Supporting Statement A_Attach #4--NMDB Technical Documentation 15-01 (Aug. 2015)

Technical
Report Series

National Mortgage Database
Technical Report 15-01

August 17, 2015

This document was prepared by Robert B. Avery, Kenneth P. Brevoort, Ian H. Keith, Ismail E.
Mohamed, Forrest W. Pafenberg, Jay D. Schultz, and Claudia E. Wood. The analysis and
conclusions are those of the authors and do not necessarily represent the views of the Consumer
Financial Protection Bureau, the Federal Housing Finance Agency or the United States.

1.0

Introduction

The National Mortgage Database project is a multi-year project being jointly undertaken by the
Federal Housing Finance Agency (FHFA) and the Consumer Financial Protection Bureau
(CFPB). The project is designed to provide a rich source of information about the U.S. mortgage
market based on a five percent sample of residential mortgages. It has two primary components:
(1) the National Mortgage Database (NMDB) and (2) the quarterly National Survey of Mortgage
Borrowers (NSMB).
The NMDB will enable FHFA to meet the statutory requirements of section 1324(c) of the
Federal Housing Enterprises Financial Safety and Soundness Act of 1992, as amended by the
Housing and Economic Recovery Act of 2008, to conduct a monthly mortgage market survey.
Specifically, FHFA must, through a survey of the mortgage market, collect data on the
characteristics of individual mortgages, including those eligible for purchase by Fannie Mae and
Freddie Mac and those that are not, and including subprime and nontraditional mortgages. In
addition, FHFA must collect information on the creditworthiness of borrowers, including a
determination of whether subprime and nontraditional borrowers would have qualified for prime
lending. 1
For CFPB, the NMDB project will support policymaking and research efforts and help identify
and understand emerging mortgage and housing market trends. The CFPB expects to use the
NMDB, among other purposes, in support of the market monitoring called for by the Dodd-Frank
Wall Street Reform and Consumer Protection Act, including understanding how mortgage debt
affects consumers.
FHFA and CFPB considered existing databases but determined that none sufficiently support the
above objectives. 2 The NMDB, when fully complete, will be a de-identified loan-level database
of closed-end first-lien residential mortgages. It will: (1) be representative of the market as a
whole; (2) contain detailed, loan-level information on the terms and performance of mortgages, as
well as characteristics of the associated borrowers and properties; (3) be continually updated; (4)
have an historical component dating back before the financial crisis of 2008; and (5) provide a
sampling frame for the NSMB (see NMDB Technical Report 15-02).
The core data in the NMDB are drawn from a random 1-in-20 sample of all closed-end first-lien
mortgage files outstanding at any time between January 1998 and June 2012 in the files of
Experian, one of the three national credit repositories. 3 The use of a sampling frame substantially
reduces the privacy risk associated with any data collection. By contrast, a universal registry can
present challenges for privacy since it is known that a particular loan must be in the dataset.
However, for a 1-in-20 sample, the odds are 95 out of 100 that a particular loan is not in in the
1

FHFA interprets the NMDB project as a whole, including the NSMB, as the “survey” required by the Safety and
Soundness Act. The statutory requirement is for a monthly survey. Core inputs to the NMDB, such as a regular
refresh of credit-bureau data, occur monthly, though the NSMB does not.

2

Please see the Appendix for a discussion of existing sources and their limitations.

3

Experian was chosen through a competitive procurement process to assist in creating the NMDB.

2

database. In addition, the sample used is large enough to support almost all types of statistically
valid analyses but small enough to manage logistically, thus dramatically reducing both contract
and personnel costs.
A random 1-in-20 sample of mortgages newly reported to Experian is added each quarter.
Mortgages are followed in the NMDB database until they terminate through prepayment
(including refinancing), foreclosure, or maturity. Information from credit repository files on each
borrower associated with the mortgages in the NMDB sample is collected from at least one year
prior to origination to one year after termination of the mortgage. The information on borrowers
and loans available to the FHFA, CFPB, or any other authorized user of the NMDB data is deidentified and does not include any direct identifying information such as borrower name,
address, or Social Security number.
This technical report is designed to provide users of the NMDB data with background on the
development of the database, as well as an assessment of the quality of its data. The remaining
sections of this report discuss the development of the contract with Experian, outline the process
of selecting the initial historical sample, describe how the initial sample data were processed,
discuss how the data are being updated, and how administrative data are being merged into the
NMDB. The final section then evaluates the NMDB sample frame.
2.0

The Experian Contract

By interagency agreement between FHFA and CFPB, FHFA leads the production of the NMDB.
Following a competitive procurement process, a five year contract for the core data of the NMDB
was signed between FHFA and Experian in September 2012. Simultaneously, FHFA and CFPB
signed an interagency agreement that codified the cost-sharing (shared equally) and
administrative arrangement.
The Experian contract has several key elements designed to ensure compliance with the Fair
Credit Reporting Act (FCRA) and to protect the privacy of both borrowers and lenders. 4 First,
while Experian will be using name, address and Social Security number for matching purposes
only, this information will not be transmitted to FHFA or CFPB when constructing the NMDB.
Second, any user of the database must sign a “terms of use agreement” that states that they will
not attempt to learn the identity of any borrower. 5 Third, all access to the NMDB must be
through a server at FHFA or CFPB and strictly controlled. Fourth, the NMDB – which is

4

The Fair Credit Reporting Act (FCRA), Public Law No. 91-508, was enacted in 1970, and substantially amended
since, to promote accuracy, fairness, and the privacy of personal information assembled by credit reporting agencies
(CRAs). The Act's primary protection requires that CRAs follow “reasonable procedures” to protect the
confidentiality, accuracy, and relevance of credit information. To do so, the FCRA establishes a framework of
requirements for credit report information that include rights of data quality (right to access and correct), data
security, use limitations, requirements for data destruction, notice, user participation (consent), and accountability.
5

Though FHFA and CFPB have not yet established policies of access or determined who may attempt to obtain
access, the contract allows access to the NMDB to be extended to employees of other federal agencies, the Federal
Reserve System, Fannie Mae, Freddie Mac, and Federal Home Loan Banks, provided the employee has signed the
terms of use agreement.

3

designed to describe the market as a whole – cannot be used for enforcement against any specific
servicer or lender.
3.0

Selecting the Initial Sample

The credit repository core of the NMDB is being developed in two phases: (1) an initial 1-in-20
random sample of closed-end first-lien mortgages active at any time from January 1998 to June
2012 (January 1998 was the earliest available date given Experian’s archive policies); and (2)
quarterly updates that add a 1-in-20 random sample of mortgages newly reported to Experian and
updated information on existing loans still active in the database.
One of the virtues of the credit repository sampling frame is that the repositories maintain records
in a credit report not only of mortgages (and other credit obligations) that are currently active, but
also of those that are closed. However, because of the FCRA, records with derogatory
information are purged from the current credit report after seven years from their point of first
continual delinquency, and Experian's policies dictate a purge of all closed accounts 10 years after
their closing.
However, since Experian retains archives of their data for 10 years or longer, data on mortgages
that have been purged from Experian’s current files can be recovered. These archives, which are
not used for credit granting decisions, contain snapshots of each credit record as it existed at the
close of business on a given day of each month, except that personal information, such as name,
address, and Social Security number, is suppressed.
The bulk of the initial sample for the NMDB was drawn from the June 2012 archive. This was
supplemented by samples from the December 2005 and July 2001 archives that captured loans
that may have been purged from the current files by June 2012.
Trade lines, which are records that contain information about specific loans or debt obligations
that are reported by loan servicers, account for most of the information contained in credit
records. Loan servicers typically update trade line information on a monthly basis using a
standardized format agreed upon by the servicers and the credit repositories (Metro 2® format).
The updates include information on the opening date of the loan, the current and original loan
balance, the type of servicer, loan term and type, payment amount, and loan repayment
performance.
However, the format agreed upon by loan servicers and the credit repositories does not perfectly
identify closed-end first-lien mortgages. Recognizing that some second liens would be sampled
and have to be removed later, trade lines falling under the following categories were deemed
eligible for the NMDB:
(1) any trade line with a Metro 2 “Enhanced Account Type Code” of: 08 (Real estate
loan, specific type unknown), 19 (FHA real estate mortgage), 2C (FMHA real estate
mortgage) , 25 (VA real estate mortgage), 26 (Conventional real estate mortgage), 27
(Real estate mortgage, with or without collateral, usually second mortgage), 85 (Bimonthly mortgage payment), 87 (Semi-monthly mortgage payment), 5A (Real estate –
4

junior liens and non-purchase money first), 17 (Manufactured home loan) , and 05 (FHA
home-improvement loan); or
(2) trade lines reported by servicers with “Kind of Business Codes” of: FB (Mortgage
Brokers), FM (Mortgage Companies), FR (Mortgage Reporters), RE (Real Estate Sales
and Rentals), BM (Bank-mortgage only), FL (Savings and loan – mortgage
department) and Metro 2 “Enhanced Account Type Codes" of: 02 (Secured loan), 04
(Home improvement loan), 66 (Government- secured guaranteed loan), 7B (Agriculture),
9A (Secured home improvement) or a “Secondary Agency Code” of: 01 (Fannie Mae) or
02 (Freddie Mac).
Trade lines in the June 2012 archive that met either of the above criteria were included in the
population from which the initial NMDB 1-in-20 random sample of mortgages was drawn. Any
open-ended or revolving loans otherwise meeting one of the criteria were excluded from the
sampling universe. No other restrictions were imposed.
The first supplemental sample was a 1-in-20 random sample of trade lines drawn from the
December 2005 archive that met the criteria for the June 2012 archive, had information reported
for some period in the past 7 years (indicated by an “Account Balance Date” of January 1998 or
later), and were opened in September 2005 or earlier. In order to exclude loans from the 2005
sample that should be present in the June 2012 archive, loans were excluded if they were last
reported after July 2002 with a reported account status of “current.”
The second supplemental sample, drawn from the July 2001 archive, was a random 1-in-20
sample of trade lines that met the criteria used for the June 2012 archive and that had “Account
Balance Dates” of January 1998 or later and “Account Open Dates” of April 1999 or earlier. Any
trade line with an “Enhanced Status Code” of “current” was excluded from the sample. Again,
these additional conditions were designed to exclude from the 2001 sample all trade lines that
should be present in the 2005 archives.
4.0

Processing the Initial Sample

For each archival pull, all available individual depersonalized credit records, including trade lines,
inquiries, and public records (collectively, TIPs) associated with all borrowers accompanying any
initial sample trade line were provided regardless of the archive from which it was sampled. The
data provided by Experian are de-identified and contain no directly identifying personal
information such as name, address, or Social Security number. The credit records were tagged
with de-identified borrower numbers (DINs) and servicer and loan numbers (both in encrypted
form). 6 These could be used (imperfectly) to link TIP files to other account-level files both
within an archive and over time.
6

The encrypted servicer identification and loan numbers are used only by the NMDB development team primarily
to update the database each quarter. They are not available to dataset users even in encrypted form. This is done to
ensure compliance with the contract restriction that the database not be used for enforcement against servicers. The
borrower DINs are unique to the NMDB and are randomized. Experian, however, maintains the mapping between
the borrower identification numbers used in their system and the DINs supplied to the NMDB team so that records
in the NMDB associated with the same DIN will be associated with the same borrower ID in the Experian records.

5

One major problem encountered with the NMDB sample frame is that a single mortgage can be
associated with multiple trade lines. This can arise when the servicing of the loan is sold or
transferred and the trade line reported by the original servicer is not properly linked to the trade
line reported by the new servicer. In such cases, borrowers may appear to have multiple
mortgages, when, in fact, they have only one. Because of these duplicates, randomly sampling
trade lines will result in mortgages with multiple records being over represented in the data. To
correct for this, a processing methodology was developed to identify and combine multiple
records that contain information about the same mortgage into one record.
The first step in the process of eliminating duplicate mortgage records (“de-duping”) was to find
multiple trade lines for the same mortgage in the same archive. From these duplicates, sample
loans were removed when the selected trade line was not the one with the latest “Account Balance
Date” (this corrects for the problem of having mortgages associated with multiple trade lines
over-represented in the sample). The second step was de-duping across archives. The June 2012,
December 2005, and July 2001 samples were treated as sequential NMDB sample frames (in that
order) whereby mortgages selected from a NMDB sample frame later in the order (e.g., July
2001) that can be found in a NMDB sample frame earlier in the order (June 2012 or December
2005) would be removed from the sample (again, this corrects for the fact that such mortgages are
over-sampled in the raw frame).
The de-duping process also dealt with the problem of ambiguous lien status for the “Enhanced
Account Type Codes” of 08 (Real estate, specific type unknown), 27 (Real estate mortgage, with
or without collateral, usually second mortgage), and 5A (real estate – junior liens and nonpurchase money first). Sample trade lines associated with these codes were removed from the
sample when they subsequently could be linked with trade lines that were unambiguously second
liens.
Once the initial samples were de-duped, it was necessary to link archival records over time to
create a composite picture of the performance of each sample loan. Semi-annual archives were
drawn for the period December 2001 to December 2011 for borrowers associated with the initial
sample loans. Data from these archives were patched together to create a temporal picture of each
loan. One issue that needed to be dealt with is that DINs for a given borrower can change over
time. There are times when a loan is first reported to the credit repositories and cannot be
connected with existing credit records for the borrower(s). This can happen because lenders make
errors in reporting names and addresses or because of changes to a borrower’s addresses or
names. In this instance Experian treats the loan as associated with a new borrower. In most of
these instances the records are ultimately reconciled with the correct existing borrower and a
“DIN-merge” occurs. However, historical archives are stored with the DINs at the time of the
archive. Thus, to properly connect borrowers (and mortgages) over time, it was necessary for
Experian to provide a DIN-merge transformation table to map historical to current DINs.
As shown in Table 1, the de-duping process substantially reduced the size of the original NMDB
sample. About 15 percent of the mortgage trade lines originally sampled from the June 2012
archive, more than a quarter of the selections from the 2005 archive, and almost three-quarters of
the selections from the 2001 archive were dropped. The percentages were higher for the older
archives since many of the loans selected from them were selected because they were not current
6

at the date of the archive and thus subject to the FCRA purge rules. However, many of these
loans subsequently became current and could be found in later archives.
Table 1
Archive Date
July 2001
Dec 2005
June 2012

5.0

Sample Tradelines
302,398
2,955,675
9,225,304

Final Loans
86,797
2,158,188
7,794,176

Final Borrowers
133,127
3,520,538
12,169,729

Percentage
Dropped
71.3 %
27.0
15.5

Updating the Sample

Under the NMDB sample design going forward, credit records for borrowers associated with
sampled mortgages are to be collected quarterly until one year after the mortgage is reported as
closed. 7 As of June 2012, approximately 3 million loans from the initial sample were still active
or had been closed less than a year. In addition, to keep the NMDB up-to-date, it is necessary to
add a representative sample of the new mortgages reported to Experian each quarter to the
database.
The initial update of the NMDB from the June 2013 archive covered a full year of newly-reported
mortgages since June of 2012. Since that date, updates have taken place quarterly drawing from
the last archive of the quarter (March, June, September or December). Each quarterly update
follows the same pattern. A 1-in-20 random sample of closed-end first-lien mortgage trade lines
is drawn. These loans, which are identified using the same criteria as was used for the June 2012
archive, are selected from among the loans that were newly reported to Experian since the date of
the previous quarterly update archive. The new sample is de-duped using the same methodology
as used for the initial sample. If multiple trade lines are identified for the mortgage and the
selected mortgage is not the one with the latest “Account Balance Date” or the mortgage is
deemed to be a second lien then it is dropped. In addition, checks are run to determine if the
mortgage was already reported in an earlier archive period (perhaps as a different trade line). If
so, the loan is dropped.
Existing sample loans are also updated each quarter. Prior to the update, the DIN-merge
transformation table is updated to account for “newly merged” DINs. To ensure that lagged
information for all DINs newly added to the dataset is collected, the year-old archive is drawn
each quarter for all active DINs for which this archive had not previously been collected.
At present between 75,000 and 80,000 new loans are added to the NMDB each quarter (see Table
2). The number of mortgages added to the database is only about two-thirds of the raw trade lines
originally selected for the update sample.

7

A partial update is done monthly collecting only limited performance data for active sample mortgages. This
allows the database to provide high-frequency information on mortgage delinquency rates.

7

Table 2
Archive Date
June 2013
Sept. 2013
Dec. 2013
Mar. 2014
June 2014
Sept. 2014
Dec. 2014
Mar. 2015
June 2015

6.0

Sample Tradelines
648,224
240,001
174,404
111,928
146,406
124,389
124,323
104,613
129,737

Final Loans
499,466
132,336
110,326
54,564
79,800
76,911
77,792
75,284
93,822

Final Borrowers
775,732
201,641
163,897
80,962
118,042
114,294
115,078
111,859
139,886

Percentage
Dropped
22.9 %
44.9
36.7
51.3
45.5
38.2
37.4
28.0
27.7

Merging with other Data Sources

Although extensive, Experian’s archive files do not contain information on a number of key
mortgage features, such as the loan’s purpose (home purchase or refinance), whether it had an
adjustable or fixed rate, its securitization status, its origination channel (broker or retail lender), or
whether it was for an owner-occupied property, vacation home or investor property. Moreover,
Experian’s archives contain no information on the property backing the mortgage, such as its
location, purchase price, characteristics, or current value. Finally, key information on borrowers
associated with the loan including income is also missing. Consequently, values of these key
variables need to be inferred indirectly or acquired from other data sources if they are to be
included in the NMDB.
The NMDB expects to obtain much of the missing information from matches to administrative
file records. Predominantly the administrative files come from government-affiliated mortgage
programs including Fannie Mae and Freddie Mac (the Enterprises), and tentatively, the Federal
Housing Administration (FHA), and the U.S. Department of Veterans Affairs (VA). Collectively,
loans associated with these programs comprise about three-quarters of the loans in the NMDB.
The most accurate means of merging information from outside sources into the NMDB would be
to use information about the borrowers, such as their names, Social Security numbers, addresses,
and dates of birth. Using such directly identifying information (DII), however, would heighten
concerns about data security and borrower privacy. Consequently, FHFA contracted with an
outside consultant to conduct a study of how such concerns might be mitigated. The third-partyblind matching process that FHFA used is consistent with the “best practices” and
recommendations from that study.
The third-party-blind matching process adheres to three guiding principles. First, neither FHFA
nor the Enterprises can receive DII from Experian. Second, Experian cannot access both
Enterprise administrative data and borrower DII in the same place. Third, FHFA must not be able
to match loans in the NMDB records to the specific administrative records from the Enterprises.

8

In December 2014, a process was initiated to supplement the NMDB data with administrative
data from Fannie Mae and Freddie Mac. The process for matching the data from the Enterprises
followed seven steps:
(1) The Enterprises created a unique anonymized identifier (AID) for each loan. This
identifier, along with the borrower-level DII associated with each loan (including name,
address, Social Security number, and date of birth), was transmitted directly to Experian
using a secure portal. FHFA did not receive this information. Other administrative data
on these loans were not sent to Experian.
(2) The Enterprises sent the AID, along with administrative data for each loan, to an
FHFA data processing unit that is separate from the NMDB development team. No
borrower-level DII was included in the information sent to the FHFA data processing
unit.
(3) Behind a secure firewall to protect FCRA-regulated data, Experian matched the DII it
received from the Enterprises to the DII maintained in its own files on the borrowers in
the NMDB to determine potential matches. When a potential match was identified,
Experian compiled the DIN for each matched borrower.
(4) For all potential matches, Experian transferred the Enterprise-supplied AID and the
matched NMDB borrower DINs to a separate unit within Experian that had no access to
the credit repository data or any DII.
(5) The second Experian unit sent the list of matched AIDs to the data processing unit in
FHFA that received the administrative data from the Enterprises in step (2). For each
AID it received, this data processing unit sent back the associated administrative data that
it received from the Enterprises.
(6) After receiving this information, the second Experian unit forwarded the
administrative data they received from the data processing unit at FHFA, plus the
matched borrower DIN that they received from the first Experian unit, to the NMDB
development team at FHFA. The information sent to the NMDB development team
included neither the Enterprise-created AID nor any DII.
(7) The NMDB team compared the characteristics of the loans associated with the DINs
received from the second Experian unit to the administrative information on the loans. If
the information from both sources was consistent, the match was confirmed. A list of
confirmed matches was sent to Experian. Upon confirmation, Experian stored the
property address supplied as part of the DII file from the Enterprises but otherwise
permanently destroyed all DII used in the match.
The figure below illustrates the third-party-blind matching process.

9

NMDB Administrative Data Merging Process
Enterprises

Admin. Data and AID
(2)

DII and AID
(1)

FHFA

Experian

Data Processing Team
Firewall
NMDB Development Team
(7)

AID of
Matched Loans
(5)

DII Matching Team
(3)
Firewall

Admin. Data and AID
of Matched Loans
(5)
DIN and Admin. Data
(6)

Second Team

AID and
Matched DIN
(4)

As of this writing, results of the Enterprise administrative file matching are still being processed.
Negotiations are also underway to use similar methods to match FHA and VA loans with the
NMDB. Contracts have also been signed to merge property record information into the NMDB,
using similar third-party blind matching techniques. Data from servicing and private-label
databases will also be matched which should provide missing data elements for most of the nongovernment-affiliated loans in the NMDB. 8
It is anticipated that additional matching will be conducted to enhance the NMDB with
information from the Home Mortgage Disclosure Act (HMDA) data, private mortgage insurance
companies, the Rural Housing Service, and the Federal Home Loan Banks. 9 These matches will
likely not involve DII and those will have to reply on less accurate techniques.
Ultimately, the NMDB will combine data from all of these sources into a common file with one
record per sampled loan. The record will contain variables reflecting all the static characteristics
8

To facilitate the property matching, the entire property database of one of the two largest U.S. property data
vendors has been placed behind the secure firewall at Experian. This allows information on borrower name and
address to be used in the matching process. Again, any DII used in the match will be discarded once the matching
process is completed.

9

Such merges will use information common to both datasets to perform a match but not DII. Most of the matches
contemplated for the NMDB will rely on the original loan balance, the opening date of the mortgage and the general
location of the property (census tract, ZIP Code or state/county). Unfortunately, mortgage servicers report the
billing address of the mortgage borrowers to Experian, but this is not necessarily the property address, particularly
for mortgages on non-owner occupied properties. Additional address information maintained within Experian’s
databases may prove useful in supplementing the repository addresses, as might historical information on borrower
location. Nevertheless it is expected that such merges will be less accurate than those employing DII because the
later are less reliant on address.

10

of the loan, culled from multiple sources, as well as vectors of dynamic data, such as the monthly
performance of the loan from origination to termination, changes to its interest rate in each month
(if a variable rate loan), and the associated loan balances. It should be noted that information
from external databases is only used to supplement information about sample loans, not to add
new loans to the sample. The NMDB sample frame will continue to be that established in the
Experian data files. All information on mortgage performance will likewise come from Experian.
7.0

Evaluating the NMDB Sample Frame

A complete evaluation of the NMDB sampling frame may not be possible until the database is
fully developed. However, at this stage of development the NMDB can be compared with
HMDA data as alternative estimates of the U.S. mortgage origination market. Table 3 compares
estimates of national quarterly origination totals from HMDA data and the NMDB. Loans are
divided into two groups, based on whether they exceeded $75,000 in real 2013 dollars. Smaller
loans are separated out because HMDA did not differentiate between first and second liens
(which are generally small) prior to 2004.
The two databases track each other remarkably well, with HMDA totals slightly below those of
the NMDB. This may stem from known gaps in HMDA’s coverage. Loan originators that are
very small or that operate exclusively in rural areas are exempt from HMDA’s reporting, so their
lending activity is not included in the HMDA data. Additionally, HMDA excludes commercial
loans and (non-purchase) loans backed by properties that were previously mortgage-free. Many
of these loans, however, may not be reported to the credit repositories either. For example, loans
to corporations, loans made as part of a seller-financed property sale, and loans made by nontraditional lenders are unlikely to be in either database. Moreover, some types of loans may be
missed by the NMDB though they are captured in HMDA. Lenders that retain all of their loans in
portfolio, particularly credit unions, are known not to report their loans to the credit repositories,
but are nevertheless still subject to HMDA reporting.

11

Table 3
Quarterly Loan Originations (1,000s of loans)

Quarter
1998-1
1998-2
1998-3
1998-4
1999-1
1999-2
1999-3
1999-4
2000-1
2000-2
2000-3
2000-4
2001-1
2001-2
2001-3
2001-4
2002-1
2002-2
2002-3
2002-4
2003-1
2003-2
2003-3
2003-4
2004-1
2004-2
2004-3
2004-4
2005-1
2005-2
2005-3
2005-4

$75,000 and Under
HMDA
NMDB *
616
696
755
893
710
835
705
816
601
736
661
862
622
776
544
669
490
619
571
779
525
698
473
608
463
607
610
872
557
769
574
757
507
665
504
702
507
699
540
717
514
678
630
889
650
911
430
634
375
314
460
392
399
318
343
299
317
266
389
310
394
317
321
273

Over $75,000
HMDA
NMDB *
1,955
1,950
2,278
2,247
2,229
2,179
2,808
2,676
2,263
2,122
2,143
2,046
1,750
1,668
1,418
1,349
1,197
1,162
1,470
1,466
1,470
1,427
1,451
1,384
1,995
1,922
2,814
2,789
2,633
2,542
3,558
3,434
2,909
2,791
2,541
2,521
3,573
3,484
4,778
4,608
4,443
4,326
5,579
5,443
5,624
5,577
2,966
2,989
2,873
2,812
3,518
3,538
2,747
2,734
2,816
2,798
2,611
2,537
3,137
3,057
3,443
3,346
2,832
2,760

12

NMDB/HMDA
(Percent Ratio)
≤ $75,000
> $75,000
88.6 %
100.3 %
84.6
101.4
85.0
102.3
86.3
105.0
81.7
106.6
76.7
104.7
80.1
104.9
81.3
105.2
79.2
103.0
73.3
100.3
75.2
103.0
77.7
104.8
76.3
103.8
69.9
100.9
72.5
103.6
75.8
103.6
76.2
104.2
71.7
100.8
72.5
102.5
75.2
103.7
75.8
102.7
70.9
102.5
71.4
100.8
67.8
99.2
119.5
102.2
117.4
99.5
125.3
100.5
114.7
100.6
119.1
102.9
125.6
102.6
124.2
102.9
117.4
102.6

Quarterly Loan Originations (1,000s of loans)

Quarter
2006-1
2006-2
2006-3
2006-4
2007-1
2007-2
2007-3
2007-4
2008-1
2008-2
2008-3
2008-4
2009-1
2009-2
2009-3
2009-4
2010-1
2010-2
2010-3
2010-4
2011-1
2011-2
2011-3
2011-4
2012-1
2012-2
2012-3
2012-4
2013-1
2013-2
2013-3
2013-4

$75,000 and Under
HMDA
NMDB *
287
245
373
289
360
273
282
232
251
210
322
245
291
222
231
198
215
195
251
220
217
191
153
142
161
156
217
197
200
183
173
167
137
135
189
176
179
176
195
195
158
166
190
186
201
203
205
217
199
207
231
235
241
249
244
258
238
251
263
272
242
247
175
185

Over $75,000
HMDA
NMDB *
2,370
2,284
2,729
2,601
2,582
2,450
2,539
2,397
2,240
1,997
2,402
2,221
1,929
1,848
1,692
1,642
1,874
1,759
1,762
1,695
1,283
1,254
1,152
1,098
2,126
2,017
2,542
2,444
1,786
1,750
1,761
1,695
1,315
1,291
1,551
1,524
1,921
1,871
2,235
2,192
1,387
1,376
1,253
1,249
1,544
1,533
1,901
1,885
1,846
1,828
2,031
2,032
2,310
2,279
2,431
2,389
2,189
2,173
2,271
2,264
1,753
1,748
1,201
1,216

*National estimate based on 1-in-20 sample.

13

NMDB/HMDA
(Percent Ratio)
≤ $75,000
> $75,000
116.8 %
103.8 %
129.2
104.9
131.9
105.4
121.6
105.9
119.6
112.2
131.6
108.1
131.0
104.4
116.7
103.0
110.0
106.6
114.3
104.0
113.8
102.3
107.7
104.9
103.1
105.4
110.4
104.0
109.0
102.0
103.8
103.9
101.2
101.9
107.7
101.8
101.8
102.7
100.2
101.9
95.3
100.8
102.0
100.3
98.8
100.7
94.7
100.8
96.4
101.0
98.1
99.9
96.8
101.4
94.5
101.8
94.8
100.7
96.8
100.3
98.0
100.3
94.5
98.7

Appendix
The primary sources explored were the Home Mortgage Disclosure Act (HMDA) data, the
Federal Reserve Bank of New York's Equifax Consumer Credit Panel, the CoreLogic property
database, the servicing databases owned by CoreLogic and Black Knight Financial Services, and
data available from the three national credit repositories—Experian, Equifax, and TransUnion.
Public survey databases, particularly the American Housing Survey (AHS), were also considered.
All of these sources share several desirable features such as: (1) the databases are de-identified
containing no direct-identifying information such as borrower name, address, or Social Security
number; (2) they are collected for other purposes, thus their use entails no new data collection
from lenders, servicers or borrowers; and (3) all of them have been collected for a period of time
and are expected to continue into the future.
However, each was also found to be deficient in significant ways.
The HMDA data include loan applications and underwriting outcomes for most mortgages with
selected information about the loan, property, and borrower. The data are arguably the most
representative publicly available existing data source about the mortgage market. However, the
HMDA data contain no information on loan performance, little information on borrower creditworthiness, and have up to a 21-month delay in release. The CoreLogic property database suffers
from similar deficiencies. Although it has widespread coverage, the database contains very
limited information on mortgage characteristics or performance and nothing on the borrower.
The Federal Reserve Bank of New York’s Equifax Consumer Credit Panel provides a nationally
representative 1-in-20 sample of individuals with credit records, observed quarterly from 1999
onward. However, mortgage loans are often represented by duplicate trade lines and important
information is missing, such as loan purpose, owner-occupancy, pricing, loan-to-value ratio,
income, and borrower demographics. Finally, these data are accessible at present only to the
Federal Reserve System.
CoreLogic and Black Knight Financial Services produce loan-level databases with performance
information collected from mortgage servicers. The servicing fields available from CoreLogic
and Black Knight are relatively comprehensive in both variables and coverage: the CoreLogic
database claims about 32 million active mortgage loans, while the Black Knight database claims
about 31 million active mortgage loans. However, these data offer no assurance of being
representative, as data are only collected from about 25 servicers each. Moreover, mortgages
cannot be tracked if servicing is transferred. Other drawbacks include minimal borrower
demographics and no information on other borrower’s obligations.
The semi-annual AHS contains comprehensive information on a nationally representative 1-in2,000 sample of mortgages of owner-occupied properties with very good information about the
property and borrower demographic. However, the AHS has only limited information about the
mortgage itself. As with the other nationally representative consumer survey data sources, AHS
contains no information on mortgage performance, provides only a small number of observations,
and is released with a significant lag.
14

The credit repository data from Equifax, Experian, and TransUnion are rich in credit information.
By construction they incorporate data on credit card debt, installment loans, credit inquiries, and
public records for the consumers they have in their respective databases. Their data can be linked
to marketing datasets that provide borrower characteristics including age, gender, and marital
status which, if validated, could be of potential use in a dataset. The credit repositories also
maintain data on borrowers' changes of address and broader geographic classifications, such as
the census tract. However, there are important areas that are not covered. They lack some
information on borrowers (e.g., race/ethnicity and income), mortgages (e.g., loan product and
contract rate), and the underlying property (e.g., location and value).
Given the foregoing, FHFA and CFPB, along with other organizations most notably HUD, the
Federal Reserve Board and Freddie Mac, decided that a modified derivative of the credit
repository data offered the best source from which to construct a nationally representative
comprehensive mortgage database. The three credit repositories all actively pursue loan servicers
as data providers. As a result, they obtain information on almost the entire population of nonprivate mortgage loans made in the United States. Furthermore, they archive their data, making it
possible to “jump start” the data collection process by going back in time, collecting data in
almost the same fashion as if it had taken place in real time.
As part of the exploratory process, using a competitive procurement process, Experian was
engaged by Freddie Mac to construct a prototype to confirm the appropriateness of using credit
repository data for the database. This effort confirmed the concept but suggested that a number of
steps needed to be taken in order to meet the design objectives.
First, it was recommended that the database should be a sample rather than a universal registry of
loans. Second, while these data contain detailed information on loan performance and other
borrower credit obligations, they are missing critical data items needed for the database such as
the location and features of the property, demographics, and loans characteristics such as whether
the loan had an adjustable- or fixed-rate mortgage and whether the loan was a refinance or for a
home purchase. Thus, it would be necessary to access other data sources and merge information
gleaned from them with the repository data in order to make the database comprehensive. Pilot
testing also confirmed that the best method of merging data would rely on third-party blind
matching conducted behind a firewall at the credit repositories.

15