interesting point. There are a host of adjustments that are subjective. I have yet to make my first adjustment for "view" in 31 years. My mentors pounded in my head that these were too nebulous to quantify with any certainty. Pulling a single paired sale from the stack makes for a guess, not an adjustment. Find the most similar view. Same with quality. Condition-age - well I can extract that from the comps although that is difficult if the house has been remodeled extensively. I can extract accrued depreciation from any sale though. So it really boils down to finding the most similar properties. I will ignore less similar properties that are closer or more recent.
In many of my assignments, I find effective age (condition+age) and SF to be the 2 factors besides land that drive values - and land is valued as if vacant and available for its HBU- thus is a dollar for dollar adjustment.
Sometimes. And sometimes it doesn't matter. I worked for a number of banks during the crunch who sent notices to borrowers saying that they were re-appraising their property and if the LTV wasn't still 80% or whatever that they would need to come up with more money. Many couldn't. I recall 2 chicken farms in particular where the people were "underwater" despite making all their payments. There were people and groups who tried to get the FDIC to relent and offer forebearance to those who were making timely payments. Unfortunately, the FDIC basically insisted the banks shore up their balance sheets period. The banks then foreclosed, sold the places for what they could, and wrote down their reserves accordingly. One bank survived, one didn't.
In this area Condition, Quality, Aesthetics and Functional Utility are big drivers of value.
But regression is a tricky thing. It is far more devious than most begin to imagine - even if they have been working with it for years.
Let's take Quality. You can look at a house in comparison to competing properties and judge its Quality with a relative value like % better or worse. MARS can give you a dollar value for that. But that adjustment will be same for all houses with a given quality measurement (I call CQA) regardless of size. So, you get the same adjustment for a house that is 2,0000 and 4,000 sf. But the 4,000 sf home has twice of everything including quality components. The question is, did the regression already take that part of quality dependent on GLA, Bath Count, Bedroom Count, etc., and put it in those adjustments?
Yea, you have just opened another can of worms.
And, ..., in MARS we can use 2-D interactions to take care of this problem partially:
So, with a given GLA and SaleAge (Date of Sale), you get a value contribution. That is to say, the value contribution of SaleAge does not just depend on how old the sale was, but also how large the size of the house was. This is the way it should be done. Well, you might add, what about Quality which is in the residual and doesn't come under this Stage I adjustment? That's another problem. (This graph by the way is for an area of San Mateo, CA, as of 5/28/2021.)
And here is related graph for Baths vs SaleAge vs SalePrice that is extremely interesting:
It appears that just after Trump was elected or rather perhaps the tax cut was passed, the homes with the smaller number of bathrooms (2.0-3.0) went up in price and then dropped off as the Fed started increasing interest rates in 2018. The larger homes were not affected. MARS used bath count, rather than GLA to create this value sub-model.
[Note this part of San Mateo borders the "aristocratic" (i.e. 'old money') areas of Hillsborough with many $10M+ homes on the hillsides.]
