Market Conditions Adjustment

Using this tool, and how it works

Two halves. The first is how to produce an adjustment. The second is what the calculation actually does, written for the reviewer who asks. If you are in a hurry, read “What to pull from the MLS” and skip to “Reading the result”.

1. What to pull from the MLS

This is the step that decides everything downstream, and it is the one most often done wrong. You are pulling market data, not your comparable selection. The model needs enough sales to see the shape of the market; you pick up to 4 comps you are actually adjusting later, on the review screen.

• Cast wider than your comp search. Include sales that are not comparable to the subject but compete for the same buyer — a broader radius, or neighbouring areas at a similar price point and quality. Searching by price range is reasonable here, even though it would not be when selecting comparables.
• As many months as your search will give you, up to 36. By default the tool fits the trend over everything in your file — you chose what to pull from your MLS, and the window only ever narrows that. 12 months is the floor and it will not go below it, because Fannie Mae requires the trend to rest on at least twelve months of data. A longer file also lets the model see last year's seasonal pattern rather than guessing at it. Past 36 months the curve becomes too stiff to follow a market that has turned, so that is the ceiling; you can narrow the window on the review screen whenever a shorter one describes your market better.
• Include the contract date column. It is the time axis — a sale is placed on the date it went under contract, not the date it closed. See “The time axis” below for why this matters more than it sounds.
• Include square footage, and whatever else you have — beds, baths, year built, lot size. These are not decoration. The market trend and the value of those differences are estimated together, which is what stops a month when larger houses happened to sell from being read as the market rising. A column you leave out cannot be corrected for.

How many sales you need

More sales let the curve take a more complicated shape. Below 25 sales in the window the tool does not report a locally derived rate at all — it reports the published index instead, and says so on the exhibit.

A straight line is not a lesser answer when the market moved steadily. It is the wrong answer when the market ran and then flattened, which is the case this product exists to handle — so pull more sales than you think you need.

2. The five steps

1. Upload. CSV, TSV or tab-delimited text. If your export opens in Excel, use File → Save As → CSV first. Column order does not matter and headers do not have to match anything; the tool recognises common MLS layouts and asks about the rest.
2. Confirm the mapping. Anything the tool is not certain about is flagged for you to confirm rather than guessed silently. Contract date and living area are always worth a look.
3. Describe the subject. The effective date of your appraisal, which is what every adjustment is measured to.
4. Review. The curve, the sales behind it, the warnings, and a per-comparable adjustment. Click any sale to exclude it and watch everything update.
5. Export. A PDF exhibit for the workfile, kept in your saved reports.

3. Reading the result

The interval is the point. Every adjustment carries a 95% interval. A wide interval is not a defect in the tool — it is the tool telling you the truth about what your data can support. A tight interval that is right half the time is worse than no interval at all, and it is what gets a report sent back.

When the interval half-width exceeds 8 percentage points, or you have fewer than 25 sales, the exhibit reports the index-anchored figure instead of the local one and labels it as such. It never refuses to produce a number. An appraiser at 5pm with a report due needs an answer with an honest caveat, not a refusal.

“From comps” tells you how much was yours. The exhibit prints the weight the local sales carried against the published index. If that number is low, the market you uploaded was too noisy to say much on its own, and the figure leans on the index. That is disclosed rather than hidden.

Warnings are not errors. They are the things a reviewer would otherwise catch: thin months, gaps in the data, a shift in the mix of what sold, concessions reported inconsistently, a local trend that disagrees with the published index, a missing contract date. Each one appears on the exhibit in plain language.

4. How the calculation works

The time axis is the contract date

A sale is placed on the date the parties agreed the price, not the date the money moved. Those differ by a month or two, and using the closing date shifts every sale later than the market event it represents — which flattens a turn and mis-times a peak. Fannie Mae frames it the same way: whether an adjustment is warranted turns on market changes “between the contract date of the comparable sale and the effective date of the appraisal report”.

If your MLS does not export a contract date, the tool estimates one at 40 days before closing and raises a warning saying it did. The estimate is disclosed on the exhibit; it is never silent.

Time and property differences, estimated together

The market trend and the value of the differences between houses are fitted in one regression, not in sequence. This is the single most important choice in the tool. If you fit price against date alone, then any change in what happened to sell shows up as market movement: a quarter when larger or newer homes closed looks like appreciation. Estimating both at once separates them.

Smoothness comes from the data volume, not from a mode switch

The curve is a natural cubic spline whose flexibility is set by the number of sales, per the table above. The model class never changes — there is no threshold where the tool quietly switches to a different method and produces a discontinuous jump in the answer.

Seasonality comes from outside your comp set

Seasonal effects are applied from published factors for your metropolitan area, never estimated from the sales you uploaded. A year of local sales cannot distinguish “the market fell” from “it is January”, and a model that tries will confidently report a winter trough as a declining market.

Your sales are blended with a published index

The local estimate is combined with the FHFA house price index for your metropolitan area, weighted by how precise each one is. This is load-bearing rather than a fallback: local comparables alone cannot resolve a six-month adjustment to useful precision at any realistic sample size, which is a measured finding and not an opinion. The weight each side received is printed on the exhibit.

Past the index's last published month, it is projected forward at its own most recent twelve-month rate rather than held flat — holding flat is measurably biased low — and the projection is both priced into the interval and disclosed.

How far past your last sale we will go

If your effective date sits more than 45 days after the most recent sale in the file, the tool stops trusting the local curve there and says so. Market movement in a gap between your last sale and your effective date is real, and a tool that silently pretends the trend simply stopped will understate it.

What the 95% interval is checked against

The claim that a 95% interval contains the true market movement 95% of the time is tested by simulation rather than asserted. Markets with a known planted trend are generated and the interval is required to contain the truth between 93% and 97% of the time — at 50, 100, 200 and 500 sales, under both a market that moved steadily and one that ran and then turned, with 1,200 repetitions per case. If any case falls outside that band the build fails and the release does not go out.

The same numbers you reviewed

The screen and the PDF are produced by the same code, so what you sign is what you reviewed. Each exhibit carries a hash of the sales that went in, so the inputs behind a given exhibit can be identified later.

5. What this deliberately does not do

• It does not analyse active listings or pending sales. UAD 3.6 requires you to analyse them; a time-adjustment exhibit is not the whole market analysis. List price is not sale price, and the ratio between them moves inside the very window being fitted, so splicing them in naively introduces error rather than removing it.
• It does not decide anything. It produces a supported rate and an interval. Whether an adjustment is warranted, and what goes in the report, is your judgement and your signature.
• It is not certified by anyone. No GSE certifies appraisal software, and any vendor implying otherwise is overstating it. What we can say precisely is in the next section.

6. What SB4-1.3-09 asks for, and where it is

The UAD 3.6 policy supplement requires the appraisal report to “summarize the supporting evidence and include a description of the data sources, tool(s), and technique(s) used”. Every element is on the exhibit, so you can cite it rather than rewrite it.

Questions this page does not answer are welcome at any time — the address is on the Terms of Service page.