1003 mortgage application data extraction: sections, checks and automation

1003 mortgage application data extraction means reading the Uniform Residential Loan Application (Fannie Mae Form 1003, Freddie Mac Form 65) and turning its nine sections into structured data: the borrower, income, assets and debts, real estate owned, the loan and property, and the declarations. When the application is taken in a loan origination system (LOS), that data already exists. Extraction, or parsing, matters when the 1003 arrives as a document, such as a broker's PDF or a signed scan, and has to be checked against the pay stubs, W-2s and bank statements in the file.

This guide covers what's on the 1003, when to parse it instead of pulling it from your LOS, the cross-checks to run and how to automate them.

What is a 1003 in a US mortgage?#

The 1003 is the standard loan application for a home mortgage in the US, published jointly by Fannie Mae and Freddie Mac. The redesigned version, marked "Effective 1/2021", became mandatory for new loans sold to the two agencies on March 1, 2021, and it's still the current form. The borrower answers sections 1 to 8, the loan originator completes section 9, and the lender fills in a separate Lender Loan Information form.

The URLA is a set of forms, not one:

• Borrower Information

  Sections 1 to 9 for the borrower: the core of the application.

• Additional Borrower

  The same nine sections for each additional borrower.

• Unmarried Addendum

  Used when an unmarried borrower has a partner with property rights similar to a spouse's, such as a civil union or domestic partnership.

• Lender Loan Information

  Sections L1 to L4, completed by the lender: property and loan details, title, the mortgage terms and payment, and the funds needed to close.

• Continuation Sheet

  Extra space when a section runs out of rows.

The 1003 is for a new mortgage. A borrower who already has one and can't keep up with the payments applies for help on a different form, Form 710.

The 9 sections of the 1003, and what to extract#

The numbers on the form follow the rows of the table below, top to bottom.

Sections 1 to 4 hold the stated income, assets and debts that the pay stubs, W-2s and bank statements in the file are checked against.

When to parse a 1003, and when to pull it from your LOS#

Fannie Mae's Selling Guide calls the version in a lender's LOS or in Desktop Underwriter the online loan application: data collected from the Form 1003. The agencies' underwriting systems take it as a MISMO v3.4 file, so for applications your own staff take in the LOS, there's nothing to extract. Parsing earns its place when the 1003 reaches you as a document.

Cross-checks to run on every 1003#

The 1003 is what the borrower says; the rest of the file is the evidence. Fannie Mae requires the final 1003 to reflect the income, assets, debts and loan terms used to underwrite the loan, so every stated figure should tie to a document.

• Income against pay stubs and W-2sBase, overtime and bonus income match the latest pay stub (dated within 30 days of the application, with year-to-date earnings) and the W-2s for the last one or two years.
• Two years of employmentEmployers and start dates in sections 1b to 1d cover at least two years and agree with the W-2s.
• Assets against statementsEach account in section 2a matches its statement: the borrower as holder, the last 4 digits of the account number and the balance, over the last two months for a purchase or one month for a refinance.
• Debts against the credit reportEvery liability in section 2c appears with the same balance and payment, and nothing on the credit report is missing from the application.
• Real estate owned against mortgage statementsEach property in section 3 has its payment and balance confirmed, with taxes, insurance and dues documented.
• Declarations against the file"No" answers on bankruptcy, foreclosure, judgments and undisclosed debt agree with the credit report.
• Loan and property against the contract and appraisalLoan amount, purpose, occupancy and property value match the purchase contract and the appraisal.
• Signed, dated and consistentThe final 1003 is signed and dated by each borrower, and a 1008 for a manually underwritten loan carries the same figures.

For the asset checks in detail, see mortgage bank statement verification.

How automated 1003 data extraction works#

The hard parts of a 1003 are the repeated rows (several employers, accounts, debts and properties, sometimes spilling onto continuation sheets), checkboxes that carry the meaning, and handwriting on paper applications. A parser has to keep each row's fields together and flag what it isn't sure about.

Docsumo is an intelligent document processing (IDP) platform. It takes loan files by email, upload or API, splits and classifies the documents (on the Business plan), and extracts the 1003 alongside pay stubs, tax forms and bank statements, handwritten entries included. Fields it's unsure about go to a reviewer with the source line highlighted, and corrections improve the model. Cross-checks like the ones above run as cross-document validation on the Enterprise plan. It doesn't make the underwriting decision or calculate qualifying income on Fannie Mae Form 1084 or Freddie Mac Form 91; it extracts the income data that goes into them. Results go to your LOS through API and webhooks, or download to Excel. More on the whole file in mortgage document processing.

• 99%99%field-level accuracy across 250+ document types
• 95%+95%+of documents processed straight through, without manual review
• <5 min<5 minper document, down from 2+ hours

The bottom line#

The 1003 is the borrower's account of their finances, and the rest of the loan file is the proof. If you already hold the application as data, check the documents against it; if it arrives as a PDF or scan, extract every section first. Either way, the value is in the cross-checks and in sending people only the fields that don't agree.

Book a demo with a few of your own loan files, or start a free trial.

Frequently asked questions#