Ginnie Mae pool factor download: the guaranteed book's speed ledger, delivered as rows

A pool factor is the share of a securitized pool's original principal still outstanding, stated per security. In Ginnie Mae's disclosure universe it arrives as the rpb_factor field on every pool and security row: remaining principal balance divided by original aggregate amount, 1.0000 at issuance and declining thereafter as borrowers pay on schedule and prepay off it. A reading of 0.4217 - the seasoned single-family pool in the sample below - says roughly 42 percent of the original dollars are still working.

The factor never travels alone. On the same row sit the nine-character cusip, the six-character pool_id, issuer_name attribution, original_aggregate_amount and remaining_security_rpb, and the weighted averages that summarise the collateral: wa_interest_rate_wac coupon, wa_remaining_months_to_maturity_warm, wa_loan_age_wala seasoning and wa_credit_score. That pairing is what makes the field analytically useful rather than merely descriptive - every factor observation arrives already attributed to an issuer, a coupon bucket and a seasoning cohort.

Two records in Datadory's thirteen-record commercial & residential mortgage finance pool carry the factor, and they solve different problems.

Ginnie Mae Disclosure Data and Reports (quality 9/10) is the bulk layer: full-universe files at pool, security and loan grain across six file families - MBS Single Family, HMBS, Multifamily, Platinum, Factor Files and Other Files - with the pool supplemental family adding 32 record types of stratification. It answers portfolio-sized questions offline, which is why prepayment desks treat it as the backbone rather than one input among many.

Ginnie Mae Investor Disclosure Portal (7/10) is the interactive counterpart: every MBS, HMBS, REMIC or Platinum security among the agency's hundreds of thousands of outstanding pools is reachable by CUSIP or pool number, returning the current factor with a twelve-month history, reorganized by WHFIT tax year when the question is tax reporting, and backed by Disclosure Plus dashboards spanning seven report areas from CPR Analysis to Cohort Analysis.

The two share keys, so they belong on the same project: a desk marking one bond tonight works interactively, while a book of hundreds of pools feeding a prepayment model wants the full-universe rows. Datadory delivers both as the same shaped output, so the altitude is a scope decision rather than an integration project.

Illustrative rows in the delivered shape - one single-family pool record, one interactive lookup result:

# pool/security record, single family
cusip                    : 36295XBY0      pool_id     : 683255
issuer_name              : Usaa Federal Savings Bank       issuer_number : 3536
issue_date               : 20191001       pool_indicator : C      pool_type : SF
original_aggregate_amount: 3,544,349      remaining_security_rpb : 3,544,349.58
rpb_factor               : 1.0000         number_of_loans_in_pool : 74
wa_interest_rate_wac     : 3.519%         wa_loan_age_wala : 349 mo

# security-level lookup, seasoned pool
cusip          : 36295XBY0     pool_factor : 0.4217     rpb : 28,450,000
security_type  : MBS Single Family           interest_rate : 4.50%
cpr            : 7.2           delinquency_rate : 2.31%  forbearance_trend : -0.40 pts

Read the shape before the digits - every value above is a documented example for its column. The first row is a pool at issuance: 74 loans, a 3.519 percent weighted-average coupon, factor 1.0000, nothing yet owed to time. The second is a seasoned security past its halfway mark, prepaying at 7.2 CPR with a forbearance tail shrinking by 0.40 points. Same key pair, two altitudes, and because both land as rows keyed on cusip and pool_id, a monitoring panel joins them without entity resolution.

Rows reflect Datadory's August 2026 review pass, and live rows for whichever pools, CUSIPs and periods you name arrive in this identical schema.

Twelve documented fields anchor the factor row set below; the full thirty-field dictionary travels with every delivery. At pool level the economics trio of original aggregate amount, remaining balance and rpb_factor sits beside the weighted averages. At loan level the Version 2.0 layout exposes what no other public source shows for this collateral - numbered elements through LL-55, including months_delinquent (LL-21), combined_ltv_cltv (LL-25), total_debt_expense_ratio (LL-26), credit_score (LL-27), state (LL-35), msa (LL-36) and removal_reason (LL-39), which distinguishes a delinquent-loan buyout from a foreclosure exit. Geography rides on every loan as state and MSA codes, so a national book cuts to a metro without any external join.

Depth is a property of the record, not something you assemble yourself. Monthly archives are retained per file prefix back to January 2020, which sets roughly a six-year floor on factor backtests for Ginnie Mae collateral, and pre-2020-format outputs continue alongside so long-run panels keep schema continuity across the boundary instead of breaking at it.

Scale tells you why normalisation matters before you write a line of modelling code. A recent single-family monthly portfolio file runs about 430 MB, its HMBS equivalent about 469 MB, the quarterly Loan Performance archive near 782 MB and the annual build near 2.3 GB - pipe-delimited text carrying pool, supplemental and loan-level records together. Delivered as typed rows, those same vintages become columns your warehouse already joins, and the archive question reduces to naming prefixes and periods.

Three altitudes come from one key space: one row per pool for fast full-universe scans, the 32-type supplemental family for stratified views, and true loan level - one row per loan per period - the finest public view of federally insured mortgage collateral anywhere. Pick the altitude per question rather than per project; the keys hold across all three.

You can compute realized speeds, and that is where the factor earns its keep - but the loan-level tapes are what turn those speeds into credit research. A pool's factor decline is an aggregate; splitting it into components needs the borrower attributes that ride on the loan rows: seasoning buckets from loan age, note-rate incentives from coupon versus prevailing rates, state and metro concentration from state and msa, distress overlays from months_delinquent and the re_performing_loan_indicator (LL-55) added in the August 2026 Version 2.0 release alongside scheduled_upb (LL-54).

Two cautions travel with the analysis. Figures are compiled from issuer submissions the agency does not independently verify, so cohort trends and cross-pool comparisons carry more signal than any single point estimate, and loan totals deserve reconciliation against pool-level factors before model output ships. Second, the program's design confines use to analysing Ginnie Mae MBS credit performance - analysis stays at pool, cohort and programme altitude, which is precisely where prepayment research lives anyway.

Definitions for the speeds themselves sit in the conditional prepayment rate entry, and the monthly CPR files in the Other Files family give you the agency-computed series to benchmark your own curves against.

The trade is depth on one guaranteed book versus breadth of the whole market, and the neighbouring records make good complements rather than substitutes.

Freddie Mac Single-Family Loan-Level Dataset (10/10) goes deeper per loan - roughly 56 million originations from January 1999 through March 2026 with monthly performance trails - but covers conventional collateral only, and its performance view stops at acquisition-quarter detail where Ginnie Mae's files carry ongoing monthly status to payoff. For government-backed speeds, Ginnie Mae is not one option among many; it is the record.

FHFA National Mortgage Database (NMDB) Aggregate Statistics (9/10) samples the whole US first-lien market from a nationally representative five percent sample, publishing performance tables quarterly from 2002 Q1 through 2026 Q1 across the 100 largest metros - the benchmark layer a single-guarantor tape cannot supply.

Federal Reserve Mortgage Debt Outstanding (8/10) splits all US mortgage debt across every holder group quarterly back to 1945 through roughly 100 successor series, and never descends below aggregates. Our head-to-head on the two works through the granularity-for-breadth swap row by row.

Context layers complete the desk. FRED Mortgage and Housing Data API (10/10) supplies the rate driver - the weekly 30-year average from its first April 1971 reading of 7.33% - and FHFA House Price Index (10/10) contributes collateral values, 90,000+ repeat-sales master-file observations from January 1975 down to census tracts. American Housing Survey National Microdata (9/10) adds household-reported mortgage terms across 16,834 records in the 2023 public-use file.

Fixed-income investors and quant researchers run factor histories and CPR curves per programme, coupon and vintage rather than programme-wide averages; the workflow breakdown lives at investors & quants use cases. Data scientists and ML engineers get a stable, documented feature source for prepayment, delinquency and forbearance models, with the 32-type supplemental family supplying ready-made stratifications; see data scientists use cases. Journalists, academics and students cite the federal guarantee programme's reach and borrower profile in auditable rows - see journalists & academics use cases. Around them, issuer and channel analysts track which issuers scale the FHA, VA and USDA channels quarter over quarter from per-pool issuer attribution, and multifamily due diligence teams read property-level exposure, prepayment penalties and terminated pools before a position is priced.

Because the hard part was never the first pull - it is the hundredth. Pipe-delimited text wearing familiar extensions that parse wrong by default. Weighted averages that move between pool and supplemental records depending on file family. A layout that reached Version 2.0 in August 2026, adding three elements and changing its delimiter in the same release. Archives whose format differs on either side of January 2020. Each is survivable once; none is fun to re-solve in every notebook.

API, files, or straight into your warehouse. Daily, weekly, or hourly - your call.

Name the pools, CUSIPs, programmes and periods when you request a sample and real factor rows come back cut to that scope, field dictionary attached - the schema in the sample is the schema you ship against.

Start with the commercial & residential mortgage finance data guide, which maps all thirteen pooled records and ranks ten of them, then browse the same catalog on the commercial & residential mortgage finance data hub or take the scored shortlist of the best mortgage finance datasets. The Ginnie Mae source profile covers the publisher behind both records.

Go deeper three ways: the Ginnie Mae Disclosure Data and Reports record documents every field in the dictionary above, the Ginnie Mae versus Fed Z.1 comparison settles the granularity-for-breadth choice, and the neighbouring threads on the loan level mortgage performance dataset and the house price index by ZIP code cover the credit tape and the collateral leg. When the rate path belongs in the same study, the 30-year mortgage rate history pairs week for week with factor data.