How home valuation tools work
Where an adjustment number comes from
4 min read
Every adjustment grid implies a method. A figure sits beside a comparable sale, the reason is printed next to it, and the document invites the reader to assume that figure came from somewhere. Sometimes it did. Often it came from a rate applied uniformly across the set, and nothing on the page separates the two cases.
The method being implied is paired sales analysis: locate two transactions that match on location, size, age and condition and differ in one respect, then attribute the price difference to that difference. It is the technique that turns a judgment into evidence, and it is the one most grids cannot perform.
What the Selling Guide says, precisely
Fannie Mae's Selling Guide, section B4-1.3-09, states that adjustment rates can be determined through statistical analysis, modeling, paired sales, or other commonly accepted methods. That sentence sits in the market conditions and time adjustment discussion, not in a general instruction about physical characteristics, and paired sales appears there as one option among several rather than as a preferred technique.
The binding requirement in the same section is broader and more useful: the appraiser's adjustments must reflect the market's reaction to the difference in the properties. A valid pairing demonstrates that reaction. So does a defensible regression, a depreciated cost analysis, or a documented account of what buyers in that submarket have paid. What does not demonstrate it is a percentage applied because it is the percentage that has always been applied.
USPAP is quieter still. It does not name the method anywhere, and the obligation that governs is Standards Rule 1-1(a), which requires the appraiser to employ recognized methods and techniques necessary to a credible result. The standard sets the bar and leaves the instrument to the assignment.
The supply problem
The Appraisal Institute's own text describes paired data analysis as theoretically sound but liable to produce unreliable results where only a narrow sampling of sufficiently similar properties is available, and states that an adjustment derived from a single pair is not necessarily indicative, in the same way a single sale does not necessarily reflect market value.
That is a direct statement of the constraint. The method is sound and the data usually is not there. In markets of heterogeneous older stock, where every property carries its own alteration history, two sales differing in exactly one respect may not exist at any point in the year. Transaction volume does not fix it, because volume in a varied market produces more unlike sales rather than more pairs.
The four methods available to support an adjustment differ in what they need before they can be used at all, which is usually what decides the choice rather than any judgment about quality.
| Method | What it requires | What it supports | Where it fails |
|---|---|---|---|
| Paired sales analysis | Several recent sales matching on location, size, age and condition, differing in one attribute, free of concessions | A figure traceable to named transactions in that submarket | Heterogeneous stock, thin volume, or any market where the matching pair does not exist |
| Published rate applied per variable | Nothing beyond the comparable data itself | Internal consistency across every line of the grid | Says nothing about local market reaction, and carries the same figure into submarkets that price the attribute differently |
| Depreciated cost | Current replacement cost and a defensible age and condition estimate | Improvements with a clear build cost, such as a pool, an addition or a roof | Location, view and functional obsolescence, none of which have a build cost |
| Statistical analysis or regression | A sample large enough for the variable count, which residential assignments rarely reach | Several attributes at once, with the interaction between them | Small submarkets, and any set where relevance has been traded away to reach sample size |
The assumption inside the method
A derived adjustment carries an assumption that rarely gets stated: that the attribute has a fixed dollar value which can be lifted from one pair and applied across a set.
Published research in the Journal of Real Estate Research found the normative paired sales method fails to account for the diminishing marginal price effects of property attributes, testing the finding on view and lot size. The practical consequence is that an adjustment derived at one point in a price distribution overstates the same feature higher up and understates it lower down. A grid that applies one figure uniformly inherits that error even when the figure was correctly derived.
What the tools do, including this one
Most valuation software applies published rates. CMAflow does the same, and the rates are stated rather than hidden: size at half the median price per square foot, bathrooms at 2% of the comparable price per full bath, bedrooms at 1.5%, age at 0.3% per decade capped at 3%, and condition where it is known. Each adjustment prints with its reason, and the recommended price is the weighted average of the adjusted values.
Those rates are defensible and they are not paired sales. They are a consistent convention applied to every comparable in the set, which is a reasonable thing for software to do and a different thing from deriving a figure from local evidence. The honest statement of the limit is this: the grid can show what was adjusted and why, and it cannot tell the reader whether a matching pair existed in that submarket to support the amount.
Naming the limit is more useful than solving it badly. A tool that claimed to derive every adjustment from matched pairs would be claiming a data condition that does not hold in most markets, and the claim would fail first in exactly the neighbourhoods where the number matters most.
Where the confidence band does the work
The part of the document that can carry this fairly is the range rather than the grid. Where CMAflow computes confidence from the dispersion of the adjusted values, a set whose members disagree earns a wider band and says so on the page. That is not the same as measuring whether the adjustments were derived, but it is correlated with it: sets assembled from properties similar enough to pair tend to cluster, and sets assembled from whatever was available tend not to.
The agent knows which of those two situations they are in. This is Context Blindness™ in its narrowest and most technical form: the tool can compute the spread, and only the person who walked the street knows whether the comparables were alike enough for the spread to mean anything.
What would close the gap
A field, not a rule. If a grid recorded, per line, whether the amount was derived from identified sales or applied from a rate, the reader could weigh the two differently and the document would stop implying a method it did not use. That is a small addition and nobody has made it, including this platform.
Until something does, the useful discipline is the question rather than the software. Before an adjustment goes out under your name, ask which sales it came from. An answer that names properties is evidence. An answer that names a percentage is a convention, and a convention stated plainly is worth more than a derivation implied quietly.
The geography of the problem is covered from the agent's side in this breakdown of two Florida counties where the evidence conditions differ sharply.
Does Fannie Mae require paired sales analysis for adjustments?
No. Section B4-1.3-09 of the Selling Guide names paired sales as one acceptable way to determine adjustment rates, alongside statistical analysis, modeling and other commonly accepted methods, and it does so in the market conditions context rather than as a general rule for physical characteristics. The binding requirement in that section is that adjustments must reflect the market's reaction to the difference in the properties. Paired sales is one way to demonstrate that. It is not the only one and it is not mandated.
What does USPAP say about paired sales analysis?
Nothing. USPAP does not name the method anywhere, because it is method agnostic by design. The relevant obligation is Standards Rule 1-1(a), which requires an appraiser to be aware of, understand and correctly employ the recognized methods and techniques necessary to develop a credible appraisal. Which technique satisfies that in a given assignment is left to the appraiser and to the evidence available.
Why do adjustment grids show precise figures for features that cannot be measured locally?
Because a grid has to print something in every cell. Most valuation tools apply a published rate to each variable, which produces an internally consistent set of figures whether or not local evidence supports any of them. The output is identical in appearance to a grid where every line was derived from matched sales. Nothing in the document distinguishes the two, which is the gap worth naming rather than the precision itself.
The Independent Agent
Substack | Spotify | CMAflow FAQ | YouTube | Insights | Blog