The Comp Selection Problem: Why Location Proximity Isn't Enough

Selecting comparable sales by distance alone ignores property type mix, condition variance, and micro-market clustering. Here is why that matters for lender-grade valuations.

The simplest automated valuation methodology is also one of the most misleading: find the N closest recent sales, average their prices per square foot, multiply by the subject's square footage, and return the result. It sounds defensible until you examine what it systematically gets wrong.

Why Distance Is a Proxy, Not a Signal

Distance from the subject property correlates with comparability only loosely, and in some geographies not at all. A property on the east side of a major arterial road may be a mile from a property on the west side -- technically closer than a comp four streets away in the same subdivision -- but the two markets have different price dynamics, different school attendance zones, and different buyer profiles. A distance-based comp selection algorithm will pick the closer sale and produce a biased estimate.

In Colorado, where Plotgleam operates, this problem is severe in urban infill and mountain communities alike. In Denver's residential neighborhoods, a four-block radius can span three distinct micro-markets with per-square-foot price differences of 15-25%. In mountain resort areas, proximity means nothing when a ski-in property and a frontage-road property are half a mile apart but transact in completely different buyer pools.

Property Type and Bedroom Mix

A three-bedroom single-family home should not be compared to a two-bedroom townhome, even if they are adjacent and have similar square footage. The buyer pools are different. The financing programs available differ. The value trajectories in different market conditions diverge. Yet many mass-market AVM systems have insufficient depth in thin markets to enforce type-and-bedroom matching, so they use what is available.

The consequences show up in appraisal review. A desk reviewer who runs a comp check and finds that the automated estimate relied on a townhome comparison for a detached single-family subject will reject the estimate, regardless of how close the comparables are geographically. The appraiser's documentation requirement under FNMA guidelines is explicit about type-matching. An AVM that cannot meet that standard produces output that cannot go in the loan file.

Condition Variance Within a Radius

Location proximity controls for neighborhood-level price signals. It does not control for condition. A recently renovated property in a stable neighborhood transacts at a meaningful premium over an unrenovated comparable a block away. If the subject property is mid-renovation and the AVM pulls a mix of renovated and unrenovated comps without adjusting, the estimate can be off by 8-15% in either direction.

Condition adjustment is the hardest part of residential appraisal to automate because it historically required physical inspection. A certified appraiser walks through the property and assigns a condition rating based on observed quality. Without a site visit, an automated system has to derive condition signals from available data -- permit history, listing descriptions, DOM patterns, and prior sale history. That is approximation, not observation, and it should be disclosed as such.

Micro-Market Clustering

Property markets cluster in ways that a simple radius does not capture. A cul-de-sac at the end of a dead-end street in a desirable school zone may have consistently higher transaction prices than the surrounding blocks because buyers specifically target that street for the combination of factors it offers. A ring drawn around the subject property at 0.5-mile radius will average in nearby sales that are not in the same micro-cluster, diluting the signal.

Detecting micro-market clustering requires enough transaction volume to identify the pattern statistically, and enough geographic granularity to draw boundaries that correspond to real buyer behavior rather than arbitrary geometric shapes. In dense urban markets, this is feasible. In suburban and exurban markets with thinner transaction volume, it requires a methodology that does not collapse when the sample is small.

What Lender-Grade Comp Selection Requires

The Federal Housing Administration and the GSEs publish detailed guidelines on comparable sale selection. The core requirements include type matching, proximity within a defined radius, recency within a defined look-back window, adjustment for size differentials, and documentation of the selection rationale. These are not suggestions; they govern whether a report format satisfies desk review standards.

An AVM that documents its comp selection in a format reviewable by a desk appraiser can go in the loan file. One that runs a black-box algorithm and returns a point estimate with no selection rationale cannot. The documentation requirement is as important as the accuracy requirement, and for many lenders it is the first criterion an automated valuation tool fails.

The Plotgleam Approach

Plotgleam's comp selection logic starts with type and bedroom matching before applying distance constraints. Within matched-type results, it applies condition scoring derived from permit history and listing-condition flags to weight comps toward comparables with similar condition profiles. The output includes the selected comps in a structured table with the selection rationale, so a desk reviewer can evaluate the methodology rather than trusting a black box. That is the documentation standard that makes a lender-grade valuation usable rather than merely fast.

When Location Proximity Should Be Expanded

The guidelines for comp selection allow the appraiser to expand their search radius when insufficient comparable sales exist within the standard distance. This is appropriate and necessary in thin markets, but it introduces a judgment call: when the expanded-radius comp is used, how much weight should it carry relative to a closely located but older or less well-matched comp?

In automated systems, this tradeoff needs to be parameterized. A fixed distance cutoff with a hard drop-off is less useful than a distance-decay weighting function that reduces the influence of farther comps proportionally. The decay function needs to be calibrated against the specific market -- how large a radius still captures comps in the same sub-market? -- and should be adjustable by the desk reviewer who knows the market well enough to judge when the algorithm's radius choice is appropriate.

Providing that reviewer control -- the ability to expand or constrain the comp radius, filter by condition tier, or override a specific comp -- is what distinguishes a tool designed for professional review from a consumer-grade estimator. It treats the desk reviewer as a participant in the valuation process rather than a passive consumer of the output, which is the role that both regulatory requirements and good workflow design assign to them.