Geographic Concentration Risk: Why a Diversified Map Can Lie

Concentration limits are among the oldest and most sensible controls in residential credit. The trouble is not the control, it is the unit. Shares by state and metro are computed against boundaries drawn for administrative and statistical purposes, and housing markets were not consulted. A book can sit comfortably inside every geographic limit it has and still be holding the same bet many times over.

What geographic concentration risk means

Geographic concentration risk is the exposure created when many assets in a book can be affected by the same local event or trend. It is a correlation problem wearing a map. The reason it gets managed geographically is straightforward: location is the most legible proxy for shared exposure that a loan tape supports, and for a large class of shocks (a regional employer contraction, a state policy change, a climate event) it is a genuinely good proxy.

The risk is real and the control is correct. What follows is about the resolution at which it is applied.

Three houses or three thousand

This is usually written up as an institutional problem, which makes it easy to assume it starts at some threshold of scale. It does not. The arithmetic is identical whether you hold three rentals or thirty thousand loans, and only the tools change.

An individual investor who buys three houses in the same metro because the first one worked has made exactly the bet a fund makes when it concentrates in one MSA. It often feels like the opposite of a bet: you know the area, you have a contractor there, the numbers penciled before. Familiarity is the most common route into correlated exposure, and it does not announce itself. Three properties on the same side of one city, bought on the same thesis, in the same price tier, will move together, and a metro-level price chart will show you nothing useful about that, because the variation is happening underneath it.

What changes with scale is not the risk but the remedy. With three houses you can hold the whole picture in your head, and the fix is to look at each property against its own local market before you buy the fourth. With three thousand rows nobody can hold the picture, so the same read has to arrive as an ordering: which rows cluster, which deserve attention first. Everything below applies at both ends; the institutional sections are the version that has to be automated.

How it is conventionally measured

Why administrative boundaries mislead

A metropolitan statistical area is defined around a core urban center and the surrounding counties tied to it by commuting patterns. It is an excellent unit for labor-market analysis and a poor one for housing. A single metro can contain a dense pre-war core, mid-century inner suburbs, recent exurban subdivisions, and semi-rural fringe, all segments that draw entirely different buyers, at entirely different price points, with entirely different supply dynamics.

ZIP codes are worse in a specific way: they are postal routing constructs, so they are shaped by delivery logistics instead of by anything about the housing stock. A single ZIP can hold neighborhoods that no buyer would consider substitutes for one another. We wrote about the property-level consequence of exactly this in why two homes in the same ZIP code appreciate differently.

Four ways concentration hides in plain sight

1. The repeated thesis

A buyer with a view (workforce housing near logistics corridors, entry-level homes in fast-growing secondary metros, build-to-rent in sunbelt exurbs) will express it across many states. The map looks diversified. The exposure is one idea bought twenty times, and the outcomes will move together if the idea is wrong. Geographic dispersion protects against local events; it does not protect against a correlated thesis.

2. The price tier

Entry-level and luxury segments respond differently to rate moves, credit availability and inventory. A book concentrated in one tier across many markets carries a coherent exposure that no geographic cut will surface, because tier is not a geographic field.

3. The vintage cluster

Loans originated in a narrow window share an underwriting environment, a rate regime, and a set of valuations struck under similar conditions. When the weakest collateral in a book turns out to cluster by origination vintage, the geographic report will have shown nothing unusual throughout.

4. Count versus balance

A market that is a modest share of a pool by loan count can be a large share by balance, especially where price levels differ sharply across the book. A concentration report on a single basis understates exposure roughly half the time. Both bases should be shown, and a large gap between them is a finding in itself.

What to check instead

None of this argues for abandoning geographic limits. It argues for treating them as a floor and adding questions that are not geographic:

• Would the diversification claim survive a finer cut? If exposure inside your largest metros clusters into a handful of submarkets, the metro-level share was never the real number.
• What is the shared thesis? Write down, in a sentence, why each large segment was acquired. If several sentences are the same sentence, the geography is decorative.
• Does composition read the same by count and by balance?
• Where does the weak collateral cluster? The tail, not the average. Weak rows concentrated in one submarket, tier or vintage is the finding that matters.
• Where is the evidence thinnest? If the properties you can say least about are themselves concentrated somewhere, that is a concentration of uncertainty, and it should be reported alongside the concentration of exposure.
• How old are the marks, by segment? Concentrated staleness behaves like concentrated exposure and is far less likely to be reported.

Where a property-level read helps

A within-market rank on every property changes what concentration analysis can be computed on, in three specific ways.

It makes rows comparable across geographies. Because each property is ranked against comparable properties in its own local market instead of against a national benchmark, a row in Ohio and a row in Arizona can be placed on the same axis. Concentration can then be measured on something other than location: whether weak-ranked rows cluster by tier, vintage or originator.

It reveals within-metro dispersion. Two metros with identical pool shares can look very different once their rows are ranked locally: one spread across the distribution, the other bunched at one end. That difference is invisible in a share table and visible immediately in a rank distribution.

It surfaces concentration of uncertainty. Coverage, the share of rows that cannot be assessed with sufficient local evidence, is itself distributed. When unsupported rows cluster in one market or one property type, that is a specific, actionable statement about where your evidence is weakest. Coverage is not a defect rate; it is the analysis reporting the limit of its own evidence.

The broader framework this sits inside is in the institutional guide to residential property risk assessment, and the pre-bid application in candidate pool assessment.

Where the read stops

A within-market rank is one lens, and it is worth being exact about which questions it answers and which it leaves to the tools you already run.

• It measures position, and co-movement stays with your risk models. A rank does not estimate how two assets move together, so it complements scenario and stress analysis instead of standing in for them.
• It is a position, and forecasting stays with forecasting. The output is where a property sits within its market today, not a predicted price path for a region.
• It works underneath your geographic limits, which stay exactly as they are. State and metro exposure remain the right control for state and metro shocks.
• It informs collateral, and credit stays with credit. Concentration reaches loss distributions through many channels, and property positioning is one input among them.

The narrow version of the argument

Geographic concentration limits do exactly what they were designed to do, at the resolution the data supports. The gap is that the resolution the data supports is coarser than the resolution at which housing markets operate, so a book can satisfy every limit and still hold correlated positions. Adding a within-market read does not replace the limits. It lets you ask the question the limits were always a proxy for: which of these rows would move together, and do we know why?