Discover Bank's loans fell 72% in 2025, from 34,694 to 9,711, ending six straight years of growth.
A sharp reversal after rapid growth
Similar lenders are state non-member banks making between half and twice as many loans; their originations rose 2% and the whole market's 10%, so its fall was the steepest of the group. Loans had climbed every year from 7,301 in 2018. A change of product or strategy could explain the drop.
Small loans, rarely for buying a home
Purchases were just 2.9% of its loans; other purposes (39.7%), home improvement (36.4%) and refinancing (21.0%) made up the rest. Its typical loan of $75,000 was the lowest of similar lenders, against $315,000.
Highest prices and first-mortgage denials among peers
12.6% of its loans were higher-priced (a rate well above the average for similar loans), against 2.5%, the highest of the group though down from 63.2% five years before; small home-improvement and other-purpose loans may carry higher rates than first mortgages. It turned down 90.3% of first-lien home-purchase applications on owner-occupied homes and 81.7% of refinance applications, against 6.5% and 13.9% for similar lenders, also the highest of the group. Such loans are a small part of its lending, and applicants seeking a full first mortgage may not fit the smaller loans it mainly makes.
Wide reach, modest LMI share
It lent in 49 states and 1,396 counties. LMI (low- and moderate-income) borrowers took 22.7% of loans, lower than about two-thirds of similar lenders, though up from 17.7%; its mix of loan purposes may draw a different set of borrowers.
Questions the data can answer next
Which states and counties account for the drop in its 2025 loans?
Similar lenders: same lender type, originations between half and twice the lender's in the data year.
Written from the 2025 HMDA data and checked figure by figure against it.
FIG. 1 — LOANS ORIGINATED, 2018–2025
9,711 loans in 2025, against a peak of 34,694 in 2024.
Originated means the loan was made: applications denied, withdrawn or not accepted are not counted. Discover Bank took 68,974 applications in 2025.
39.7% of its 2025 loans were other loans; 100.0% were Conventional.
Other 39.7% · Home improvement 36.4% · Refinance 21.0% · Purchase 2.9% Conventional 100.0% · of 9,711 loans. Purchase: buying a home. Refinance: replacing an existing mortgage, including cash-out. FHA: insured by the Federal Housing Administration. VA: guaranteed by the Department of Veterans Affairs. Conventional: neither.
SOURCE: FFIEC HMDA 2025, LOANS ORIGINATED
Where it lends, 2025
49 states, 390 metros, 1,396 counties
State
Loans
Share
California
813
8.4%
Florida
771
7.9%
Texas
649
6.7%
Illinois
493
5.1%
New York
467
4.8%
Georgia
466
4.8%
New Jersey
451
4.6%
Ohio
397
4.1%
Pennsylvania
392
4.0%
Virginia
317
3.3%
Largest metro areas, 2025
Loans originated
Metro area
Loans
Chicago-Naperville-Schaumburg
300
Atlanta-Sandy Springs-Roswell
245
Phoenix-Mesa-Chandler
221
Houston-Pasadena-The Woodlands
189
New York-Jersey City-White Plains
167
Riverside-San Bernardino-Ontario
167
Dallas-Plano-Irving
166
Lakewood-New Brunswick
151
Los Angeles-Long Beach-Glendale
146
Nassau County-Suffolk County
146
Identifiers
LEI (Legal Entity Identifier, the global company code HMDA has used since 2018)
X05BVSK68TQ7YTOSNR22
2018–2025
RSSD (the Federal Reserve's identifier for banks and their parents)
30810
2004–2023
Respondent ID (the HMDA filer code before 2018)
35649
1990–2009
Respondent ID (the HMDA filer code before 2018)
930810
2014–2017
Established 30 August 1911. Reported under HMDA 1990-2025.
Discover Bank and its peers, 2018–2025: every application in the HMDA data, cleaned and corrected, as Parquet files, with the Jenner programs that made these pages.
Every mortgage application reported under HMDA from 1990 to 2025, cleaned and corrected, every lender followed across name changes and mergers, with the market-share and fair-lending tables behind these pages and the Jenner programs that built them: the warehouse and the datamart, as Parquet files.
Jenner runs programs written in the SAS language. It drew every chart on these pages and fitted the fair-lending model, PROC GLIMMIX on 3.5 million applications. Run the programs that come with the data, or your own.