Summary
A matched pair tells you what two buyers paid for one difference. It's the strongest evidence you can put on the page, and the narrowest. Build pairs that differ in one thing, time-adjust before you compare, report the spread across your pairs instead of the average, and check the answer against a method that uses every sale you have.
What a pair proves
Two sales alike in everything but one feature. The price difference is what that feature was worth. The logic is simple, which is why every reviewer asks for paired sales and every textbook starts there.
It proves less than it usually gets asked to prove. A pair is two transactions. It carries the tastes, the financing and the bargaining of two buyers and two sellers on two days. Take a different pair from the same street and the answer moves, sometimes a little and often a lot.
That movement isn't a flaw in the method. It shows how much room there is around the number, and that belongs in the report, not averaged out of it.
Building a pair that differs in one thing
The method rests on "alike in everything but one." In residential work that is never quite true. The job is to get close enough that what is left over is small next to the thing you're measuring.
- Time-adjust first. Two sales six months apart differ by the feature and by the market. Apply the market conditions adjustment before you extract anything else, or the two stay mixed together. See supporting a market conditions adjustment.
- Match location tightly. Same subdivision, same schools, same side of the main road. Location will swamp whatever you're measuring, and there's no good way to correct for it afterward.
- Match the big physical items. Living area, age, condition, site. A pair that differs by one garage bay and four hundred square feet isn't a garage pair.
- Check the terms. Concessions, unusual financing, a sale between family. Any of these can be worth more than the feature you're trying to price.
- Write down what still differs. Every pair has something left over. Naming it is what separates a documented pair from a convenient one, and it tells the reader which way the error runs.
Two pairs don't give you a rate
One pair gives you a point. Two pairs give you two points, and then it is tempting to average them and call the rate derived. With two numbers you can't tell a tight answer from a loose one, and how loose it is turns out to be the part somebody challenges.
Report a range instead. Run as many pairs as the data supports, show the spread, and pick a number out of it for a stated reason. "Six pairs indicated between forty and seventy dollars per square foot. The four closest in age and condition fell between fifty and sixty. Fifty-five was selected." That is support. One averaged figure with nothing behind it claims a precision you didn't earn.
Sometimes the pairs scatter so far that no range means much. That's still a finding. It usually means the market doesn't price this feature consistently, or the pairs were less matched than they looked, or there are too few of them. Any of those beats a number chosen to fill in the cell.
You picked the pairs yourself
You chose the pairs, out of a set you assembled, to measure a variable you selected. That is three chances for your own judgment to shape the answer instead of the market. None of it shows up in the result, because a pair chosen to agree with you looks exactly like a pair that matches well.
Set the screening rules before you look at prices, so pairs aren't kept or dropped based on whether the answer suits. Then report the pairs you threw out and why. If the method never once surprises you, the data isn't really testing it.
Check it against a method that uses every sale
Paired sales uses a handful of transactions and ignores the rest. So pair it with something that uses the whole set and puts up with more noise in exchange for more observations. Grouped medians across the set, a regression on the variable in question, or a cost check where cost is a fair stand-in for what the feature contributes.
Neither method settles it alone. When the pairs and the wider analysis land close together, the conclusion is well supported and you can say so in a sentence. When they disagree, work out why. It usually points to a market split in two, a variable that moves with the one you're measuring, or a feature the market prices unevenly.
What goes in the exhibit
- The pairs, with both sales identified and the leftover differences named.
- The time adjustment applied to each, and the date it runs from.
- What each pair indicated, and the range across them.
- The second method and what it produced.
- The rate you selected, and why it sits where it does in the range.
That's one page. Write it as you work and it falls out of the analysis. Write it later and it costs you an afternoon of remembering. See writing support a reviewer can follow.
Common questions
How many pairs are enough?
Enough that the range holds still when you add one or drop one. That depends on your data, so any fixed number quoted as a standard would be made up. If one pair moves the conclusion, the conclusion rests on that pair, and the report should say so.
Can I use listings to build pairs?
A listing is an asking price, not an agreement, and asking prices run high. Listings can bracket a conclusion or show which way the market is moving. They are weak as the main evidence for a rate, so if you use them, say that is what they are.
What if the only good pair is outside my comparable neighborhood?
A pair from a competing area can work if the two markets really are alike, and it is on you to show that rather than assume it. Usually a wider range from local sales is more defensible than a tight one borrowed from somewhere else.
