Chapter 15 — Further Reading

Grouped by the book's three tiers. Tier 1 is verified canonical material we can stand behind. Tier 2 is real, attributed material whose exact current values or specifics you must verify yourself. Tier 3 is this book's own constructed teaching material.

The single most important habit this chapter can give you is going to the primary source. Everything about automated underwriting that matters operationally — eligibility parameters, ratio maximums, documentation requirements, waiver criteria — is published, free, and revised continuously.

If you read only one thing

Pull your last three findings reports and read all of every one of them.

Not a book. Not this chapter. Three real reports from your own pipeline, read from page two to the end, with your income worksheet next to them. Reconcile the five figures. Read every verification message and ask why each one was generated. Find the message that was produced by something you typed and did not notice.

Nothing on this page will make you better at this job as fast as that will, and it takes about forty minutes. If you have no live files yet, ask a colleague for a redacted one; every loan officer in your office has a folder full of them and most of them have never read one either.

TIER 1 — Verified canonical

The Fannie Mae Selling Guide. Free, public, continuously updated, and the authority behind every Desktop Underwriter eligibility parameter. Read the eligibility, credit, income, liabilities, and asset sections that correspond to the pages of a findings report. Fannie Mae also publishes DU release notes describing what each engine version changed — subscribe to them; a version release is one of the two things that can move a recommendation without anyone touching the data.

The Freddie Mac Single-Family Seller/Servicer Guide. The corresponding authority behind Loan Product Advisor. Reading the two guides side by side on the same topic — student loan payments, self-employment history, reserves — is the fastest way to understand why the two systems sometimes disagree.

HUD Handbook 4000.1, the Single Family Housing Policy Handbook. The authority for FHA, including the TOTAL Scorecard, the Accept/Refer distinction, the manual underwriting benchmarks, and the compensating factors that permit higher ratios on manually underwritten files. Chapter 16 lives here; §15.3 is a doorway into it.

The FHA TOTAL Mortgage Scorecard documentation published by HUD. Explains the scorecard's role, that it operates through an approved automated underwriting system rather than standalone, and what a mortgagee's obligations are on an Accept and on a Refer.

The Uniform Residential Loan Application (Form 1003 / URLA). Every field on the form becomes a field in the AUS. Reading the form as a data schema — rather than as a form — is a genuinely useful exercise and directly supports §15.9's data audit.

The Equal Credit Opportunity Act and Regulation B (12 CFR Part 1002). The adverse action requirements, the prohibition on discouraging prospective applicants, and the valuations rule (notice of the right to receive copies of appraisals and other written valuations, and the timing of delivery). Chapters 18, 24, and 25 go deeper; read the adverse action and discouragement provisions now, because they are what §15.10 and Case Study 2 rest on.

CFPB Circular 2022-03, on adverse action notification requirements in connection with credit decisions based on complex algorithms. Short, direct, and the clearest statement available that model opacity is not a defense. Check the Bureau's circulars page for subsequent guidance in the same line.

The Home Mortgage Disclosure Act and Regulation C. Relevant here for what the public HMDA data does and does not contain — notably that it does not contain credit score, which is the crux of the measurement debate in Case Study 2.

The Federal Housing Finance Agency (FHFA) for conforming loan limits, which change annually and are county-specific, and for published research and analyses on GSE lending and appraisal data.

TIER 2 — Attributed; verify the specifics yourself

Your own lender's overlay matrix and AUS policy. The single most useful non-public document you have access to, and almost nobody reads it. It tells you which system your shop runs first, whether you may run both, whether the lender exercises value acceptance offers, and every place where the lender's answer differs from the agency's. Ask your underwriting manager for the current version and ask again in six months.

Agency release notes and product bulletins, for both enterprises. These announce engine version changes, eligibility changes, and new capabilities such as validation services and collateral options. They are the reason a professional is never surprised by a version release.

Gates, Susan; Perry, Vanessa; and Zorn, Peter. Automated Underwriting in Mortgage Lending: Good News for the Underserved? Housing Policy Debate, 2002. The clearest articulation of the access argument for automated underwriting. Note that the authors were affiliated with Freddie Mac; read it as a well-argued position from an interested party, which is a useful skill in itself.

Munnell, Alicia; Browne, Lynn; McEneaney, James; and Tootell, Geoffrey. Mortgage Lending in Boston: Interpreting HMDA Data. Federal Reserve Bank of Boston working paper, 1992; revised version published in the American Economic Review, 1996. The study that made the discretion question unavoidable. Read the critiques alongside it — the methodological debate is genuinely instructive about what regression on lending data can and cannot show.

Martinez, Emmanuel, and Kirchner, Lauren. The Secret Bias Hidden in Mortgage-Approval Algorithms. The Markup, 2021, distributed with the Associated Press. Read the published methodology and the responses to it, not a summary. Case Study 2 explains why the credit-score limitation matters and why it is not a dismissal.

Freddie Mac Economic and Housing Research, note on racial and ethnic valuation gaps in home purchase appraisals (2021), and FHFA's published analyses of appraisal data. Relevant to §15.7's open question about whether automating valuation helps or hides.

The PAVE Task Force Action Plan (Interagency Task Force on Property Appraisal and Valuation Equity, 2022). Chapter 18 uses this properly; it is here because value acceptance and appraisal equity are the same conversation.

Announcements from both enterprises regarding rental payment history in DU and in LPA, and the nontraditional credit provisions of both guides. Scope and mechanics have been revised since announcement — verify current eligibility before you rely on any of it with a borrower.

Supervisory guidance on model risk management issued by the federal banking agencies. Applies to supervised institutions rather than to you personally, but it is the framework your employer's risk department is working within, and knowing it exists makes you a better colleague.

The Urban Institute's Housing Finance Policy Center, for regularly updated, generally careful analysis of credit availability, access, and GSE policy. Useful for context; check the date on everything.

TIER 3 — Illustrative and constructed

The Linden Street file. All figures in this chapter — the \$10,500.00 income, the 706 representative score, the \$4,479.72 of total obligations, the 42.66% back-end ratio, the \$12,623.66 of reserves, the day-44 jump to 48.48%, and the day-47 re-run — are this book's frozen constructed teaching figures. The rendered findings summary page and verification-message block in §15.4 and §15.6 are written for this book. No agency's actual report wording, message numbering, or page layout is reproduced.

The Harlow Street file. The 641 representative score, \$4,150.00 income, \$395.00 of debts, \$1,721.57 housing payment, and the 41.48% / 51.00% ratios are constructed. The 31% / 43% manual underwriting benchmark is real and belongs to HUD; the permitted exceptions and their compensating factors are in Handbook 4000.1 and must be read there.

The Fulton Avenue file. The \$9,020.83 and \$8,916.67 income figures and the 2.3% decline are constructed teaching figures; Chapter 32 owns the worksheet. The \$3,870.00 obligation figure used in §15.9's half-a-point illustration is a hypothetical introduced only for that arithmetic.

Every guideline value quoted in this chapter — the 97% LTV maximum, the ten-payment installment exclusion threshold, the 90% second-home LTV in the exercises, the \$766,550 loan limit in exercise 15.19 — is illustrative and dated the moment it was written. The structure is what transfers. The values are perishable, and a loan officer who quotes a textbook figure to a borrower instead of the current published one has learned the wrong lesson from a chapter about data integrity.