Datasets:

FinanceRateCalc
/

fha-denial-timeseries-2018-2025

Error code:   DatasetGenerationCastError
Exception:    DatasetGenerationCastError
Message:      An error occurred while generating the dataset

All the data files must have the same columns, but at some point there are 1 new columns ({'state'}) and 3 missing columns ({'originated', 'denied', 'lender'}).

This happened while the csv dataset builder was generating data using

hf://datasets/FinanceRateCalc/fha-denial-timeseries-2018-2025/fha_state_denial_timeseries_2018_2025.csv (at revision e260fe28607fbfcbd838eb9aa2b3800a8b6bc0d2), ['hf://datasets/FinanceRateCalc/fha-denial-timeseries-2018-2025@e260fe28607fbfcbd838eb9aa2b3800a8b6bc0d2/fha_lender_denial_timeseries_2018_2025.csv', 'hf://datasets/FinanceRateCalc/fha-denial-timeseries-2018-2025@e260fe28607fbfcbd838eb9aa2b3800a8b6bc0d2/fha_state_denial_timeseries_2018_2025.csv', 'hf://datasets/FinanceRateCalc/fha-denial-timeseries-2018-2025@e260fe28607fbfcbd838eb9aa2b3800a8b6bc0d2/lender_persistence_2018_2025.csv']

Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1848, in _prepare_split_single
                  writer.write_table(table)
                  ~~~~~~~~~~~~~~~~~~^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 765, in write_table
                  self._write_table(pa_table, writer_batch_size=writer_batch_size)
                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 773, in _write_table
                  pa_table = table_cast(pa_table, self._schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              year: int64
              state: string
              denial_rate_pct: double
              -- schema metadata --
              pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 625
              to
              {'year': Value('int64'), 'lender': Value('string'), 'originated': Value('int64'), 'denied': Value('int64'), 'denial_rate_pct': Value('float64')}
              because column names don't match
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1850, in _prepare_split_single
                  raise DatasetGenerationCastError.from_cast_error(
                  ...<4 lines>...
                  )
              datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
              
              All the data files must have the same columns, but at some point there are 1 new columns ({'state'}) and 3 missing columns ({'originated', 'denied', 'lender'}).
              
              This happened while the csv dataset builder was generating data using
              
              hf://datasets/FinanceRateCalc/fha-denial-timeseries-2018-2025/fha_state_denial_timeseries_2018_2025.csv (at revision e260fe28607fbfcbd838eb9aa2b3800a8b6bc0d2), ['hf://datasets/FinanceRateCalc/fha-denial-timeseries-2018-2025@e260fe28607fbfcbd838eb9aa2b3800a8b6bc0d2/fha_lender_denial_timeseries_2018_2025.csv', 'hf://datasets/FinanceRateCalc/fha-denial-timeseries-2018-2025@e260fe28607fbfcbd838eb9aa2b3800a8b6bc0d2/fha_state_denial_timeseries_2018_2025.csv', 'hf://datasets/FinanceRateCalc/fha-denial-timeseries-2018-2025@e260fe28607fbfcbd838eb9aa2b3800a8b6bc0d2/lender_persistence_2018_2025.csv']
              
              Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

year int64	lender string	originated int64	denied int64	denial_rate_pct float64
2,018	Planet Home	1,019	444	30.35
2,018	loanDepot	30,200	20,120	39.98
2,018	Guild	15,311	1,187	7.19
2,018	Rocket	66,007	35,212	34.79
2,018	NewRez	5,469	5,238	48.92
2,018	UWM	20,704	3,861	15.72
2,018	Mr. Cooper	9,697	5,072	34.34
2,018	Freedom	19,454	18,703	49.02
2,018	CrossCountry	8,746	824	8.61
2,018	Wells Fargo	2,303	2,020	46.73
2,018	PennyMac	3,138	3,510	52.8
2,019	Mr. Cooper	33,390	10,470	23.87
2,019	Freedom	40,677	9,626	19.14
2,019	PennyMac	8,333	7,683	47.97
2,019	UWM	49,197	6,857	12.23
2,019	Guild	18,904	1,281	6.35
2,019	loanDepot	25,472	12,727	33.32
2,019	Planet Home	3,514	891	20.23
2,019	Rocket	93,525	39,496	29.69
2,019	NewRez	8,098	4,371	35.05
2,019	Wells Fargo	2,289	2,018	46.85
2,019	CrossCountry	12,968	760	5.54
2,020	loanDepot	25,814	8,389	24.53
2,020	PennyMac	13,631	9,844	41.93
2,020	Mr. Cooper	41,451	10,447	20.13
2,020	Rocket	83,289	36,256	30.33
2,020	Guild	21,063	1,473	6.54
2,020	CrossCountry	23,396	1,352	5.46
2,020	Wells Fargo	1,503	1,598	51.53
2,020	Freedom	137,000	14,924	9.82
2,020	NewRez	13,840	5,012	26.59
2,020	Planet Home	5,688	1,157	16.9
2,020	UWM	24,042	5,306	18.08
2,021	Wells Fargo	1,272	1,362	51.71
2,021	Freedom	117,575	9,702	7.62
2,021	NewRez	17,147	8,396	32.87
2,021	Mr. Cooper	33,738	8,597	20.31
2,021	UWM	38,516	7,130	15.62
2,021	CrossCountry	25,540	1,793	6.56
2,021	Rocket	127,771	49,320	27.85
2,021	loanDepot	25,844	10,429	28.75
2,021	PennyMac	31,773	10,508	24.85
2,021	Guild	18,543	1,519	7.57
2,021	Planet Home	5,609	1,314	18.98
2,022	Rocket	73,003	34,165	31.88
2,022	Guild	11,221	1,198	9.65
2,022	loanDepot	25,288	14,644	36.67
2,022	Planet Home	2,906	1,127	27.94
2,022	NewRez	5,627	9,044	61.65
2,022	UWM	47,735	9,884	17.15
2,022	PennyMac	12,930	6,386	33.06
2,022	Freedom	17,217	4,978	22.43
2,022	CrossCountry	20,009	1,471	6.85
2,022	Wells Fargo	1,746	2,666	60.43
2,022	Mr. Cooper	11,245	4,232	27.34
2,023	CrossCountry	21,064	1,450	6.44
2,023	Guild	11,986	1,054	8.08
2,023	UWM	64,537	13,529	17.33
2,023	Mr. Cooper	6,120	1,980	24.44
2,023	Planet Home	2,237	895	28.58
2,023	Rocket	68,367	27,894	28.98
2,023	PennyMac	9,251	4,568	33.06
2,023	Freedom	9,384	4,642	33.1
2,023	loanDepot	19,411	12,128	38.45
2,023	NewRez	8,901	7,996	47.32
2,023	Wells Fargo	1,501	1,937	56.34
2,024	CrossCountry	22,994	1,342	5.5
2,024	Guild	17,932	1,375	7.1
2,024	UWM	74,402	18,484	19.9
2,024	Planet Home	3,246	997	23.5
2,024	Freedom	23,149	7,488	24.4
2,024	Mr. Cooper	5,517	1,862	25.2
2,024	PennyMac	15,078	5,466	26.6
2,024	Rocket	66,721	26,429	28.4
2,024	loanDepot	20,996	12,339	37
2,024	NewRez	9,596	8,804	47.8
2,024	Wells Fargo	848	1,197	58.5
2,025	CrossCountry	27,608	1,877	6.37
2,025	Guild	20,919	1,609	7.14
2,025	UWM	87,294	24,022	21.58
2,025	Planet Home	5,823	1,303	18.29
2,025	Freedom	27,636	8,159	22.79
2,025	Mr. Cooper	5,480	2,024	26.97
2,025	PennyMac	24,044	7,444	23.64
2,025	Rocket	65,337	28,619	30.46
2,025	loanDepot	22,700	12,191	34.94
2,025	NewRez	10,547	12,022	53.27
2,025	Wells Fargo	1,295	1,226	48.63
2,018	null	null	null	24.34
2,018	null	null	null	24.81
2,018	null	null	null	25.05
2,018	null	null	null	25.48
2,018	null	null	null	25.56
2,018	null	null	null	25.82
2,018	null	null	null	28.08
2,018	null	null	null	28.09
2,018	null	null	null	31.22
2,018	null	null	null	32.19
2,018	null	null	null	32.29
2,018	null	null	null	32.48

End of preview.

FHA Lender Denial Time Series — Eleven of the Largest Lenders, 2018–2025

Annual FHA mortgage denial rates for eleven of the largest US FHA lenders across the eight HMDA reporting years 2018–2025 — a balanced panel of 88 lender-year cells computed from the public CFPB Home Mortgage Disclosure Act record. Lenders: CrossCountry, Freedom, Guild, Mr. Cooper (Nationstar), NewRez, PennyMac, Planet Home, Rocket (Quicken Loans through 2021), UWM, Wells Fargo, loanDepot. Entity continuity across rebrandings is maintained via Legal Entity Identifier.

Headline finding: the door hierarchy barely moves. CrossCountry is the softest of the eleven doors in 7 of 8 years and Guild is the runner-up in 7 of 8 (the two swap in 2018) — while Wells Fargo and NewRez occupy the strict end throughout the second half of the window. 80.8% of pooled variance in the panel is between lenders rather than within lenders over time; the year-over-year Spearman rank correlation of the lender ordering has a median of 0.918.

Denominator — read this before comparing to other tables

Rates in this series are denied / (originated + denied): HMDA action 3 divided by actions 1 + 3. Applications approved but not accepted (action 2 — the lender said yes, the borrower walked away) are excluded, because they record a borrower's choice rather than a lender's decision posture, and their share varies across institutions.

This intentionally differs from the {1, 2, 3} "decisioned" universe used in FinanceRateCalc's 2025 cross-sectional tables and papers. Mixing the two denominators in one comparison is a common error in published work — do not do it. Levels differ modestly for most lenders and most where the approved-not-accepted share is large (Freedom 2025: 22.8% under this definition vs 19.9% under {1,2,3}). Every within-series comparison (ranks, variance shares, breaks) is unaffected by the choice.

Files

fha_lender_denial_timeseries_2018_2025.csv

year — HMDA reporting year (2018–2025)
lender — short institution name (see list above for legal names)
originated — loans originated (action 1)
denied — applications denied (action 3)
denial_rate_pct — denied / (originated + denied) × 100

lender_persistence_2018_2025.csv

Per-lender statistics from the companion paper (below):

classification — systematic_shift / systematic_shift_with_trend / emerging_shift / stable, by published magnitude thresholds (≥5.0 pp best-break shift; Mann–Kendall p < 0.05 with |Sen slope| ≥ 1.0 pp/yr)
level_tier — low (<15%), mid (15–35%), high (>35%) by eight-year mean
eight_year_mean_pct, break_between, shift_pp, mk_p, sen_pp_per_yr, sd_pp, rates (2018→2025, pipe-separated)

Classification is by magnitude, not p-values: cells aggregate large counts, so nearly any difference is statistically significant; test statistics are reported for transparency.

Companion working paper (August 2026)

Yetiş, Z. (2026). Persistent Doors: Denial-Rate Stability and Structural Breaks Across Eleven of the Largest FHA Lenders, 2018–2025. SSRN Working Paper 7341481. https://ssrn.com/abstract=7341481· doi:10.2139/ssrn.7341481

Key results computed from exactly this series: the persistence statistics above, plus four single regime shifts of ≥5 points — Freedom −19.0 (2019–20) and PennyMac −18.6 (2020–21, the panel's only significant monotone trend, Mann–Kendall p = 0.009, Sen −4.1 pp/yr) around the refinancing boom; NewRez +18.1 and Wells Fargo +7.9 both at the 2021–22 rate-regime boundary. All estimates are associational; see the paper's Limitations.

The cross-sectional companion: Yetiş, Z. (2026). The Door Effect (38% of explained variation in 2025 FHA denial outcomes attaches to lender identity). SSRN 7309319, doi:10.2139/ssrn.7309319 · dataset: https://huggingface.co/datasets/FinanceRateCalc/door-effect-fha-2025

Source, caveats, citation

Underlying data: public CFPB HMDA loan/application records, 2018–2025 (https://ffiec.cfpb.gov/). Denial rates partly reflect applicant mix, channel and product composition (notably refinance share through the 2020–21 boom); a high rate is a historical observation, not evidence of wrongdoing, and nothing here supports a prediction about any individual application.

Yetiş, Z. (2026). FHA Lender Denial Time Series, 2018–2025. FinanceRateCalc. https://financeratecalc.com — companion papers: SSRN 7341481, SSRN 7309319.

License: CC BY 4.0. Independent; no lender partnerships, no advertising. A denial is a data point, not a verdict on the applicant.

Downloads last month: 10

Total file size:

50 kB