Case Study 1 — The Maps: How a Risk Assessment Became a Policy

A real, public case. The HOLC residential security maps and their accompanying area descriptions have been digitized and published by academic projects and are available to read directly. This account describes the documented record; readers are strongly encouraged to look at the maps for their own city.

Background

In 1933 the Home Owners' Loan Corporation began refinancing distressed mortgages at national scale. To do that it needed something that did not exist: a systematic way to assess property risk in cities it had no local knowledge of.

Between roughly 1935 and 1940, HOLC field agents worked with local real estate professionals, lenders, and appraisers to produce residential security maps for more than two hundred American cities. Each neighborhood received a grade — A, B, C, or D — rendered in green, blue, yellow, and red, accompanied by a written area description recording the factors behind the grade.

The stated purpose was mundane. A national lender needed a consistent framework, and the framework had to be portable across cities.

The operating issue

The area description forms asked for things a modern appraiser would recognize: the age and condition of the housing stock, the terrain, the occupations and income levels of residents, the presence of nuisances, the rate of construction, and the trend of values.

They also asked, directly, about the racial, ethnic, and national-origin composition of the residents, and about whether the neighborhood was undergoing what the forms termed "infiltration."

The completed descriptions are explicit. Neighborhoods with Black residents were graded D — the lowest grade — with reasoning that named the residents' race as the basis, frequently without reference to housing condition and sometimes explicitly despite good housing condition. Areas with substantial immigrant populations were commonly downgraded to C. The presence of racially restrictive covenants was recorded as a favorable factor supporting a higher grade.

This is not an inference from outcomes. It is what the documents say.

What happened

Two features of the system made it consequential far beyond the HOLC's own lending.

First, the framework migrated. The FHA — the agency whose insurance made high-loan-to-value long-term lending possible at all — adopted underwriting standards embodying the same logic. Its underwriting manual of the period advised against insuring in neighborhoods undergoing racial transition, treated proximity of what it called "inharmonious racial groups" as a valuation risk, and endorsed restrictive covenants as a stabilizing device. Private lenders and appraisers, who took their standards from the FHA because FHA insurance was what made the loans marketable, absorbed the same criteria.

Second, the grading was self-fulfilling. A neighborhood graded D received less mortgage credit. Less mortgage credit means fewer buyers, deferred maintenance, and falling values. Falling values confirm the grade. The map did not merely predict decline; it was one of the causes of it, and the resulting data then validated the map.

The programs themselves worked extraordinarily well for the households that could use them. The postwar expansion of American homeownership — and the middle-class wealth built on it — was substantially a product of FHA and VA lending. The point of this case is not that the programs failed. It is that they succeeded, and that access to them was allocated on a basis that is now illegal.

What it shows

1. A risk model can encode a policy. The maps were presented as an empirical assessment. They incorporated a variable — race — that had no causal relationship to loan performance and enormous correlation with the grading, and then produced outcomes that appeared to validate the model. This is the exact failure mode that modern fair lending analysis of automated underwriting and pricing models is designed to detect. Chapter 36's discussion of AI in underwriting is this case in current form, and the reason explainability is treated as a compliance requirement rather than a technical nicety.

2. Facially neutral criteria can carry the same freight. Once naming race became unacceptable, the same exclusions could be produced by criteria correlated with it — minimum loan amounts that exclude low-value housing stock, branch and marketing footprints that stop at particular boundaries, appraisal methodologies that treat a neighborhood boundary as a market boundary. This is precisely why disparate impact exists as a legal theory (Chapter 25). Requiring proof of intent would make the doctrine unenforceable against the mechanism that actually operates.

3. The wealth effect compounds and does not reset. A thirty-year amortizing loan converts income into an asset that appreciates and transfers at death. Three decades of differential access to that instrument produces a gap in the next generation's down payment, and the generation after that. This is why the modern discussion concerns gift funds, intergenerational transfers, and down-payment assistance rather than only current-year lending decisions — and why Chapter 33 exists.

4. "Everyone used this standard" was true and is not a defense. The maps were the professional standard of their era, produced with the participation of local real estate professionals and lenders. Universality of a practice is evidence about the era, not about the practice.

The outcome for the practitioner

You will encounter this history in three concrete places in your work.

Appraisal. The comparable-selection question — which sales are "in the market" for a subject property — is exactly where neighborhood-boundary reasoning lives. Appraisal bias is an active federal enforcement priority, and Chapter 18 covers both the mechanics and the reconsideration process.

Marketing and footprint. Where you prospect, which agents you build relationships with, and which neighborhoods your employer's marketing reaches are the raw material of a modern redlining analysis. Chapter 38 covers business development; Chapter 25 covers what a regulator looks at.

Assistance and equal treatment. The most common individual-level fair lending exposure is not a denial. It is unequal effort — spending forty minutes helping one applicant restructure a file and five minutes telling another they do not qualify. Chapter 25 is specific about this, and it is worth reading before you take your first application rather than after.

Discussion questions

1. The maps were a genuine attempt to solve a real problem: a national lender needed portable risk criteria. Describe a current practice in mortgage lending that is a genuine attempt to solve a real problem and that could produce a similar effect. How would you detect it?

2. The grading was self-fulfilling — the model's output changed the world in a way that confirmed the model. Name one place in modern origination where an assessment can change the outcome it is assessing.

3. This case argues that the success of the FHA and VA programs is what makes the exclusion consequential. Restate that argument for someone who says "that was ninety years ago."

4. Restrictive covenants were recorded as a favorable factor. Explain to a colleague why that detail is more revealing than any single grade on any single map.

5. Look up the residential security map for the city you work in, if one exists. Compare a D-graded area to a modern map of your employer's branch locations or your own referral network. Write two paragraphs on what you find — including, honestly, if you find nothing.