> run a regression on comparable sales and take human judgment out of it.

It's extremely rare to find enough comparable sales of parcels of land only to run a regression on. In an intensely developed area, the overwhelming majority (like 98+%) of sales will be inextricably linked sales of land and improvements of that land.

I have some intellectual agreement with LVTs, but the practical implementation is daunting and it seems to make it much harder to appeal a bad assessment.

▲

nostrademons 8 hours ago | parent | next [-]

You wouldn't run the regression on parcels of land only. You'd run the regression on parcels of land + improvements, run a PCA on the various attributes of both the land (acreage, viewshed, proximity to amenities, transit access) and the improvements (square footage, # BR/BAs, year of construction, presence of pool, presence of deck) to determine which factors materially affect valuation, and use the regression coefficients to subtract that out. So for example, you might find that going from 1BR->2BR costs an extra $200K, going from 2BR->3BR costs an extra $600K, a 9-rated school is an extra $500K, etc.

Now you take a sample of sales within the area of the parcel in question, subtract out the regressed values for the improvements, and smooth & average it, and you have a rough approximation for the value of the land. It's not going to be perfect: it assumes that features are linearly independent, for example. But it should be close enough, and doesn't suffer from the same incentive problems as letting a human assessor put in values for those features.

▲

sokoloff 8 hours ago | parent [-]

When I was shopping for homes, I found several things: the city's records of improvements was comically disconnected from the ground truth (admittedly much of that was due to taxation of improvements which would wash out in the long-run) and that the inherent human judgment ("this one was done by a blind, half-assed house-flipper", "this one needs $100K of maintenance soon", "this one was last updated in 1970", "this one was designed and built by absolute master craftspeople", "this one is next door to a known problem house", "this school is a GreatSchools 9, but buyers believe it sucks") could swing the price between two spreadsheet-identical homes by 40%.

Either you need to bring in human judgment to account for those factors, or have enough data to hope that people will accept that these differences will reliably come out in the wash. If you could, Zillow probably wouldn't have lost around $1B robo-buying during a massive real-estate boom.

▲

nostrademons 8 hours ago | parent [-]

Could also introduce a mechanism for homeowners to correct the official city record, with documentation. Under a LVT the incentives would be in place for this, because undocumented improvements generally improve the structure value of the house, which homeowners are not taxed on, yet if the structure value is subtracted from the recent sale price it would reduce the land value, which homeowners are taxed on. (This is not the case with a straight property tax, where homeowners have every incentive to lie about improvements to avoid a re-asssessment.) And it's in the city's interest to have accurate records about the state of each home.

▲

sokoloff 7 hours ago | parent [-]

Right, the incentive just has the opposite sign now. Find a way to take basement storage and turn it into just barely legally qualifying as bedrooms and bathrooms, even though every human would immediately classify and value it as if it was ordinary basement storage. Same with the living room/den/family room. In a lot of places you’d just need to make a small back-to-back closet and ensure egress is met (or grandfathered as-built) and they’d legally be able to called bedrooms. My 4 BR, 2.5 bath house becomes a 7 BR, 3.5 bath place, saving the new buyer (or me) on land taxes forever, making them (or me) willing to pay more to do that pointless remodeling.

“Why does this basement hallway have 4 half-baths and 4 tiny bedrooms full of shelves, 2 bedrooms right off the main entry, and no living room? Taxes.”

I just can’t see how to eliminate human judgment from the valuation process of the parts when humans are unavoidably the ones valuing the combination.

▲

nostrademons 7 hours ago | parent [-]

It comes back to "AI" meaning "aggregated intelligence" rather than "artificial intelligence". Statistical valuation approaches basically mean replacing the judgment of one assessor with the aggregated (through some smart averaging process) value of many purchasers. Aggregates are much more robust to both manipulation and misjudgment than the individual data points used to make them. Ergo, simply replacing one person's judgment with thousands of peoples' judgment will make the process more accurate.

	▲	sokoloff 5 hours ago \| parent [-]
		Assuming that works for the determining of the standardized, publicized assessment method, I think it stops working as well when people look at the published assessment method and have huge incentives to find ways to boost the assessed value of improvements, thus reducing the value assigned to the land when they appeal based on the notion that the arms-length purchase for the combination was $X and the publicized value of the improvements calculates out to $Y, therefore leaving only $Z for the value of the unimproved land. Actually, now that I think about it more, the output of the model would have to be a direct estimate of $Z (rather than an estimate of $Y, which is surely easier), making some of those concerns moot when the most recent sale was long ago (but not for new transactions). I still can't figure out what the appeals process when the model gets it wrong (or is perceived to be wrong) is going to be based on. It seems like we're likely to end up right back at "human judgment", though perhaps that of an actual judge trying to interpret the parcel's value taking into account the model's output and the arguments of the plaintiff land owner instead of a city assessor trying to argue comparable sales with a land owner, which is a process most everyone can wrap their head around and have a pretty good feeling how they'd fare if they took it all the way to court.

▲

fernmyth 8 hours ago | parent | prev [-]

Here's a dumb idea:

- Take the most recent sale prices of all neighboring parcels within a quarter mile

- Take the land-area weighted average price per square foot

- Assess the given property as its land area times that average

Practically, this can be assessed without ever visiting any of the properties, and there are no games for either the assessors or the owners to play to manipulate it. Each neighbor serves as an example of the potential use (and therefore value) of local parcels. Yes, your hole-in-the-ground gets taxed the same as a skyscraper of the same size. Have you considered building something useful on it?

(Yes, as usual, parks, churches, and other non-economically-extractive community amenities would be exempt from _paying_ the tax, but that doesn't change the assessment)

	▲	sokoloff 7 hours ago \| parent [-]
		It’s simple, hard to game, and executable which puts it ahead of many other ideas on the topic. But: Land above a subway station or in a main business square is worth A LOT more than the same amount of land 1300 feet away. Waterfront land is wildly more valuable than the land just across the street. In districts with front setbacks, corner lots are less valuable per square foot than mid-block lots. And I’d really hate to own a normal mini-skyscraper 4 blocks away from the skyscrapers overlooking Central Park and have my building’s land be valued as the average of the sum of all the units overlooking the Park. But your “dumb” idea is smarter than most I think.