Many large American assessment jurisdictions still treat land valuation as an afterthought, and whenever deficiencies in land valuation are brought up, these jurisdictions respond to critics with a perfectly sensible question:
Well, what do you suggest we do instead?
Unfortunately, the IAAO, the governing body that sets standards, isn’t much help in this regard. The entire chapter on “Land Valuation” in their standard on Mass Appraisal handbook is two whole paragraphs:
Furthermore, whenever I propose some method to assessors, they respond with genuine concerns:
We need to be able to explain and defend land valuations It needs to be straightforward to retrain our staff It needs to not be a bunch of extra work
Fancy computer models have their merits, but even when they perform well they tend to score poorly on exactly these three points, so adoption tends to be an uphill battle. Is there perhaps a simple and straightforward method we’ve been overlooking? What if we flipped the entire question of land valuation on its head?
What’s the simplest thing a jurisdiction can do to improve land valuation?
As always, we’re not looking to make the perfect the enemy of the good. What we really care about is improvement over the status quo, which in many cases is a simple but misguided rule of thumb. Is that setting our sights too low? On the contrary, history strongly suggests “good enough” is an acceptable target; the historical German colony in Qingdao, China enjoyed great economic success directly in line with theory despite a fairly crude valuation method. Meanwhile in the USA, we can draw similar confidence from the long legacy of the Somers System, a popular wisdom-of-the-crowds approach that was used to power land valuations in large American cities across, and which persisted for many decades.
We also have a simple, objective test that will tell us for sure when we’ve hit the mark:
An LVT based on these land values will yield the benefits predicted by theory.
Either that will happen or it won’t. If it does, we know we have adequate land values.
Now, we want to be reasonably confident in our valuations before we run a live economic experiment on real people, so we should go a bit further to justify ourselves. Let’s do that now by thinking through the question of land valuation specifically in terms of incentives.
Features we expect
The concrete, observable, testable, benefits we’re constantly touting about Land Value Tax boil down to two main things:
Building is not punished Land speculation is punished
These points expand into several testable corollaries:
If a landowner improves their property, their land values should not increase based on that fact alone
Owners of vacant and underutilized land will pay more than under conventional property tax
Side-by-side owners of physically/legally identical land will pay the same taxes regardless of what is built on their properties
Assessed land value is proportionate to how desirable and useful the land is
These corollaries are all pretty easy features to test for: a land valuation will either have them or it won’t. With these tests in hand, let’s now start with the simplest, dumbest possible way to value land, then improve it one step at a time until it satisfies all of our criteria.
The Simplest, Dumbest Possible Method
The simplest and easiest way to value land is with an entirely flat valuation. Pick a number, any number, and multiply that by each parcel’s size. If we imagine a city laid out on a rectangular grid, our valuation map could look something like this:
Let’s evaluate this incredibly simple and dumb method. Surprisingly, it somehow already passes two of our criteria:
Land values don’t go up just because you built something: TRUE ✅
Side-by-side land pays the same: TRUE ✅
…however, it fails the other two:
Vacant/underutilized land pays more than property tax: FALSE ❌ (Downtown vacant lots will pay much less)
Land value is proportionate to desirability/usefulness: FALSE ❌ (The most valuable land is massively undervalued)
The only thing this method really achieves is horizontal uniformity, or “consistency,” one of the three virtues of Mass Appraisal for the Masses. As for the other two virtues, “accuracy,” and “fairness,” this method falls flat.
This method’s chief deficiency is in treating all land the same; land in the urban core becomes massively undervalued, and land on the rural fringes could easily be overvalued. Besides being both inaccurate and unfair, it likely yields a tax break to the worst land speculators in town.
The Next Simplest, Dumbest Method
The first method’s only virtue is that it makes local land values uniform. However, it fails to capture any values that accrue due to location. Let’s improve on it.
We know that land in the center of the city tends to be more valuable than land far away. A simple way to model this would be to just assign local values based on distance from the city center. We look at some market evidence to calibrate the high and low end of the scale, as well as the falloff (exponential or logarithmic or something). Then we establish circular bands of land value based on how far away each location is from the city center, evoking the von Thünen bid rent model:
This would allow our land valuations to have local side-by-side consistency and capture the strong central value concentration that we tend to see in big cities. How does this method perform?
We maintain our first two points:
Land values don’t go up just because you built something: TRUE ✅
Side-by-side land pays the same: TRUE ✅
And add another one:
Vacant/underutilized land pays more than property tax: TRUE ✅
However, we fail spectacularly on the fourth:
Land value proportionate to desirability/usefulness: FALSE❌
Although we’re likely to properly capture high-value land in the city center, we are going to have major geographic distortions applying a simple model like this to an actually existing city. Just look at this map of Baltimore:
This is a simple map of sale price per square foot (ppsf), revealing how whacky real spatial patterns of urban real estate prices can get. Here’s the same kind of map for Philadelphia:
I don’t really need to show any more cities to drive the point home—every county and city is unique. Even if there’s strong value concentration in the central business district, it’s not uncommon to see large drops in value just a few blocks away.
If we were to paint a simple circular distance gradient over either of these cities, we would wind up with wildly unfair land valuations in any location where local prices surge or dip. A simple circular distance trend cannot capture the fact that property values (and thus land values) can jump sharply from neighborhood to neighborhood.
Just treat land as a % of total value
The next simplest thing we could do is to simply treat land on a fixed percentage basis.
First, we would focus all our efforts on correctly estimating the total market value. This is something most assessors do a pretty good job of already, and for which there are already ample tests to grade them on. Then, we would work out land value on the back end as a simple % of the total market value of every parcel. This target percentage is called a “land allocation,” and it’s incredibly simply to use—just multiply against it.
Many jurisdictions will even use an entirely flat land allocation % for the entire jurisdiction. That is to say, a 15%, or 20%, or 30% (or whatever) land allocation for land regardless of location, land use, or any other characteristic. Land is just always X% of total assessed value, always, everywhere.
This is kind of a crazy thing to do, but let’s not even challenge this practice for a minute and just follow it where it leads, because even when laboring under such a blanket restriction, there is a single, simple, reform that can make a big difference in land valuation quality.
One weird trick
The key thing is that there’s a “Right Way™” and a “Wrong Way™” to apply land allocation percentages, as we uncovered in our report on Baltimore vacant land assessments. Here’s a quick illustration, using a 20% allocation:
The Wrong Way™ is to take the target land allocation and apply it directly to every individual lot’s total assessed value. This is wrong because it will generate large side-by-side differences in land values within an area, solely because of differences in improvements, not land.
The Right Way™ is take the target land allocation, then determine the prevailing price and size of the most typical kind of improved lot for the area, and use those to work out a prevailing local land rate. That standard land rate is then used to value all land in the area, ensuring horizontal uniformity.
What’s interesting is that, as long as you do it the Right Way™, even a single flat allocation percentage results in surprisingly decent land valuations, at least from the perspective of economic incentives.
However, doing it the Wrong Way™ destroys those same incentives, underscoring the importance of getting this detail correct.
The Right Way™ vs. The Wrong Way™
First, let’s consider the Wrong Way™, where every parcel’s land is valued at exactly 20% of its market value:
In this simple toy example, we have a nice house valued at $200K, an average house valued at $150K, and an older house valued at $100K. We also have a vacant lot, and direct market evidence showing us its value is $30K.
The vacant lot is nothing but land value to start with, so its value remains at $30K. The nice house’s land, even though it’s sitting right next to the vacant lot, gets overvalued at $40K because the allocation implicitly sucks in some improvement value. The older house’s land, having less improvement value to start with, winds up valued less than the vacant lot at $20K. The average house’s land gets valued at about the same level as the vacant lot, $30K.
What are the incentives in this scenario?
Land values don’t go up just because you built something: FALSE ❌ (Under the Wrong Way™ , a parcel’s land value is directly proportional to its total property value, which increases with improvements.)
Side-by-side land pays the same: FALSE ❌ (Under the Wrong Way™, side-by-side differences in improvements cause immediate side-by-side discrepancies in land value.)
Vacant/underutilized land pays more than property tax: UNKNOWN ❔ (This depends entirely on how vacant land values are properly calibrated. The Wrong Way™ provides no method at all for vacant land, which opens the door to rock-bottom nominal values, as we saw in Baltimore.)
Land value proportionate to desirability/usefulness: SORT OF❔ (More valuable improvements tend to be built on more valuable land. By definition, this tags land in the most desirable places with a higher value than in the least desirable. But only if it is improved land!)
The only incentive the Wrong Way™ comes close to hitting is to generally assign a higher value to land in proportion to its desirability. This will only be true at a zoomed-out level, and only if the land is improved. Meanwhile it creates all kinds of distortions on the local level, creating ample incentive for speculation, disincentives for building, and encouragements to hold the most valuable land out of use.
Okay, now let’s consider valuing land using the allocation method The Right Way™, even in a situation where we are stuck with a single blanket land allocation figure like 20% for the entire city.
In this situation we don’t take each parcel’s value and apply 20% to it, we instead carve our jurisdiction up into local areas or neighborhoods, and then within each area, we establish a “local prevailing price” per unit land size. We can do this either by finding the median price and size of the most typical kind of property, or just take the median price and size of everything within local bounds. We take 20% of that prevailing market value, and divide it by the prevailing size to establish a local land rate. Then we paint all local parcels’ land values with that same local land rate.
The vacant lot, as before, remains at $30K. The big change is that now every improved lot’s taxable value converges on the same value as a local vacant lot’s. Even if we did not have direct land evidence to derive the vacant lot’s value, this would still be the case; we would just paint the vacant lot with the same local land rate as everything else.
What happens to the incentives?
Land values don’t go up just because you built something: TRUE ✅ (A parcel’s land value depends only on the local land rate, which depends not on any individual parcel’s total market value, but on that of the the typical parcel)
Side-by-side land pays the same: TRUE ✅ (All land in an area is valued using the same local land rate)
Vacant/underutilized land pays more than property tax: TRUE ✅ (Vacant/underutilized land is valued the same as the typical local improved parcel, a massive shift from improved property to vacant/underutilized property)
Land value proportionate to desirability/usefulness: TRUE ✅ (Higher value areas yield higher land rates)
It’s pretty fascinating what a big difference a small change like this can make. The best part is, it’s perfectly in line with existing IAAO recommendations, it doesn’t require a ton of new work, and it’s perfectly easy to explain and defend. In fact, because neighboring properties will have the same land values, there’s fewer side-by-side discrepancies to generate protests over.
Also, by following a simple, auditable rule, the assessor’s individual degrees of freedom are limited, so there’s fewer ways to game and influence the system.
Judging Neighborhood Quality
On that note, how do we know if the neighborhoods are well drawn? Fortunately, these can be held to an empirical standard and tested accordingly. A well-drawn neighborhood is one in which:
Building typology is more or less uniform Variation in ppsf of improved property is small Major building characteristics—building size, age, etc.—exhibit minimal variation.
A large single-family neighborhood with a commercial corridor running straight through it where ppsf jumps 300% is an objectively poorly drawn neighborhood, as are two regions jammed together where the west half trades around $50 ppsf and the east half trades for around $100. If multiple land uses are mixed together in a neighborhood in a way that’s not easy to separate contiguously, a parallel neighborhood should be developed for each land use, and each should be used to set a corresponding land rate. We should also consider that in most cases these neighborhoods have been drawn for us already by the local assessor, are used to drive location adjustments under conventional mass appraisal, and we can apply these kinds of tests to them right now.
If there’s any doubt, we can simply chart these things on a public map and the most egregious mistakes will quickly become visually apparent.
Good Enough Already?
Ironically, this investigation has likely revealed a lower-bound “simplest method for land valuation that could possibly work”—just carve your jurisdiction up into areas where you have good reason to expect land values to be uniform, then assign a local land rate based on a fixed % of the market value of the prevailing property type.
It’s a promising sign that this method produces land valuation maps of the same basic form as Qingdao’s, which simply assigned uniform land rates within specified areas:
I’m willing bet that if we were to implement an LVT off a system that used the method we’ve arrived at, we would see most if not all of the predicted beneficial effects, even though land value is calculated in such a simple way. One can also see how simple and easy it would be for an assessor to implement.
That said, there are various ways in which it could be improved.
Going Further
It’s unlikely that prevailing land value is going to stick to a single % throughout the entire city, because certain areas are going to be more or less well utilized relative to their highest and best currently-legal potential use. The areas that are the best utilized will have the lowest land allocation %, and the areas that are the least utilized will have the highest land allocation %. It would be useful if we were able to dial in the land allocation based on local evidence.
Calibrating local land allocations
To better calibrate local areas, we could employ a Bayesian approach; setting a baseline land allocation for the whole city based on general zoomed out evidence, and departing from that locally wherever we have better evidence for local land values, such as vacant sales, teardown sales, and sales of new construction (where it’s easiest to estimate replacement cost of the improvements).
Furthermore, once we’ve established some local land rates, we can identify neighborhoods that have similar characteristics and prevailing prices, and match them:
If we assume that two neighborhoods with similar physical characteristics and similar prevailing prices have similar land values, we can apply a matched land value rate by applying a simple sales-comparison approach at the neighborhood level rather than the at the parcel level. This should be backed up by constantly looking at everything on a map and employing some common sense tests.
Handling different kinds of land within an area
Next, not all land, even on a given street, is going to necessarily be the same value. For townhomes or condos, “end units” might have different values than “center units,” and so on. Additionally, land value can be affected by features such as: size, shape, land use restrictions (i.e., zoning), topography, flood plains, and the like. Let’s go over these briefly.
Land size
This is the single biggest issue to deal with. You might naively expect that if a one-acre lot sells for $100,000 in a particular neighborhood, that a three-acre lot would sell for $300,000, because it’s three times as big. What actually tends to happen is that large lots are cheaper on a $/sqft basis than similarly-located small lots, even while large lots remain more expensive in absolute terms.
This makes perfect sense. Unless we’re talking about agricultural land, once you get excess land past a certain point doesn’t increase the utility of a building you might put there. You would naturally expect a minimum amount of land that makes the lot “viable” for the prevailing land use of the area (and which is surely anchored to local land use restrictions), and some threshold beyond that at which any additional land becomes merely “nice to have.”
Wherever we have evidence of this, the local rate should not be flat, but instead a curve or table that takes land size as its principal input. This should be calibrated in such a way that the most common size of land in the area gets the normal land rate, and exceptional sizes deviate from there.
As an aside, if land area boundaries are drawn tightly enough, some of this becomes taken care of automatically, as parcel size tends to be strongly spatially correlated, with similarly sized parcels tending to cluster together within neighborhoods.
Land use restrictions
Land use obviously has a strong effect on the price of land, because it literally dictates what you can and can’t do with the land. In most places in the US, zoning tends to be fairly uniform across neighborhoods, but one should pay special attention to sudden changes, particularly within an otherwise homogenously zoned area.
A particularly pernicious kind of land use restriction is the deed restriction. These are individual requirements attached to the deed itself that seek to govern what all future purchasers of the property are allowed to do with the land. A common example is a golf course, country club, or resort, attaching a deed restriction to itself that demands that all future buyers must use the land for the same purpose. This is a common tax dodge, a way to “have your cake and eat it too”—the location remains as desirable and useful to you as ever, but now you’ve legally outlawed demand from anyone who might want to use the land for something else (typically to build housing), and likewise removed their bids from consideration when the assessor comes knocking. In this way a golf course in the middle of a residential area is able to secure a “market based” assessment for itself far below that of what all its land would be sold for if it was carved up and sold as residential lot developments. Unfortunately, the assessor’s hands are often tied when it comes to deed restrictions; dealing with the inequities they cause is a problem to must be solved at that political level.
In many assessment neighborhoods, zoning and land use restrictions will be more or less uniform across its boundaries. When they aren’t, parallel neighborhoods should be built for each major land use category, assuming there’s enough samples of each. When there is instead just isolated spot variation, this can be accounted for with adjustments to land value; the more restrictive, the lower the land value.
Other issues
The shape of a lot can greatly affect its usefulness, because the land use you had in mind might not literally fit on it, or hook up properly with the rest of the neighborhood/city. Additionally, the shape of the lot might render some portion of its square footage unusable, so the lot might be valued as if its “effective size” were smaller than its “nominal size.”
Topography can obviously make a big difference in the usefulness of land. The physical topography of a parcel can be modeled quite simply by downloading the latest Digital Elevation Model from the US Geological Survey and then calculating two statistics per parcel: average slope of the parcel, and bumpiness. A parcel that is tilted too steeply off grade, and/or one that is insufficiently smooth, is going to be much more difficult to build on than a perfectly flat and level lot.
Flood plains can obviously cause hyper-local changes in value over short distances. A good example is my own house, which sits next to a creek and is just a few feet lower than everyone else’s house in the neighborhood, so when it rains super hard, water will tend to come into our garage but not anybody else’s. Of course, when we bought the house we were assured that it was “well outside” the 500-year flood plain…
Hyper-local effects encompass any number of other things that can cause a particular lot be worth more or less than what you would typically expect for the area. An experienced assessor with good local knowledge of their area, receiving constant feedback from their community, should build up a continuously updated map of these over time.
Handling all of this stuff in detail is beyond the scope of this essay, but a wealth of literature has been published on them already. There are two broad approaches; the first is “paired sales analysis”, where you find comparable sales with and without the feature in question and directly look at changes in sale price. The second approach is to use any number of computerized methods for regression, principal component analysis, or machine learning. Any of these methods can be employed to generate simple local adjustments that can be justified by data.
The Least We Can Do
Right now, one of the greatest bang-for-buck reforms in property assessment is to just get assessors to start applying land allocations the Right Way™—as locally uniform land rates within each assessment neighborhood—even if they are totally plucking their target land allocation percentage out of thin air and applying it to the whole city. Doing so will generate land values that have the correct economic incentives, and, at least in cases where the assessor is otherwise doing a good job, probably aren’t all that far off from “true” platonic land values anyways. This would be a perfectly adequate basis for taking an existing property tax and applying a Universal Building Exemption to it.
The next best thing we can do is to calibrate local land allocations based off of the best direct land value evidence, and follow that up with sensible adjustments for all the ways in which land values can and should differ within a local area. And in a perfectly ideal world, we’d not just have sale prices to work with, but also direct rental figures for everything (and a chocolate pony).
Either way, this method is simple and easy to explain. It also slots right into prevailing practices that many assessors are doing already, and doesn’t require anyone to blindly trust the output of a black box computer algorithm.
The only necessary prerequisites to deploy this method, right now, in any assessment jurisdiction in America, are the following:
Total market value assessments that pass basic scrutiny Decently well-drawn assessment neighborhoods
As long as we have those two things, we can use the “least we can do method.” Any amount of direct land value evidence lets us improve upon it by degrees. The next step is to feed it directly into LVTShift and model a revenue-neutral tax shift under Universal Building Exemption, and this will immediately tell us whether we can craft a pitch to the local public that will be politically popular.
Finally, we should consider how things might change moving forward once an LVT is in place. I submit that land becomes easier to value after the institution of an LVT, because as soon as buildings are exempted from taxation, the incentive for taxpayers to hide the value of those improvements doesn’t just disappear, it flips—now they want to run to the assessor with their itemized receipts for new construction. This floods the assessor with not just up-to-date building characteristics, but also fresh evidence for directly calculating land values, with which they can calibrate their existing neighborhood-level land allocation percentages. Furthermore, long-held vacant land now starts to hit the market in much higher volumes, and older properties will be redeveloped, so we will see more new construction. All of this increases the evidence we need.
Much work remains to be done, but I think there’s a lot of value to be had simply by starting with small but meaningful improvements to the status quo.