Land Stewardship and Development Behaviors Under an Ecological-Impact-Weighted Land Value Tax Scheme: A Proof-of-Concept Agent-Based Model

Abstract

Sprawling land development patterns have exacerbated ecological degradation, social fragmentation, and public health problems. Perverse incentives arise from the ability to privatize collectively created value in land rents and socialize ecological costs. Land value taxation (LVT) has been shown to encourage urban infill development by reducing or eliminating rent-seeking behavior in land markets. However, despite its purported benefits, this tax reform is value monistic in its definition of optimal land use and, therefore, does little to address the lack of non-market information to inform land use decisions. We propose an ecological-impact-weighted land value taxation policy (ELVT) which incorporates the ecological footprint of land use into one’s land value tax burden. We test both proposed policies (LVT and ELVT) relative to a “status quo” (SQ) property tax scheme, utilizing a conceptual spatially explicit agent-based model of land use behaviors and housing development. Our findings suggest that both tax interventions can increase the capital intensity and decrease the land intensity of housing development. Furthermore, both tax interventions can lead to a net profit loss for speculators and a decrease in the average housing unit price. The ELVT scheme is shown to significantly increase urban nature provisions and dampen the loss of ecological value across a region.

1. Introduction

Suburban sprawl has become a defining feature of the modern landscape in the U.S. and other countries [1,2]. It is characterized by dispersed, low-density residential buildings reliant on automobile travel for access to amenities and commercial areas [3]. Today, most U.S. citizens reside in suburban neighborhoods [4], places which account for over 90% of new impervious development [1]. Such rates of development continue to outpace population growth, suggesting an increase in per capita land consumption [1].

The negative effects of sprawling land development span the social [5,6], public health [7,8], economic [9,10], and ecological domains [11,12]. Concerns about sprawl have been raised as far back as 1958, particularly regarding its impact on open-space amenities and public expenditures [13]. More recently, research has demonstrated the economic inefficiencies of sprawl, which can increase public expenditures, as it necessitates an extension of the boundary of public services and infrastructure [14,15]. Suburbs have also contributed to the United States’ oversized greenhouse gas emissions due to their associated higher residential energy burden [16] and longer commuting distances [17]. Jones and Kammen [18] found that suburban households have 25% higher carbon footprints than urban households and comprise around 50% of the total U.S. household carbon footprint.

Of particular relevance to this paper are the environmental consequences of sprawling land development—arguably the most central concern about sprawl today [19]. Suburban development remains a leading cause of the loss of forested rural lands, along with the essential ecosystem services they provide. At the current development rates, an estimated 18 million acres of U.S. agricultural land could be lost, fragmented, or degraded by 2040, with 12 million acres projected to be converted for low-density residential use [20]. However, much of this research simplifies ecological footprint according to land use categories, overlooking the variability in the impacts that different suburban developments can have. For example, turf grass lawns, which cover three times more area than any other irrigated crop in the U.S. [21], are a major contributor to habitat and biodiversity loss. Despite their ubiquity, lawns are not essential to suburban life, but rather a mere landscaping choice shaped by cultural norms. Reforming such practices could significantly reduce the ecological impact of suburban development, even if the trend toward suburbanization continues.

Despite the contribution of each new development to these broader social dilemmas, individual actors remain unresponsive to the collective costs they pose—an evident example of a market failure. In response, governmental agencies at the local, county, state, and in some cases, national level have implemented urban containment policies. These efforts include comprehensive planning, public land acquisition, and various regulations and incentives to establish clear boundaries between urban and rural areas, promoting urban infill and preserving rural land [19]. Common policies include zoning restrictions, development impact fees, property tax adjustments, density bonuses, brownfields redevelopment subsidies, and the public conservation of peripheral lands, among other things [19]. Some research has found these policies to have at least some effects in promoting compact development [22,23], particularly when enforced at the state level [24] or enacted as a suite of regulations, incentives, and public investment policies [19]. Others note the counterproductive effects of these policies, such as increased housing costs [25], which lead to spillover or leapfrog sprawl in surrounding municipalities [26,27]. Despite their best efforts, urban planners and governmental bodies at various scales have not been able to redirect the inertia of development away from suburban forms. Exploring more effective urban containment policies remains paramount and has been explicitly called for by local and state governments [19].

Many argue that prior policy interventions have ignored a crucial driver of suburban sprawl—speculation, e.g., [27,28]. Land, being fixed in supply, derives its value from the demand it garners, its desirability for use or financial returns, and the number of demanders [29]. Increases in a parcel’s desirability and value may result from municipal development or rising consumer interest, neither of which stems from actions made by the parcel’s owner. Thus, it is considered as economic rent [29]. Speculative investment in land prioritizes economic rent capture over capital investment, resulting in “leapfrog development”. Strategies typically involve purchasing large swaths of land outside the city periphery and holding that land out of use until its value appreciates due to the extension of city services, at which point, speculators may sell and purchase land beyond the new periphery [27].

The rent-generating potential of land creates perverse incentives for land use decisions, encouraging land hoarding based on speculative value rather than the land’s immediate use for housing or commercial purposes. In other words, land prices are increasingly set according to their potential to appreciate rather than their use value, which can embed boom–bust cycles in land markets and force many to seek housing on the more affordable city outskirts. It also undermines collective welfare through private appropriation of the public and nature-derived values of place [30]. Policies and public investment, which aim to make city centers more livable and amenable to development, can instead lead to displacement and growing wealth inequality, so long as publicly created value remains available for private appropriation [31]. As just one example, the benefits of urban greenspace have been shown to capitalize on property prices and result in the displacement of existing communities and market-based exclusivity of the subsequent benefits of greenspace [32,33]1. Thus, addressing the speculative dynamics in land markets is essential for crafting more effective urban containment policies.

Land value tax (LVT) is a policy tool for curbing speculation-driven suburbanization by redistributing the benefits of public improvement and local land stewardship toward the community at large, rather than to neighboring parcels alone via price appreciation [30,35]. Originally proposed in the late 19th century by Henry George [30], land value taxation consists of a tax levied solely on the unimproved value of land. As an alternative to the conventional property tax, LVT shifts taxes off capital improvements (which stem from private investment) and on to land rents (which stem from public investment). Consequently, it removes the deadweight loss associated with taxes on land improvements, and removes an incentive to speculate by redirecting the fruits of public service investments back to the municipality. This policy has sustained interest among researchers and policymakers, and has been implemented in over 30 countries, including many places in the U.S. [36]. Researchers have demonstrated the effectiveness of LVT policies in promoting urban infill over urban sprawl [37,38,39], preserving exurban land for nature [40,41], and reducing housing costs [42,43].

However, despite their improvements both to equity and land use efficiency by buffering the exploitation of land rents, conventional land value taxation schemes remain reliant on markets to define the “optimal use” for land and, therefore, lack the ability to relay non-market benefits and costs to landowners in order to, for example, safeguard ecological thresholds [44]. Thus, this sustained reliance on markets and their associated value monism presents a shortcoming regarding the applicability of land value taxation to ecologically sound land stewardship.

Moreover, conventional land value taxation falls short of directly incentivizing private urban or suburban land stewardship due to the non-discriminate distribution of the social welfare created. While a landowner would bear the full cost of land regeneration (including the direct cost and potentially increased tax burden from land value appreciation), they would receive a negligible portion of any monetary benefits (via the broad distribution of new tax revenue).

Although a slew of public action has focused on conservation or incentivizing private land stewardship, such endeavors typically engage in large-scale conversions in rural regions [45]. Urban and suburban areas, where the majority of Americans live, present an underutilized opportunity for the broader adoption of ecological stewardship [45,46,47]. Addressing this gap requires a paradigm shift away from human settlements as sacrifice zones countervailed by far-off conserved nature and towards a view that recognizes the potential for human–nature relationships to be mutually enhancing [48,49]. Urban ecological stewardship also directs the benefits of ecosystem services to within urban and suburban centers, with co-benefits including food access [50], cultural transformation via crowding-in pro-environmental behavior [51], urban resilience [52], and the variety of other benefits associated with proximity to greenspace [53].

1.1. Policy Proposal

This research seeks to introduce and justify the theoretical expansion of a land value taxation scheme aimed at encouraging the broader adoption of land stewardship practices within and outside urban centers. We propose an ecological-impact-weighted land value tax scheme (ELVT) which penalizes land degradation and subsidizes ecological stewardship in high-ecological-importance areas. Under the ELVT scheme, a conventional LVT rate is first applied and then scaled up or down according to non-market information and landowner decision making. The ecological weighting considers the following two factors: the value that a particular parcel would have to collective ecological health and service provisions (deemed “ecological potential”), and the degree to which the parcel, in its current state, is meeting that potential (deemed “ecological value”). For example, a parking lot or industrial site directly abutting unfragmented natural land may have a high ecological potential and low ecological value. So too might a parking lot that is far from any greenspace. Ultimately, such designations of ecological potential and ecological value would be at the discretion of the enacting municipality.

Under ELVT, ecological value would have an inverse relationship with tax burden, the magnitude of which is mediated by ecological potential. Those who live in a location with high ecological potential and have converted their lawn into a garden or native habitat (for example) would face a lower tax rate, which is subsidized by the higher tax rate of those who maintain ecologically degrading land use practices but still benefit from the ES provisions around them. For parcels with a low ecological potential, the tax would mirror that of a conventional LVT.

ELVT provides a means to align individual action with collective value creation. It coheres with the original justification for land value taxation, namely re-socializing collective values, but allows for a broader definition of such values. This policy also aligns with broader policy suggestions such as common asset trusts, which builds an institutional capacity to manage common goods toward collective benefit [54], and has received some recent attention in political spheres (see [55]). Finally, ELVT could provide a passive means of agglomeration incentives, in which land stewardship increases the land value and benefits of restoration among neighboring parcels, thus incentivizing their restoration and bolstering the contiguity of intact land.

While the effects of LVT have a long history of examination, to the authors’ best knowledge, researchers have yet to explore the outcomes associated with the particular mechanisms of the proposed ELVT scheme. Nevertheless, limited research on similar proposals supports further examination. Lafuite et al. [41] showed that a natural land depletion tax has the effect of internalizing the value of biodiversity, resulting in more diversity being preserved at equilibrium through more labor-intensive agricultural practices that reduce natural land conversion. The tax was also predicted to mitigate the vulnerability of the economy to overshoot and collapse given the time-delayed externalities of biodiversity loss [41]. Xiong and Li [56] proposed an ecological deficit tax that compensates the difference between land’s ecological carrying capacity and its current ecological footprint. Using a computable general equilibrium model, the authors found that the ecological deficit tax had the effect of reducing the region’s ecological footprint in the long run. Similarly, Muellbauer [57] made a case for a “Green Land Value Tax”. Though not focused on ecological land stewardship, it entails a conventional land value tax scheme, except that property owners would receive a discount for measures which lower energy use.

1.2. Study Overview

In this study, we utilize a spatially explicit agent-based model of land use behaviors and housing development as a proof-of-concept exploration of the emergent outcomes associated with LVT and ELVT policies. We employ boundedly rational decision making among homeowners, developers, and speculators operating within a regional land and housing market. Our analysis focuses on four experiments in which the model is tested under different parameter settings. In experiment 1, we model the three tax schemes (with approximate revenue neutrality) under three different configurations of homeowner preference, which influence the resulting development patterns toward suburban sprawl, compact urban infill, and something in between. In experiment 2, we perform a sweep of the percentage of speculator agents in the model to explore their role in mediating spatial and socio-economic outcomes. In experiment 3, we perform a sweep of the magnitude of divergence between the LVT and ELVT tax rates to explore the effect of ELVT under increasingly extreme operationalizations. In experiment 4, we relax the assumption of revenue neutrality and perform a sweep of the LVT rates to assess the spatial and economic effects of more aggressive rent socialization tax policies. Our key finding is that both LVT and ELVT tax interventions increase housing density and decrease land-intensive housing, while ELVT further densifies the urban core by moderating the loss of natural beauty, which can otherwise drive prospective homeowners away from the urban core.

2. Methods

Urban development is an inherently complex process; emergence results from non-linear interactions between strongly coupled social, environmental, and economic systems, featuring spatial and multi-scale interactions and context-adaptive decision making amongst a heterogenous population of land use decision makers [58]. Such complexity precludes the use of traditional economic or statistical models [59]. Instead, researchers have increasingly turned to agent-based models (ABMs) [60]. ABMs take a micro-lens of analysis, controlling only individual decision making rules and system contexts, therefore facilitating the examination of how complex emergent macrophenomena can result from individual action, as well as the outcomes associated with intervening at the individual level [61].

This research takes inspiration from a robust foundation of academic study using spatially explicit ABMs to understand residential locational choices and urban sprawl. Magliocca et al. [62] incorporated endogenous, spatially heterogenous land and housing price formation to demonstrate the occurrence of urban sprawl and leapfrog development driven by consumer preferences and various economic influences on the profitability of farmland conversion. Brown and Robinson [63] utilized empirical residential locational preferences and showed how more realistic heterogeneity in preferences led to greater urban sprawl in location choice. Jackson et al. [64] demonstrated the emergence of gentrification as mediated by residential migration and resulting changes in land rents. Still, others have shown the important dynamics that exist between land use activity, ecological services, and resilience, e.g., [65,66].

Researchers have also utilized ABMs to explore the effectiveness of tax interventions associated with land value on urban sprawl. Hosseinali et al. [67] found that a policy that increased the land cost by 50% (presumably by way of a tax) in areas planned for growth was the most successful tested policy in encouraging denser development patterns. Furthermore, Chen [68] demonstrated the effect of differential bargaining power on the effect of land taxes. He found that increasing exurban land costs through a development tax was ineffective in a “thinly traded” land market, because it increased buyer bargaining power such that they could capture a greater surplus from the exchange and were, thus, encouraged to purchase exurban land.

Our ABM simulates a regional housing market and resulting land use patterns to the extent necessary to explore the effect of property taxes on developer, speculator, and homeowner decision making. We built the model from scratch using Python (Version 3.10.8). The complete code and instructions to reproduce our results are publicly accessible in our code repository (https://github.com/DakotaBWalker/Speculation-and-Land-Tax-ABM) (accessed on 30August 2024). A cellular automata structure is used to represent a hypothetical geographic region as a grid of cells (parcels) with land use states and underlying environmental characteristics. Spatially heterogenous cell characteristics provide information from which agents choose actions that fulfill their motivations and maximize their utility, constrained by imperfect information and bounded rationality.

2.1. Environment

The hypothetical geographic region is represented by a 100 × 100 grid of cells, with each cell representing a plot of land with underlying locational, economic, and environmental characteristics. These cells take one of the following four discrete land use states: conserved land that is unavailable for agent purchase or development, undeveloped land that is also un-owned, land that is undeveloped but owned by an agent, and land that has been developed with housing. All cells falling into the lattermost state also contain values which describe the number of housing units in the cell, as well as the number of those units currently available for purchase. The environment also characterizes ecological value (EV) as discrete states where conserved or undisturbed forest land is equal to 1, disturbed but undeveloped land is equal to 0, and developed land is equal to −1, unless restoration has occurred, in which case, EV is equal to 0.5.

Four environmental characteristics form the foundation of agent preferences. Based on the hedonic price principle, we assume that a particular property can be understood as a bundle of characteristics from which consumers derive utility based on their particular preferences and importance. Thus, variation in property prices can be used to reveal preferences for property attributes and associated willingness to pay. In the case of residential choice, price is understood as a function of physical, neighborhood, and locational attributes, e.g., [69,70]. For example, empirical research has demonstrated a strong association between real estate prices and surrounding forest amenities or natural beauty [71,72,73], neighborhood population or building density [74,75], proximity to a central business district [70,76], and property size [75,77,78].

The natural beauty (NB) matrix represents the hedonic value of a plot for environmental, aesthetic, or recreational purposes. Pre-specified forested cells are given the highest NB value of 1. A distance decay function is then used to gradually lower the NB value with increasing distance to a forested cell. NB values are also reduced by one sixth of the cell’s density value (however, with a fixed NB min value of 0.3) to represent the typical loss of natural beauty in high-density urban locations. The proximity to the central business district (CBD) matrix represents each cell’s proximity to the central cell, normalized to between 0 and 1. The housing density (density) matrix represents the development intensity of the surrounding cells, which is calculated with the sum of housing units in the surrounding 7 × 7 cell grid divided by the maximum value of surrounding development, assumed to be an average of two units per cell. The lot size (lot_size) matrix describes the acreage of each cell, which is assumed to increase with distance to the central cell (dist) as follows:

$$lot\_size=0.075\times \left(dist+1\right)+0.25+\epsilon$$

(1)

$$\epsilon =N\left(o,{\displaystyle \frac{lot\_size}{4}}\right)$$

(2)

Initial lot sizes are specified, then given random variance (ε), with the standard deviation being proportional to the initial lot size. As such, we assume that lot sizes fall between approximately 0.325 acres at the center and approximately 6 acres at the outermost cell. lot_size affects agent preferences in the following two ways: first, it is used to calculate the total land tax burden, which directly affects agent willingness to pay; second, it is used to calculate the lot size per housing unit once development occurs, which affects homeowner housing preference.

Other environmental characteristics speak to the costs and profitability of land or housing ownership. Per acre land values are specified for each cell, representing the cost per acre of purchasing a particular plot of land devoid of improvements. Because our model does not endogenize price formation in the land market, we approximate land values based on homeowner preferences, neighborhood housing density, and total regional housing as follows:

$$LVPA=land\_pref\times ({nbr\_housing+1)}^{1.4}\times {total\_housing}^{0.9}+LVmin$$

(3)

where land_pref is the average homeowner land preference (0,1) for each cell across 50 randomly sampled homeowners, nbr_housing is the sum of occupied housing units in the surrounding 13 × 13 neighborhood, total_housing is the sum of all occupied housing units across all cells, and LVmin is the specified minimum land value per acre. Upon calibrating the LV equation with the specified exponents, the approximation reproduces a characteristic power law distribution of land values across a region, as well as the land price appreciation that results from nearby or aggregate development (See Supplementary Materials; [79]).

The environment also includes separate tax rate matrices for improvement value (IVT_rt) and land value (LVT_rt). For the SQ tax scheme, both tax rate matrices are equal and uniform across space. For the LVT tax scheme, LVT_rt is spatially uniform, while IVT_rt is homogenously equal to zero. This also applies to the ELVT scheme, however, with LVT_rt varying across space according to EV and EP. The formula for calculating LVT_rt under an ELVT regime is as follows:

$$LVT\_rt=\tau +E_{burden}$$

(4)

$$E_{burden}={\displaystyle \frac{EP\left(-EV\right)}{eco\_burden\_denom}}$$

(5)

where τ is the baseline LVT_rt and eco_burden_denom is a scalar used to arbitrarily control the magnitude of divergence from the base rate at EP and EV extremes. Due to the tax rate variation that results from land use changes under ELVT, each cell also includes hypothetical tax rate values if each parcel is restored or developed to allow agents to anticipate the taxes associated with their actions.

The EP matrix was specified to represent one of the many ways a municipality might communicate the locations in which land stewardship is most important. We used a distance decay function to create a gradient of values from 0 to 3 as a function of distance to any conserved land. In practice, ecological potential could be a far more sophisticated description of ecological integrity, essential habitats, etc., or a more arbitrary description of, for example, neighborhoods in need of better stormwater management.

2.2. Agents

Agents represent land users of various types who can buy, sell, develop, and restore parcels of land. Agents spend money buying land, developing land, restoring land, and paying taxes. Agents gain money by way of an assumed fixed exogenous income (for homeowner agents) and land or housing sales. Agents are heterogeneous in their preferences, wealth, and altruism. They make decisions based on the expected benefits of different actions relative to their own motivations, constrained by their wealth. Decision making is implemented under imperfect information and bounded rationality, in which agents are given a randomly generated subset of all possible options and choose randomly from a smaller subset of those options that satisfy at least a given threshold of relative benefit. Agents interact indirectly, through sensing and reacting to the environment, which is impacted by other agents, as well as through a housing marketplace where purchase price is directed from the buyer to the seller.

“Homeowner” agents seek to buy housing that meets at least a given threshold of utility, constrained by their wealth. To model agent utility with respect to a housing unit, we use a modified multiplicative Cobb–Douglas function [80], which has been used extensively to model consumer choices by way of preferences and product characteristics [81] and is commonplace within the urban economic modeling of residential choices [63,82,83,84,85]. The function takes the form shown in Equation (6), where the utility that a particular homeowner (r) gains from a particular housing unit (h) is a function of three vector variables, each with a length of four (i.e., m = 4). One vector comprises objective cell characteristics (γ_h); one comprises the agent’s preferred values for each characteristic (β_r); and the third vector comprises the preference weight (α_i) that the agent assigns to each cell characteristic.

$$U_{r,h}={\prod }_{i=1}^m{\left(1-\left|{\beta }_{r,i}-{\gamma }_{h,i}\right|\right)}^{{\alpha }_{r,i}}$$

(6)

Homeowner agents draw each preference weight (α_r,i) by randomly sampling within a given range from a uniform probability distribution. Agent preference weights are then normalized such that they sum to 1, which allows for utility to be compared across agents. The objective cell characteristics and preferred values (also sampled at random from a given range) fall between 0 and 1. Thus, agent utilities are constrained to between 0 and 1, where a utility of 1 would occur only through a perfect match between the preferred and actual values for all four cell characteristics. For natural beauty (NB) and CBD proximity (CBD) cell characteristics, the idea value (β) is assumed to be the max value (i.e., 1) for all agents, which allows the equation to be simplified to the objective cell characteristic raised to the preference weight. For density and lot size cell characteristics, agents are given unique preferred values (ideal_density and ideal_lot_size), such that the same lot size may fall close to the preferred lot size of one agent, but may be far from ideal for another agent. Another way of representing the operationalized utility function is shown in Equations (7)–(9). This particular implementation of the Cobb–Douglas utility function most resembles that of Brown et al. [84].

$$U_{i,j}={{NB}_h}^{{\alpha }_{NB,r}}\times {{d\_match}_{r,h}}^{{\alpha }_{density,r}}\times {{CBD}_h}^{{\alpha }_{CBD,r}}\times {{lot\_match}_{r,h}}^{{\alpha }_{lot\_size,r}}$$

(7)

$${d\_match}_{r,h}=\left(1-\left|{ideal\_density}_r-{density}_h\right|\right)$$

(8)

$${lot\_match}_{r,h}=\left(1-\left|{ideal\_lot\_size}_r-{lot\_size}_h\right|\right)$$

(9)

The utility of agent r for a particular housing unit h is then multiplied with their wealth (assumed to represent the maximum amount of money they would be willing to spend on a home) to determine their willingness to pay (WTP_r,h) for the unit. WTP is also affected by the present value of the unit’s property taxes, given a homeowner future discount rate (future_disc_rt_homeowner), as well as the neighborhood vacancy rate, as follows:

$${WTP}_{r,\mathrm{h}}={U\_hu}_{r,\mathrm{h}}\times {wealth}_r\times \left(1+vacancy\_multiplier\right)-present\_value\_tax$$

(10)

$$vacancy\_multiplier=\left(0.25-{vacancy\_rt}_h\right)\mathrm{\ast }0.25$$

(11)

Because the vacancy rate is fixed at between 0 and 0.5, vacancy_multiplier can reduce or increase WTP by as much as 6.25%.

Under the ELVT scheme, homeowners that already own a home have the option to restore their land. Restoring land refers to the act of investing a pre-determined sum (restoration_cost) per acre to change the land use attributes of a cell such that it then provides greater ecosystem services (EV changes from −1 to 0.5). The probability of choosing to restore (p_restore) for a homeowner (r) and owned parcel (h) is as follows:

$${P\_restore}_{r,h}={altruism}_r\times {economic benefit}_{r,h}\times {neighborhood adoption}_h$$

(12)

In which altruism (0,1) reflects the agent’s own devotion to providing collective goods, economic benefit is the expected present value of savings on tax payments divided by income, and neighborhood adoption represents the percentage of plots in the surrounding 7 × 7 cell grid that have restored their land. The homeowner chooses to restore if their P_restore value is greater than a random number between 0 and 1 drawn from a uniform distribution.

Homeowners also have the option to sell their current home if they become overburdened by the cost of ownership or if the associated utility of other housing units on the market at or below their willingness to pay exceeds the utility that they derive from their current home by at least a given threshold (pref_dif_to_sell).

“Developer” agents seek to supply homeowners with housing by buying and developing land. They make decisions about where to buy land for development based on expected profit, represented by the percentage difference between the 75th percentile of the aggregate willingness to pay across a random sample of homeowners and the expected cost of developing. Costs include the present value of the tax burden after development, given developers’ future discount rate (future_disc_rt_developer). Potential development sites are constrained to a random subset of plots (rnd_off_limits) to invoke imperfect information, as well as to locations in which the cost of development would not exceed their wealth and where the vacancy rate is less than 0.25. Upon calculating the expected profit for each potential development site, developers then choose randomly between the top x% (rnd_choice_pct) of plots with regard to their associated expected profit, invoking bounded rationality.

Upon buying land, developers determine how many units to build. Developers are each given unique values for square footage per unit, building costs per square foot, and the desired improvement value to land value ratio by drawing randomly with uniform probability from given parameter ranges (unit_sf_range, build_cost_psf_range, and desired_IV_LV_ratio_range, respectively). The agent (d) can then calculate how many units to build (N_units) on a plot (h) as follows:

$${N\_units}_{i,j}={\displaystyle \frac{{LC}_{d,h}}{{IC\_per\_unit}_{d,h}}}\times {desired\_IV\_LV\_ratio}_d$$

(13)

$${LC}_{d,h}={LV}_d\times \left(1+{\displaystyle \frac{{LVT\_rt\_developed}_h}{{future\_disc\_rt}_d}}\right)$$

(14)

$${IC per unit}_{d,\mathrm{h}}={unit\_sf}_d\times {build\_cost\_psf}_d\times \left(1+{\displaystyle \frac{{IVT\_rt}_h}{{future\_disc\_rt}_d}}\right)$$

(15)

Under the ELVT tax scheme, developers may consider the profitability of building half the units they intend and restoring the other half of the site. They consider the half restoration if their personal altruism value is greater than a randomly drawn value between 0 and 1. If so, they estimate the cost per unit of developing half the units, given a reduced tax rate. If the cost per unit plus a profit premium of 15% remains lower than homeowner willingness to pay, they choose to conduct the half-restoration development.

Developers then estimate willingness to pay per unit by sampling 20 homeowner agents at random. If the cost per unit is less than or equal to the 75th percentile of homeowner willingness to pay, the development occurs. The price per unit is set somewhere between the cost per unit and the estimated willingness to pay, with the exact point between the two based on the neighborhood vacancy rate as follows:

$$price=cost\_per\_unit+{\displaystyle \frac{WTP-cost\_per\_unit}{2}}\times \left(1+vacancy\_multiplier\right)$$

(16)

$$vacancy\_multiplier=\left(0.25-vacancy\_rt\right)\times 4$$

(17)

Therefore, we assume that a vacancy rate of zero would set the price at WTP and a vacancy rate of 0.5 would set the price at the cost_per_unit, which reflects how differential bargaining power between buyers and sellers affect price and how increasing supply can drive profits to zero [86]. Upon setting a price, the housing units are added to the market and made available for homeowners to buy.

“Speculator” agents seek to buy and sell land or housing units in such a way as to capitalize on land value appreciation that results from other agent actions. Their decision making is based on expected profit of buying, holding, then selling a plot, calculated as follows:

$$BC\_ratio={\displaystyle \frac{dLV}{LVT\_rt}}$$

(18)

where dLV is the average percentage change in land value per year over the past 3 timesteps. If the maximum BC_ratio across all available plots is greater than 10%, the speculator chooses at random between the plots with the x% (rnd_choice_pct) highest BC_ratio. Speculators continue buying plots at each timestep, so long as the aforementioned profitability condition is met.

At each timestep, speculators sell one of their owned plots either randomly with a probability of 0.1 or if their wealth is less than or equal to zero. Under the ELVT tax scheme, speculators can also consider land restoration. For all owned unrestored land, they look at the profit potential of restoring the land (as a result of lower taxes), as follows:

$$P\_restore={\displaystyle \frac{\left(\frac{LV\times \left(LVT\_rt-LVT\_rt\_restored\right)}{future\_disc\_rt}-restoration cost\right)}{restoration cost}}$$

(19)

2.3. Feedback

Several key feedback loops between agent actions and environment form the basis of recursivity in the model (Figure 1). With regard to preferences, development affects both density (positively) and natural beauty (negatively), which then affects the preferred locations at the next timestep. Newly restored cells also affect natural beauty (positively). Changes in development and locational preferences impact land values and the resulting cost of purchasing and developing, tax burdens, and returns for existing owners of land or housing. Fast-appreciating land value attracts speculators to the area, with whom developers must compete for access to desirable land and homeowners must compete for access to housing.

2.4. Scheduling

The model is run by first initializing the starting number of agents, as well as the cell grid environment with all corresponding geographic characteristics. At each timestep, agents act by buying, selling, or restoring cells. After all agents act, the ancillary cell characteristics (e.g., density and vacancy) are then updated based on the changed land use or housing ownership. A pre-specified number of new agents are then added to the simulation before the timestep advances and the agents act again. The high-level process sequence is visualized in Figure 2.

2.5. Parameters and Initialization

As a conceptual model primarily for the purpose of exploratory analysis, our model features limited empirical calibration. Like many other conceptual ABMs, e.g., [87], we hesitate to set some parameter values for empirical data without emulating other aspects of the context in which those empirical data are based. Rather, we seek to employ general rule-of-thumb values, where necessary, to form parameter ranges from which agents draw at random.

Table 1. Model parameters under “baseline” configuration.

Parameter	Description	Baseline Configuration
max_timestep	Number of timesteps for which the model runs	20
P_initial_development	Used to scale the randomly generated initial housing development	0.2
N_init_agents	The number of agents initialized into the simulation at timestep 0	300
N_new_agents	The number of agents added to the simulation at each subsequent timestep	30
ag_type_wt	The weights used to condition randomly chosen agent type when each agent is initialized. Organized as [developer, homeowner, speculator]	[0.15, 0.80, 0.05]
ln(starting_wealth_homeowner)	The mean(std dev.) of the normally distributed values from which homeowner agents draw then exponentiate to derive starting wealth	13.25 (0.25)
ln(starting_wealth_developer)	The mean(std dev.) of the normally distributed values from which developer agents draw then exponentiate to derive starting wealth	16 (0.5)
ln(starting_wealth_speculator)	The mean(std dev.) of the normally distributed values from which speculator agents draw then exponentiate to derive starting wealth	15 (1)
rnd_off_limits	The probability from which each available land cell draws to determine whether it is not available for purchase (re-drawn at each timestep)	85%
rnd_choice_pct	The top percentage of cell location choices from which agents randomly choose	10%
altruism_homeowner_range	The range of uniform probability altruism values from which homeowners draw	[0.2, 1]
altruism_developer_range	The range of uniform probability altruism values from which developers draw	[0.1, 0.5]
future_disc_rt_homeowner	The rate at which homeowners discount potential income in future years	10%
future_disc_rt_developer	The rate at which developers discount potential income in future years	25%
future_disc_rt_speculator	The rate at which speculators discount potential income in future years	33%
pref_dif_to_sell	The minimum difference in utility between the most preferred on-market housing unit and the currently owned housing unit above which the homeowner sells their current property	0.15
NB_pref_range	The range of values from which homeowners sample uniformly to determine their natural beauty preference weighting	[0.01, 0.4]
CBD_pref_range	The range of values from which homeowners sample uniformly to determine their CBD preference weighting	[0.2, 0.6]
density_pref_range	The range of values from which homeowners sample uniformly to determine their density_match preference weighting	[0.2, 0.4]
lot_size_pref_range	The range of values from which homeowners sample uniformly to determine their lot_size_match preference weighting	[0.01, 0.2]
Ideal_density_range	The range of uniform probability ideal_density values from which developers draw	[0.1, 0.7]
Ideal_lot_size_range	The range of uniform probability ideal_lot_size values from which developers draw	[0.05, 1]
LVT_rt	The rate at which land value is taxed. Formatted as [SQ rate, LVT rate, ELVT rate] and chosen to approximate revenue neutrality	[0.035, 0.10, 0.10]
IVT_rt	The rate at which improvement value is taxed. Formatted as [SQ rate, LVT rate, ELVT rate]	[0.035, 0, 0]
eco_burden_denom	The denominator of the eco burden calculation. Used to arbitrarily control the magnitude of divergence from the base rate at EP and EV extremes	50
unit_sf_range	The range of uniform probability unit_sf values from which developers draw to calculate their cost_per_unit	[1000, 2500]
build_cost_psf_range	The range of uniform probability build_cost_psf values from which developers draw to calculate their cost_per_unit	[100, 200]
desired_IV_LV_ratio_range	The range of uniform probability desired_IV_LV_ratio values from which developers draw to calculate how many units to build	[4, 7]
restoration_cost	The cost to restore one acre of land	USD 5000

summarizes the parameters and associated values or value ranges used for the model. The model begins by initializing a specified number of agents (n_init_agents), whose type is drawn at random from a weighted list (ag_type_wt), along with the agent’s initial characteristics depending on the agent type. A notable assumption regards development costs and the preferred amount of capital investment. We assume a unit square footage range between 1000 and 2500 and a per square foot cost range from USD 100 to USD 200, which represents all project costs except for land costs [88,89,90]. Therefore, the average cost to build a unit (not including the land cost) is USD 225,000. We also assume that developers seek to build units such that the land cost represents between 12.25 and 20% of the total costs of development on average (desired_IV_LV_ratio_range = [4, 7]) [89,91,92,93].

We then initialize the environment’s land use states and ecological values, which were randomly generated, but remained fixed for all simulation runs (Figure 3). From this, we derive ecological potential as a distance decaying value from conserved land, and natural beauty as a distance decaying value from any forested cell (EV = 1), with stronger values weighted toward those that are conserved. Finally, tax rates are chosen based on the total tax revenue of initial runs of the model, such that the tax revenues in the final timestep under the LVT and ELVT scenarios are roughly revenue neutral in relation to the status quo tax scenario.

2.6. Experiment Design

We perform four experiments to assess the relative outcomes under different parameter assumptions.

Table 2. Unique parameter values used in model experimentation.

Experiment 1: Land Use Patterns
Parameter	Baseline Scenario	Sprawl Scenario	High-Density Scenario
NB_pref_range	[0.01, 0.4]	[0.3, 0.5]	[0.01, 0.2]
CBD_pref_range	[0.2, 0.6]	[0.1, 0.3]	[0.5, 0.7]
density_pref_range	[0.2, 0.4]	[0.2, 0.5]	[0.4, 0.6]
lot_size_pref_range	[0.01, 0.2]	[0.2, 0.4]	[0.05, 0.1]
Ideal_density_range	[0.2, 0.5]	[0.01, 0.3]	[0.3, 0.8]
Ideal_lot_size_range	[0.05, 0.75]	[0.25, 1.0]	[0.05, 0.15]
LVT_rt ([SQ, LVT, ELVT])	[0.035, 0.1, 0.1]	[0.035, 0.135, 0.135]	[0.035, 0.1, 0.1]
IVT_rt ([SQ, LVT, ELVT])	[0.035, 0, 0]	[0.035, 0, 0]	[0.035, 0, 0]
Experiment 2: Speculator Count Sensitivity Sweep
Parameter	Min	Max	Increment
ag_type_wt_speculator	0.01	0.29	0.07
N_init_agents	300	420	30
N_new_agents	30	42	3
Experiment 3: ELVT Magnitude Sensitivity Sweep
Parameter	Min	Max	Increment
eco_burden_denom	20	140	30
Experiment 4: LVT Rate Sensitivity Sweep
Parameter	Min	Max	Increment
LVT_rt	0.05	0.25	0.05

Note. All experiments use the baseline parameter settings (Table 1) except where specified under each experiment.

summarizes the parameters used in each experiment, with all other parameters being fixed, using the values specified in Table 1. In experiment 1, we model three land use scenarios via changes in homeowner preferences, as follows: a baseline scenario represents middle-of-the-road urban development; a sprawl scenario represents a stronger preference for larger lot sizes, low-density, and high-natural-beauty plots; and a high-density scenario represents a stronger preference for locations in close proximity to CBD and with a high density.

In experiment 2, we assess the impact of a growing speculator population on sprawl and housing availability. We perform this via a sweep on the probability of choosing a speculator agent (ag_type_wt) when randomly assigning initial and new agents. We also increase the total number of agents so as to keep the homeowner and developer counts relatively constant for each iteration. In experiment 3, we assess the impact of ELVT under increasing divergence from LVT rates at the extremes (i.e., the magnitude of the subsidy and penalty) via a sweep of eco_burden_denom values. In experiment 4, we assess the system outcomes under increasing LVT rates.

Model stochasticity arises from the various randomly drawn elements of the model (e.g., homeowner preferences, available land at each timestep, and agent cell choice), as well as through the path dependence from the effect of prior agent actions on environmental conditions. To account for this, we utilized the Monte Carlo average and standard error over 20 runs of each unique initialization to converge on a robust estimate of each system metric. Therefore, for the multi land use tax comparison, a total of nine unique initializations yielded 180 runs. For each sensitivity analysis, a total of five unique initializations yielded 100 runs.

2.7. Analysis

Given the limited calibration and conceptual framing of the model, we adhere to an exploratory analytical approach, focusing on broad system dynamics and emergent patterns to build intuition and inspire future inquiry, rather than attempting to forecast specific regional land use outcomes or provide a means of decision support for policymakers [94]. More specifically, we direct attention to the relative changes among both tax interventions (as compared to status quo) pertaining to the following questions: (1) To what degree do changes in homeowner preferences lead to differing land development patterns? (2) Can differing tax schemes reinforce or redirect land use patterns and emergent outcomes for three common land use patterns as enacted by changes in homeowner preferences? (3) How does the magnitude of speculator interest affect such emergent patterns, and can LVT and ELVT dampen their effect? (4) How does ELVT differ from LVT under various tax scheme operationalizations?

Table 3. Measured land use, economic, and ecological outcomes.

Category	Metric	Description
Urban Densification	Avg HU lot size (acres)	The average lot size per housing unit for housing which is occupied by homeowners
	High-density housing rate	The percentage of homes which occupy cells with a density above 0.4
	Largest patch size	The largest number of contiguous developed cells across the region
	Housing within urban boundary	The percent of housing units that are within 20 cells from the center
Housing Availability	Total housing	The sum of all housing units in the region
	Avg HU price	Average price for one housing unit
	Avg homeowner without home	Percent of homeowners who either have not found a home or have recently sold their home to move
	Avg vacancy rate	The average rate of vacancy across all cells and all timesteps
Ecological Integrity	Ecological value change	The percentage change in the sum of ecological value relative to initial ecological value
	Urban ecological value	The sum of regional ecological value
Land Use Incentives	Avg tax burden by home type	The average tax burden for homeowners with high-density housing (housing_unit_lotsize < 0.125), mid-density housing (0.125 < housing_unit_lotsize < 0.25), and low-density housing (housing_unit_lotsize > 0.25)
	Avg tax burden by agent type	The average tax burden of homeowner, developer, and speculator agents
	Avg change in wealth	The average change in wealth for each agent type from timestep 0 to the final timestep

Note. Metric is calculated for values of the final timestep unless otherwise stated in the description.

summarizes the specific metrics used to investigate our research questions pertaining to urban densification, housing availability, ecological integrity, and land use incentives.

3. Results

3.1. Experiment 1

Under the conventional SQ tax scheme, our model broadly reproduced archetypal development patterns via changes in agent preference weightings (Figure 4). The “baseline” land use scenario demonstrated a central urban core with a downward-sloping gradient of housing density extending out to scattered suburban arms of development. The “sprawl” land use scenario demonstrated low-density decentralized housing clusters with minimal urban core housing. The “high-density” land use scenario demonstrated a tight core of high-density housing with relatively few suburban dwellings beyond it.

For the baseline scenario, the initial development tended toward the center. The broad ranges of homeowner preferences seemed to reinforce project costs as the primary driver of location. As initial development increased land value in the center, developers moved just outside to capture lower land costs with similar homeowner willingness to pay. However, as land values increased further, developers could increase the number of units built, and, therefore, profitability again favored the urban core. This dynamic continued until the loss of natural beauty or the increase in density lowered homeowner preference for the urban core, at which point, broader exurban development ensued.

In the spawl scenario, development locations heavily favored locations with a high natural beauty, initially in close proximity to the urban core. However, such development increased density and reduced natural beauty, such that subsequent development moved farther from the center toward higher-natural-beauty locations.

In the high-density scenario, homeowner preferences steeply declined with distance from initial development because of their high ideal_density preference. As the central core was further developed, homeowner preference for these locations increased, resulting in a positive feedback loop of tight urban development. Developments moved outward only when all more central land was unavailable.

Figure 5 visualizes the average and standard error for each model outcome under each land use scenario and tax scheme. Relative to the SQ scheme, both LVT and ELVT significantly increased the total housing count and were associated with higher-density dwellings on average. Furthermore, exurban housing development was less frequent under both tax interventions (Figure 6). Both tax interventions were also associated with a significant decrease in housing prices on average (Figure 5). In the baseline scenario, the average housing unit price decreased by 27.8% and 30.8% under LVT and ELVT, respectively. In the sprawl scenario, the average housing price decreased by 18.9% and 2.5% under LVT and ELVT, respectively. In the density scenario, the average housing price decreased by 36.8% and 37.0% under LVT and ELVT, respectively. We attribute this to the increase in development spurred by the tax interventions, which not only allowed developers to spread development costs over more units, but also acted to increase the vacancy rate and, therefore, push prices down. As a result, far more homeowners found houses in this model, with an average of 64% of homeowners unhoused in the status quo scenario as compared to averages of 27.5% and 28.1% under LVT and ELVT, respectively, averaged over all land use scenarios.

Both tax interventions also had the effect of greatly decreasing the profitability of speculative investment (Figure 7). Relative to SQ, the average tax burden of speculators increased by 512% and 580% under LVT and ELVT, respectively. As a result, the change in speculator wealth, on average, decreased from 6.7% under SQ to −1.8% and −1.0% under LVT and ELT, respectively. Figure 7 also shows a significant decrease in homeowner wealth, despite a small increase in tax burden. However, coupled with the finding that the average housing prices decreased precipitously, this may speak to a broader alignment of housing with use value and, therefore, becoming less of an investment asset amongst homeowners. Finally, despite a large increase in tax burden, developer’s wealth, on average, increased under both tax interventions. This, too, may encourage developers to favor high-volume, capital-intensive investment in housing above leapfrog development strategies.

Some notable differences emerged when comparing the effects of the LVT and ELVT schemes. Broadly, ELVT acted to increase urban infill densities, likely as a result of higher urban natural beauty and subsequent homeowner preference. Also, where LVT tended to exacerbate regional ecological value loss relative to SQ, ELVT greatly decreased EV loss to almost neutral levels.

3.2. Experiment 2: Speculator Count Sensitivity Analysis

The results of experiment 2 demonstrate the significant sway speculator agents had in the model (Figure 8). As the number of speculators increased, they tended to decrease the availability of high-value land and housing, which forced broader, low-density exurban development. Moreover, given the increased competition for available housing, housing prices tended upward, as did the percentage of agents unable to find housing. Even though the LVT and ELVT tax interventions increased the average tax burden of speculators, those tax interventions also led to faster-appreciating land values as a result of more development. As such, speculators were still able to profit from the region’s land and housing market.

3.3. Experiment 3: ELVT Magnitude Sensitivity Sweep

The results of experiment 3 signify the need for greater attention directed toward the decision-making criteria of adopting restoration practices. A decrease in the eco_burden_denom value increased the divergence between ELVT_rt and LVT_rt (i.e., increased the magnitude of the increase or decrease in tax burden in high-ecological-potential areas), which should have led to a broader adoption of land restoration. Yet, Figure 9 shows a slight decrease in the number of restored parcels between the lowest and highest eco_burden_denom values. This may have been the result of a higher urban density rate, which meant less homeowners had the minimum lot size for which restoration was possible. Nonetheless, the higher tax rate on ecologically important land acted to decrease the average acres per housing unit and increase the percentage of housing with the urban boundary.

The results also show a significant decrease in the speculator tax burden that resulted from decreasing the eco_burden_denom values. Land-owning speculators were again able to capture a large portion of land rents so long as they restored the land parcel to receive a lower tax burden. Such restorations would increase the surrounding natural beauty and, therefore, increase the land value for the benefit of speculative investment. This, coupled with the finding that the average tax burden on low-density housing decreased and the average tax burden on high-density housing increased, suggests that, as the ELVT_rt diverges further from the LVT_rt, some of the beneficial effects of land value taxation may diminish.

3.4. Experiment 4: LVT Rate Sensitivity Sweep

In experiment 1, we set the LVT and ELVT tax rates such that resulting tax revenue in the final timestep was relatively close to revenue neutral, as compared to SQ. In experiment 4, we showed that land value taxation can further decrease urban sprawl and speculator interest as tax rates increase. We found that, as LVT_rt increased, many of the beneficial effects of the revenue-neutral tax intervention were magnified. As shown in Figure 10, the total housing and housing density continued to increase while housing prices continued to fall. Moreover, where the revenue-neutral tax intervention did not directly address the urban boundary size, a higher LVT_rt significantly decreased the largest patch size and increased the percentage of housing within the urban boundary (Figure 10). Taken together, these results suggest that, with a high enough rate, land value taxation can lead to dense, walkable cities that exert less pressure on surrounding exurban land.

4. Discussion

In this paper, we presented a novel policy tool aimed at buffering the problematic trajectories imparted by rentier capitalism and resulting land use tendencies. The expanded land value tax scheme not only limits the financial viability of land speculation, but also provides a mechanism by which communities can price in the social costs of ecological degradation and subsidize land stewardship. With this conceptual agent-based model, we were able to conduct exploratory analyses in order to build our understanding of the broad changes in outcomes and land use patterns that result from the LVT and ELVT tax interventions.

The results of this study show general alignment with many of the theoretically and empirically demonstrated effects of land value taxation on land development, including higher housing development rates and urban housing capital density [42,95], decreased urban sprawl [37,42], and lower housing unit prices [42,43]. Under a revenue-neutral tax shift (experiment 1), both LVT and ELVT acted to increase the capital intensity of housing development, thereby increasing density and decreasing housing prices. Furthermore, speculation was taxed heavily and became a profit-losing endeavor. However, under the revenue-neutral scenario, the location of developments was still more strongly mediated by household locational preferences than costs. Given a preference for low or medium density, urban sprawl persisted. As we increased the rate at which land value was taxed (experiment 4), homeowners favored highly land-efficient developments in the urban core, more so than any exurban lands they may have favored at a lower tax rate.

We also showed that speculators exert significant sway in land patterns and housing prices (experiment 2). Under the SQ tax scheme, a higher speculator count decreased the total housing and the percentage of housing within the urban boundary, thereby increasing the average acres per housing unit. However, under both the LVT and ELVT schemes, all such metrics stayed relatively consistent as the speculator count increased. While at low speculator count, both tax interventions resulted in a negative wealth change for speculators, as the count increased, wealth change as well as housing prices trended upward.

Our findings also suggested many beneficial effects of ELVT. The proposed tax intervention broadly acted to increased urban housing densities, increase urban greenspace provisions, and greatly decrease the loss of ecological value across the hypothetical region. However, the decision-making criteria for the adoption of land restoration did not show much response to a growing subsidy for doing so. We partially attribute this to homeowners having less means for restoration as housing densifies. Finally, we found that the disincentive to speculate on land value under LVT was diminished under ELVT as the subsidy of land restoration increased; speculators were again able to profit off land rents, so long as they restored the land.

One cautionary result from the model was the faster appreciation of land values under both the LVT and ELVT tax scenarios due to the increased development and density they spurred. Despite the expectation that much of this appreciation would be taxed away and, therefore, attract less price appreciation due to speculative interest, a higher property tax burden can still displace existing residents [96]. This is particularly worrisome under the ELVT scheme given the well-documented evidence of green gentrification dynamics [32]. However, the degree to which gentrification would be dampened were speculation eliminated remains a question. Furthermore, the validity of this result remains questionable, given our proxy land value estimation; further examination is needed with the incorporation of the many drivers of upward or downward pressure on land values, including vacancy, commercial and rental interest, buyer and seller bargaining power, and speculator demand.

By establishing this proof-of-concept study, we hope to provide direction and inspiration for future empirically grounded inquiries. Such empirical data would aim to calibrate this model to a particular region according to local consumer preferences, developer decision-making behavior, land stewardship practices and adoption likelihood, and land values and development costs, as well as more detailed estimations of locally defined ecologically important land and the effects of small-scale land stewardship on ecosystem service provisions. Such local calibrations would also present an opportunity to test how an ELVT policy would interact with an already complex legal framework of land use and property rights.

Furthermore, future research could improve the model’s realism by including a wider array of land use decision criteria and actor types. For example, the inclusion of existing landowners may help to shed light on willingness to sell, tax capitalization in land, and farmland conversion dynamics [62]. A primary feature of land value taxation is to increase the cost of holding urban land idle; therefore, this study’s omission of pressure on existing vacant landowners missed an important aspect of the story. Furthermore, the assumption of one central business district was overly simplistic, and may have decreased sprawl potential as compared to when commercial areas and jobs follow suburbanites [84]. Likewise, the redevelopment of land was not incorporated into the model, which may have unnecessarily locked neighborhoods into a low-density form when redevelopment would have increased the housing capacity in desirable locations. Finally, a comparison of the impacts of the tax scheme vs. other parameter variables (or other policy approaches) can be utilized to shed light on the relative importance of the tax intervention. One example of this is Bell et al. [87], who used a Random Forest Classifier model to describe the relative importance of various parameters for particular outcomes.

While the implementation of the proposed tax reform is not recommended until it can be tested in the context of the locality, such local calibration requires detailed social, economic, and ecological accounting of the region, which can be the focus of local or state policy. As an important proxy for the capitalization of public investment in nearby parcels of land, assessing land value, separate from a parcel’s improvement value, can shed light on the potential tax revenue, distributional, and equity impacts of a switch to land value taxation. Municipalities can instruct property tax assessors to list improvement value and land value separately, even if the tax structure remains the same. While accurate land value estimation is often noted as a hindrance to the implementation of a LVT, recently the use of mass appraisal techniques has drastically streamlined the process, e.g., [97,98]. Likewise, given the significant impact of speculative investment on sprawl, municipalities might consolidate public information on property purchases and local investment behaviors to assess the ubiquity of speculation-based land purchases. This, in tandem with land value data, can help to identify the localized effects of speculative purchases on housing price appreciation and gentrification. Finally, states can reform legislation to allow municipalities to implement a differential rate of taxation on improvement value and land value (a “split-rate tax”) if they so choose. This has long been the case in Pennsylvania, which now has 20 jurisdictions using a split-rate tax model [36] and is currently being considered in Minnesota [99].