We now consider the second “time adjustments” method noted  in the FHFA paper (24-07).  It, too, is noted as being “commonly” used in appraisals. The first method we reviewed was “paired sales”.  Now we look at the second method, called “grouped data.”

This “related technique, grouped data analysis, involves grouping data by an independent variable such as date of sale and calculating equivalent typical values. The grouped sales are studied in pairs to identify the effect on a dependent variable such as the unit price of comparable properties.”  (The Appraisal of Real Estate, 15^th ed. p.372.)

However, it goes on to say:  “Although paired data analysis of sales or rents is a theoretically sound method, it may be impractical and produce unreliable results when only a narrow sampling of sufficiently similar properties is available.”  Darn.

It continues for several paragraphs to discuss the “theoretically sound,” but “time consuming,” “difficult to apply,” data-lack problems, similarity issues, special-care requirements, unreliable results.  And finally:  “The difference measured may not represent the actual difference in value attributable to the characteristic being studied.”  Darn again

Finally, it states that these methods can be useful if:  “their weaknesses are appropriately recognized, and they are applied in combination with other methods.”  And they:  “may be impractical … when only a narrow sample of sufficiently similar properties is available.”

Other methods . . .

All this, if and only if, you take “special care.”

It gets worse.

The use of “grouped pairs” for time adjustments is problematic on other counts.

• Comps at different dates need different groups.
• Grouping sales by date means “discarding data” (the exact sale date).
• People demand grouping by month or quarter, rather than around each comp sale date.
• Common sense and intuition can be wrong. For example, the statistical fallacy called “Simpsons Paradox.”  (Which shows how the trend direction can be wrong from grouping).
• Common sense and lack of training in data science combine to make this a “feel good” believable (credible) groupthink solution, which flat out gives wrong results.
• Common sense and legacy appraisal training – can result in fiascos such as the 1004MC, which worsened the bias by grouping by unequal group widths. (6-3-3 month groups.)

Please see my two related peer-reviewed journal articles in the Appraisal Institute The Appraisal Journal.

So, what have we learned so far?

• One pair, even if it is a “pure” pair, is biased by known and unknown “influences.”
• Multiple pairs are better, but variation within each set of pairs creates uncertainty.
• Grouped pairs will seldom coincide with the date and times back to each comparable.

We have learned that we are nowhere near the ideal, best practice we want.

The method we want should use all available information, be relevant to the competitive market segment, be quick, visual, and easy to calculate, to understand, and to explain.  It should be easily reviewed and checklist audited.  It should actually work.  No magic.  No wishful thinking.  No appraisal groupthink.

Goal:  The ideal method will:  1) apply the most relevant market segment; 2) is visual self-explanatory;  3) is model/judgment transparent, and; 4) is distinct/easy to audit and review.