HFPC is also provided by the Housing Finance Council, a group of
firms and individuals supporting high-quality independent research that informs
evidence-based policy development. Funds raised through the Council provide flexible
resources, allowing HFPC to anticipate and respond to emerging policy issues with
timely analysis. This funding supports HFPC’s research, outreach and engagement, and
general operating activities.