Paired sales is the fourth of five “acceptable” methods for time adjustments, according to the Fannie Mae selling guide. Once again, the method is neither defined nor explained.
Editor’s Note: Read the entire series (so far) here.
A look at USPAP shows no reference to grouped or “paired sales” as a way to support an opinion of trend, nor is there any mention of “adjustment” at all! None.
We must look at legacy established repeated education for support of time adjustment. The main source of appraisal education is the Appraisal Institute’s The Appraisal of Real Estate.
“Paired sales” is only mentioned as an “intense” method to support comparable adjustments.
“Paired data analysis is based on the premise that when two properties are equivalent in all respects but one, the value of the single difference can be measured by the difference in price between the two properties.” (p. 372)
“Paired data analysis should be developed with extreme care to ensure that the properties are truly comparable and that other differences do not exist.” (p. 372)
The TARE goes on to say that although paired data analysis is a theoretically sound method, it may be impractical and produce unreliable results when only a narrow sampling of sufficiently similar properties is available. Note that none of the above statements refer to market movement, only to property adjustments. None.
What seems to be missing in all the literature is: How is similarity measured or estimated?
The above “premise” of equivalence in a pair fails when seeking “time” pairs! To apply “time adjustment” to individual property sale prices — the two sales would need to have:
- Timespan equal to, or very close to, the timespan to the comparable time back.
- Identical buyer and seller motivation. (Not likely, and not measurable.)
- Identical condition — with the typical sale/resale being a fixer rehab.
Worse yet, the above conditions must apply to each of the “comps” used. Unlikely. But it must be acceptable if — as per TARE p. 78 — “Great care must be exercised when using paired data.” Unfortunately, there is no help on how to analyze with “great care.”
The data science approach to appraisal (EBV- Evidence Based Valuation)© goes much more precisely into matched-pair or grouped-pair methods. EBV applies measurable similarity, with the contrast/comparison being against the means or medians and other descriptives of each group.
Grouped and paired methods are problematic. (See Simpson’s paradox.) To compare two time groups of similar sales, we need one group centered on the comparable’s sale date. Then we need another group time-centered on the effective date of value! Is this possible? Yes. It is possible for a retroactive date of value, where we can time-bracket with sales before and after that date of value.
It is not possible for a current date of value. We have no sales into the future to bracket the date, future and past. 
We can gather a group of sales that occurred before the effective date of value, (say a three-month, 90-day, group). Unfortunately, this group would then be date-averaged around 45 days before the needed effective date of value.
EBV eliminates these difficulties and biases of the “pairing” methods. While pairing may work for some other “elements of comparison,” it returns wrong answers in every case. Stick to time-series methods!