You are on page 1of 20

Regional Science and Urban Economics 81 (2020) 103484

Contents lists available at ScienceDirect

Regional Science and Urban Economics


journal homepage: www.elsevier.com/locate/regec

Self-driving cars and the city: Effects on sprawl, energy consumption, and
housing affordability☆
William Larson a, ∗ ,1 , Weihua Zhao b
a
Federal Housing Finance Agency, United States
b
University of Louisville, United States

A R T I C L E I N F O A B S T R A C T

JEL classification: We adapt the classic monocentric city model to consider three main topics related to the possible widespread
R11 adoption of autonomous vehicles (AVs): sprawl, energy consumption, and housing affordability. AVs are modeled
R28 to reduce marginal commuting costs and in some cases, reduce demand for center-city and residential parking.
C60
This creates opposing forces that lead to sprawl in some models and increasing density in others. All models
point to welfare increases, but also increases to energy consumption due to longer commutes, greater traffic
Keywords: congestion, and higher productivity, calling into question claims that autonomous vehicles will save energy.
Autonomous vehicle In most models, AVs lead to greater housing affordability by making suburban areas more accessible, and by
Transportation
reclaiming land that was previously used for parking. Effects of AVs on cities are substantial and depend on the
Energy consumption
manner in which this new technology is implemented.
Land use

1. Introduction acceleration and breaking,4 greater road throughput,5 and higher auto-
mobile utilization rates through car sharing networks such as those pio-
A new technology is under development that has the potential to neered by Uber and Lyft.6
reshape the city–the autonomous vehicle (AV).2 Passengers in these However, unlike other recent transportation innovations such as
vehicles, freed from the need to attend to driving, are likely to face telework or electric vehicles, which simply reduce marginal transporta-
dramatically lower fixed and marginal costs of commuting. These cost tion costs, AVs have a secondary benefit—they reduce the need for com-
reductions are due to greater leisure possibilities while commuting,
reduced collision risk,3 greater fuel efficiency through more optimal


The views expressed in this paper are solely those of the authors and not necessarily those of the Federal Housing Finance Agency or the U.S. Government. The
authors declare that they have no relevant or material financial interests that relate to the research described in this paper.

Corresponding author. Federal Housing Finance Agency, Constitution Center, 400 7th St SW, Washington DC, 20024, United States.
E-mail address: larsonwd@gmail.com (W. Larson).
1
Larson contributed to this research in his personal capacity and not as part of his official duties at the Federal Housing Finance Agency. The analysis and
conclusions are those of the authors alone and should not be represented or interpreted as conveying an official Federal Housing Finance Agency position, policy,
analysis, opinion or endorsement. Any errors or omissions are the sole responsibility of the authors.
2 We define an autonomous vehicle as the U.S. National Highway Traffic Safety Administration’s (2013) Preliminary Statement of Policy Concerning Automated

Vehicles definition of a Level 4 Autonomous Vehicle: “The vehicle is designed to perform all safety-critical driving functions and monitor roadway conditions for an
entire trip. Such a design anticipates that the driver will provide destination or navigation input, but is not expected to be available for control at any time during
the trip. This includes both occupied and unoccupied vehicles. By design, safe operation rests solely on the automated vehicle system.”
3
Blincoe et al. (2002), in a National Highway Traffic Safety Administration (NHTSA) report, estimate the costs of traffic collisions to be $230 billion in 2000.
KPMG (2015) estimates that 80% of all traffic collisions could be eliminated through widespread adoption of AVs.
4
A RAND Corporation (2016) report cites research suggesting AVs can increase fuel economy by 4–10% through more optimal acceleration and deceleration.
5
AVs are shown in traffic simulations to increase throughput by upwards of 500% (Fernandes and Nunes, 2012), and total vehicle miles travelled (VMT) by 35%
(Bierstedt et al., 2014). Shoup (2005) finds that nearly half of all congestion delays are caused by crashes, which AVs would reduce.
6 Santi et al. (2014) estimates that the New York City taxi fleet could be cut by 40% or more with ride sharing. Were an AV used in a commute to be re-purposed

to an autonomous taxi during the workday, both the number of vehicles and the number of parking spaces could be presumably reduced.

https://doi.org/10.1016/j.regsciurbeco.2019.103484
Received 17 November 2018; Received in revised form 9 August 2019; Accepted 8 October 2019
Available online 23 December 2019
0166-0462/Published by Elsevier B.V.
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

muter parking, freeing up land for other uses.7 According to estimates increased traffic congestion, causes energy consumption to increase on
in Meyer et al. (1965) and Shoup (2005), parking typically occupies a net basis. This result runs counter to engineering studies that argue
between 20% and 40% of all land in center-cities in the U.S., and large AVs will save energy. We show that this finding, in particular, high-
amounts in residential districts elsewhere in the city. With widespread lights the need to consider emerging technologies with a model that
adoption of AVs, cities and households may choose to reduce or elimi- allows for general equilibrium effects on the structure of a city.10
nate most CBD commuter parking as car sharing and automated parking Finally, the model predicts that house prices will unambiguously
becomes ubiquitous. fall for a city of a given size if parking is reclaimed for other uses. This
There has been intense research on the subject of AVs across a vari- occurs because the effective land supply is increased by the introduction
ety of disciplines, including engineering, urban planning, and urban of AVs: suburban land becomes more accessible to the center-city due
economics. As outlined above, researchers have made great strides to the reduction in commuting costs, and land within the city is made
in understanding the technical aspects of AV cars and their perfor- available through repurposing land that was previously used for park-
mance. However, it is much more difficult to apply partial-equilibrium ing. Due to the increase in land supply, additional interior space can
approaches to the general equilibrium of cities in which location of res- be added to the city at low cost, reducing housing prices on average,
idences and patterns of land use change in response to this new tech- and in nearly all locations due to the spatial equilibrium condition. In
nology. models where migration occurs and the city population increases, the
In order to address questions of a general equilibrium nature, we average house price per square foot still falls. This result from the model
adapt the standard monocentric city model of Alonso (1964), Mills suggests autonomous vehicles may help increase housing affordability,
(1967), and Muth (1969). This model focuses on firm and household an increasingly salient issue in U.S. cities.
location, commuting costs, and can incorporate labor mobility between Beyond these new results, we also offer findings that corroborate
cities, making it the ideal platform on which to explore new technolo- those in the literature, especially regarding the effects of AVs on traffic
gies that affect transportation costs and land use. Using this model, cali- congestion in a world occupied by endogenous actors. Duranton and
brated numerically to an average mid-size U.S. city as a basis, we intro- Turner’s (2011) hypothesis of a “fundamental law of road congestion”
duce autonomous vehicles by altering parameters and functions, includ- is reinforced by our work. The idea is that any short-run reductions in
ing marginal commuting costs and the land used for parking. Finally commuting times in large cities will simply induce in-migration until
we layer onto this framework autonomous taxis, which we implement the roads are nearly as congested as before. Rappaport (2016) also
through further reductions in parking needs and commuting costs. reaches this conclusion in his simulated response of high-productivity
We have identified three research questions that are not adequately cities to the introduction of AVs. Autonomous vehicles are perhaps no
addressed in the current literature, to which we are able to contribute different than adding additional highway capacity in this regard.11
using this model: 1) will AVs make the physical footprint of cities larger Overall, the cumulative effect of AVs and car-sharing technologies
or smaller; 2) will AVs save energy; and, 3) what are the implications of is dramatic. Household utility increases under all AV assumptions con-
AVs for housing affordability? We seek to answer these questions under sidered, as reflected directly in the utility calculation for cities with
different assumptions regarding the manner in which the technology is no migration and population increases for cities with migration. The
implemented. model predicts lower house prices, higher productivity, greater popula-
We have several main findings related to these questions as they per- tion density, and higher energy consumption and carbon emissions. AV
tain to mid-sized U.S. cities.8 The first is that the effect of autonomous cars have the potential to reshape cities, and as we demonstrate in the
vehicles on urban form is ambiguous for a city of a given size, and following sections, the standard urban model continues to be an ideal
depends heavily on the manner of AV implementation. The main rea- platform on which to consider emerging urban technologies.
son for this ambiguity is the tension between reductions in commuting
costs due to the direct effect of AVs, increases in commuting costs due 2. Background
to increased traffic congestion, and greater density due to land previ-
ously used for parking that is now used for additional commercial and The standard urban model (SUM) of Alonso (1964), Mills (1967),
residential purposes. If car-sharing technologies are not adopted and and Muth (1969) treats as endogenous housing production, housing
substantial center-city land is still used for parking, then the model consumption, commuting costs, and location choice within an urban
result is substantial sprawl. However, if parking land is reclaimed, the environment. Extensions within this framework are numerous, with
geographic footprint of the city shrinks.9 those particularly relevant to the study of autonomous vehicles includ-
The second main finding is that in virtually all parameterizations, ing the allocation of roads (Wheaton, 1998), space for parking (Brueck-
autonomous vehicles increase energy use by households. While the par- ner and Franco, 2017, 2018), and energy consumption (Larson et al.,
tial effect of AVs is to reduce energy consumption, the reallocation 2012) within the city. Its foundations and extensions have demon-
of the household budget towards housing and the numeraire good, strated that the SUM is an ideal basis for modeling and predicting the
combined with agglomeration externalities in goods production and general equilibrium effects of changes to transportation technologies.
Here, we present a brief, conceptual overview of the textbook SUM and
how it relates to AVs.
7
The distinction between commuter parking and other forms of parking is
intentional. This paper considers only parking and transportation costs borne
by commuters. While commuter trips are only a fraction of all trips made by
10
automobile, travel times for other purposes are likely correlated with commute A number of studies consider transportation and residential location in an
times. endogenous system. Studies which predict AVs will reduce energy consumption
8
Our model considers only a single city size calibration (about 1,000,000 in cities include Brown et al. (2014), Greenblatt and Saxena (2015), Fagnant
people) and is not designed to model small cities without congestion, or large, and Kockelman (2015), Wadud et al. (2016), and many others. Studies that
polycentric cities with alternative transportation technologies such as subways consider energy use in the standard urban model include Larson et al. (2012),
or light rail. Larson and Yezer (2015), Larson and Zhao (2017), with inclusion of model
9
Zakharenko (2016) models parking and autonomous vehicles in a mono- choice in Borck and Brueckner (2018). To our knowledge, no study exists which
centric framework, and finds the net effect to be sprawl for a city with a fixed considers the energy implications of AVs considering housing, commuting, and
population. While an important contribution, this paper presents the results of numeraire consumption together.
11
only a single parameterization. Rappaport (2016) calculates simulations of sys- It should be noted that we do not consider the intensive margin of vehicle
tems of cities, and predicts AVs to cause sprawl for a city with a fixed population use. As marginal transportation costs fall, the number of trips will also likely
and no changes in demand for parking. rise, further adding to congestion. In our model, we only consider commuting.

2
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

While many papers on the SUM present analytical solutions, this include the possibility of land repurposing from parking to other uses,
is not always possible. Because of the increasing layers of complex- and Zakharenko’s model does not include traffic congestion or residen-
ity needed to augment the base model in innovative ways, extensions tial district parking.
are often solved numerically, or “simulated” (e.g. Muth, 1975, Altmann These complexities flow through to the effects of AVs on energy con-
and DeSalvo, 1981, Bertaud and Brueckner, 2005, and Borck, 2016, to sumption. While energy consumption may fall due to effects of lower
name several). The strategy in this numerical simulation literature is congestion and faster commuting speeds, it may rise because sprawled
to calibrate the model to a real-world city or an average of cities, and cities include larger houses and longer commutes. Greater productivity
then alter the core parameters in the model to produce counterfactual, and output increases goods consumption, increasing energy consump-
general equilibrium predictions. tion, all else equal. Finally, and autonomous parking may save land for
In the classic rendition of the SUM, the city lies on a feature- parking, but it results in additional trips that consume energy. Some
less plane, with no geological or regulatory features that would of these rival forces are considered by Borck (2016), Larson and Zhao
inhibit development. Firms occupy the Central Business District (CBD), (2017), and others in the context of height limits, greenbelts, and tele-
and they exogenously demand identical workers to produce a single work. But as with urban form, the effect of AVs on energy consumption
exported good, which provide the impetus for households to locate and is theoretically ambiguous.
remain in the city. An agricultural hinterland determines the reserva- What is needed is a thorough treatment of both parking and traffic
tion land rent at the edge of the city. Between the CBD and the hinter- congestion, and an evaluation of the sensitivity of modeling decisions
land is the residential district, which houses the workers who commute on outcomes in order to evaluate the various effects on the city. The
to the CBD, always by car. There is typically no mixing of market-driven model in the next section is based on the SUM, but incorporates func-
land use at a particular location, but there are some exceptions, includ- tional forms that allow the general equilibrium effects of AVs to be
ing Wheaton (2004). modeled.
Housing producers and households receive a reservation profit and
level of utility, respectively, at every location inside the city. The house- 3. Baseline model structure
hold iso-utility condition, as given by “Muth’s (1969) equation,” gov-
erns the relation between house prices, housing consumption and trans- The purpose of the baseline model is to produce solutions that
portation costs. At locations where commuting costs are low, house- numerically mimic a present-day city, while providing a platform on
holds have greater willingness to pay for housing. The production func- which experimental scenarios related to AVs can be explored. The first
tion for housing involves structure and land inputs. Because areas near step is to assume functional forms for the behavior of the agents, pro-
the CBD are valued higher by households, and the land endowment duction functions, and other attributes of the city such as traffic conges-
is fixed at each location, density and house prices are higher near the tion and land use. Values and functions are then taken from the prior
CBD, and quantity consumed is lower. economics literature or engineering relationships, and then calibrated
The model is typically presented as a long-run solution that is fully with respect to a vector a city attributes. The following model describes
“putty-putty.” When parameters in the model are altered, all endoge- a baseline city in which each commuter drives a car to the CBD for
nous variables react without respect for initial conditions, including work. In later AV scenarios, attributes of the baseline city are modified,
road networks and urban density. The result of changing parameters giving new counterfactual cities.
and functions in the model is to construct a completely new and dis-
tinct city.12 3.1. The central business district
The key features in this model as it pertains to AVs are transporta-
tion costs and land use for parking. Because the city is an endoge-
The role of the CBD in the standard urban model is to exist as a
nous system, a change to either affects the character of the entire city.
location that demands labor in the city. Households then seek to locate
Changes to parking land use or transportation cost parameters have
near the CBD to economize on commuting costs. Though there are some
directionally predictable effects based on the existing SUM. However,
notable exceptions, little attention is typically paid to the treatment of
magnitudes and net effects when both are incorporated simultaneously the CBD in models that require numerical simulations. A singularity at
along with endogenous road congestion are a significantly greater chal-
the center of the city is often assumed; in other models where a singular-
lenge, as we now discuss.
ity would interfere with the model solution, such as in the case of con-
Changes to transportation costs have been examined in the litera-
gestible roads, the CBD typically occupies a fixed land area. Recently,
ture. It is well-established that Muth’s equation predicts lower trans-
there have been some attempts to model the production, land use, and
portation costs result in a city with greater land area and lower
agglomeration economies present in the CBD (Rappaport, 2016). Oth-
density—in a word, sprawl—and AVs are no different. But there is a ers have developed models with parking production and land use, either
countervailing effect of AVs on urban form that must be considered.
without (Brueckner and Franco, 2017) or with production (Brueckner
AVs will reduce demand for parking compared to a baseline city, thus
and Franco, 2018). Models including endogenous land use within the
resulting in a city with less land used for parking and more land for
city include Fujita and Ogawa (1982) and Lucas and Rossi-Hansberg
goods and/or housing production. AVs have two main channels that
(2002), but these do not include parking land.
might cause lower demand for commuter parking, both in the CBD and
It is necessary for us to model production, parking, and land use in
in the residential district. First, AVs will not need to be parked near the the CBD because of the parking land that may be reclaimed once AV
destination, with automation enabling parking in less costly areas of
cars are introduced. AV cars are expected to reduce the need for center-
the city. Second, car sharing may increase automobile utilization rates,
city parking, and in some solutions of the model, induce migration into
reducing the number of necessary parking spots, holding population
the city. The CBD therefore needs to have both a production function for
constant. This parking demand effect increases density.
goods that can incorporate land, labor, and agglomeration economies.
When combining these effects, we see that the effects of AVs on We then assume the CBD is a single, homogeneous area, in order to
urban form are complex and ambiguous. While both Rappaport (2016) avoid a singularity at the center. Finally, we assume the CBD is a fixed
and Zakharenko (2016) predict sprawl, Rappaport’s model does not
land area, starting at a point in the center of the city and extending to
radius kCBD . While the CBD area may expand or contract due to changes
in relative prices and bid-rent curves, this is a standard assumption in
12
It is common in the literature to discuss the results of counterfactual simu- the literature and one that greatly simplifies the model and allows us to
lations as “changes” to a city. To be clear, simulations here are not meant to be focus on residential location.
changes to an existing city.

3
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Land use in the CBD is fixed and divided between fixed shares of The fractions of land for roads 𝜃 road and other uses 𝜃 oth within the
parking, export goods production, and other uses, with values of 𝜃 indi- residential district are constant across k, as found empirically by Duran-
cating land use shares. The CBD extends from radius zero to radius kCBD . ton and Puga (2015), who show land used for transport, and the fraction
of developed residential land (𝜃 hous + 𝜃 park in our model), is roughly
𝜃park + 𝜃prod + 𝜃other = 1, 0 < k < kCBD (1) constant between 5 and 15 km from the center of Paris.
The land used for parking and housing is determined endogenously,
We now turn to the two activities that are modeled within the CBD,
parking and goods production. and varies by k. To derive these land shares, we begin with the follow-
ing expression, where c is the (constant) size of a parking space, 𝜖 is the
(constant) number of parking spaces per household, and h(k) the size of
3.1.1. Goods production
each housing unit at radius k.13
Because the city is circular and radial, land used for goods produc-
c𝜖
tion (and other areas in the city) can be expressed using the standard 𝜃park (k) = 𝜃hous (k) (6)
h (k )
formula for the area of a circle. The radius of the CBD is kCBD and the
share for production is 𝜃 prod , so the land endowment for production is The proportion of land used for housing can then be expressed as a
function of a single endogenous variable, the average dwelling size, and
Lq,CBD = 𝜃prod 𝜋 k2CBD (2) then solved as part of the final system of partial differential equations
in Section 3.3.
Following Glaeser and Gottlieb (2009), the export good in the city is
produced by a single perfectly competitive firm using a Cobb-Douglas 1 − (𝜃road + 𝜃oth )
𝜃hous (k) = (7)
production function F(N, L, K), where N is an endogenous number of 1 + c𝜖∕h(k)
workers and is equal to the city population, L is the CBD land exoge-
nously devoted to commercial production, and K is the endogenous
3.2.1. Housing and parking production
capital input. Localized agglomeration externalities in labor inputs are
Housing H at distance k from the CBD, is produced by combining
present, giving a technology function B(N) = BN𝜓 , where B′ > 0,
structure, S, and land, L, inputs under a constant returns to scale tech-
B′′ < 0, and 0 < 𝜓 < 1. The production function is:
nology according to a CES production function:
𝜈
F (N , L, K ) = BN 𝜓 N 𝜈1 Lq2,CBD K 1−𝜈1 −𝜈2 (3) [ ]1∕𝜌
H (k) = A 𝛼1 S(k)𝜌 + 𝛼2 L(k)𝜌 (8)
The firms’ first-order conditions result in a city-level labor demand
equation, Structure inputs are perfectly elastically supplied. Land is supplied from
( ) −1 𝜃 hous (k).14 Housing is rented at a price r(k).
𝜈1 +𝜈2
𝜓𝜈1 1 pK Parking has the same production function as housing, following
W = pK 𝜈2 (4)
1 − 𝜈1 − 𝜈2 N (1 − 𝜈1 − 𝜈2 )BN 1 L
𝜈 +𝜓
Brueckner and Franco (2017). The logic in using this production func-
q,CBD
tion for parking is that at high land prices, liveable space above
The key feature in this expression as it relates to the model concerns
the garage will be standard. As land prices fall, stand-alone garages
relations among parking land, commercial land, and labor demand. As
will become more common. Effectively, the producer has a cost-
land is reallocated from parking towards goods production, demand for
minimization problem, as 𝜖 parking spaces per unit of size c is fixed.
labor increases, resulting in some combination of higher wages and/or
The rental price for a parking space is assumed r(k) per square foot.
population.

3.2.2. Household budget constraint


3.1.2. Parking production
All households are identical and maximize a CES utility function
Parking is supplied at a fixed price pparkCBD in the CBD. We do not
consisting of two goods, rental housing h and a numeraire consumption
explicitly model the production of parking spaces, but we arrive at a
good y, subject to a budget constraint.
constant price under the following logic. Suppose parking is produced
[ ]1∕𝜂
using capital and land inputs that are perfectly substitutable. Given an U = 𝛽1 y𝜂 + 𝛽2 h𝜂 (9)
exogenous land input, the land endowment is used to supply a fixed
number of parking garages that may add or subtract floors under con- 𝛽 1 and 𝛽 2 are related to consumption shares between the two argu-
stant marginal cost in order to fully satisfy demand. While this may be ments, and 1∕(1 − 𝜂) represents the constant elasticity of substitu-
unrealistic in the case of small cities with CBD surface parking, or large tion between housing and the numeraire good. Household income, w, is
cities where increasing the capacity of garages becomes problematic, divided among a basket of goods and costs that vary based on residen-
for moderate city sizes, this assumption should approximate reality. tial location, and include the numeraire good, y(k), housing purchases,
There are numerous extensions possible within this CBD modeling r(k)h(k), parking, tp (k), and total transportation costs given by the prod-
framework. One of possible first-order concern is modal choice in com- uct of workers per household, 𝜖 , and transportation costs per worker,
muting (e.g. walking vs automobile transport). While modal choice T( k ) .
would add richness to the model, the solution would become more com-
w = y ( k ) + r ( k ) h ( k ) + tp ( k ) + 𝜖 T ( k ) (10)
plex along a number of dimensions. For instance, modal choice would
create differentials in the effects of private versus employer-paid park- The treatment of parking and transportation costs is one of the key
ing (Brueckner and Franco, 2018). Additionally, optimizing behavior features of the model. Parking, tp (k), includes both CBD and residential
would result in an equilibrium price and quantity of parking spaces parking. The price of parking in the CBD is pparkCBD and the price at
supplied, rather than the fixed values we assume here.

3.2. The residential district

The residential district occupies land between kCBD and k. While 13 To illustrate, let c = 300, the size of each parking space, 𝜖 = 1.25, the

kCBD is set exogenously, the city boundary, k, is endogenous. Land use number of commuters per household, and h(k) = 1500, the size of each house
in the residential district is divided among parking, roads, housing, and for some k. Then, the land share for parking is 0.25 the land share for housing,
other uses. or 20% (0.25/1.25) of a typical residential lot.
14
This model ignores the role of maintenance, rehabilitation and durability of
𝜃park (k) + 𝜃hous (k) + 𝜃road + 𝜃oth = 1, kCBD ≤ k ≤ k (5) structures in housing production.

4
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

home is r(k). In both places, households have 𝜖 commuters each requir- tem of equations described above can be solved and reduced to a system
ing c units of parking. with two simultaneous nonlinear differential equations with initial val-
ues. After a solution is obtained, the remaining gradients can be found
tp (k) = c𝜖(pparkCBD + r (k)) (11)
recursively. The two-equation system of equations includes differential
Transportation costs, T(k) are borne by each commuter 𝜖 , and commuting costs and the household density at radius k, with known ini-
include a fixed cost tf , which includes taxes, insurance, and obsoles- tial values at the CBD. This system is solved numerically using Matlab’s
cence; pecuniary costs t(k), which includes both the constant marginal ODE 45 solver with a differential of 0.1 mile and default convergence
cost per unit distance m and a variable marginal cost which depends tolerance.
upon traffic congestion; and a time-cost, tw (k). There are two trips each [ ] [ ]
dT (k) T (kCBD )
day, so each of these costs is multiplied by two. and (15)
dN (k) N (kCBD )
T ( k ) = tf + t ( k ) + tw ( k ) (12)
After solving this system, it is possible to derive house prices, hous-
where ing demand, land prices, structure/land ratios, and housing and parking
k
1 land shares as a function of commuting costs and housing unit density,
t (k) = 2(mk + pg d𝜅) following Altmann and DeSalvo (1981). Then it is possible to calculate
∫0 G(V (M (𝜅)))
energy consumption using methods found in the next section.
k
1 There are two different model closing conditions that can be
tw (k) = 2(𝜏 w d𝜅)
∫0 V (M (𝜅)) assumed depending on the purpose of the model. The first is the “closed
In t(k), the first term is the constant marginal cost of distance mul- city” model closing condition, which assumes a fixed population but
tiplied by distance to give mk. The constant marginal cost of distance allows utility to vary. This solution can be interpreted in several ways,
includes costs based on miles travelled, i.e. tires, oil changes, and other including a single city in autarky, the response of a nation-wide inno-
routine maintenance. The variable marginal cost of distance is the price vation that has an equal effect on all cities and has no effect on net
of gasoline, multiplied by the fuel spent going through each annulus of migration, or as a snapshot of a city within a given size class under
the commute, contained within the integral. This integral contains the different parameterizations.
inverse of the fuel economy of the vehicle G, a function of the speed The second set of closing conditions considers an “open city” model,
of the vehicle V. The speed is a function of traffic congestion, which which assumes a fixed utility, allowing population to vary. This is the
includes the argument M, the ratio of commuters to roads. typical assumption in a regional equilibrium or system of cities frame-
We follow Larson et al. (2012), who specify G(V(k)) in Equation (12) work, and is useful to analyze a city-specific technology shock. Because
as a 4th degree polynomial.15 our model is putty-putty and there is no durable housing, we caution
against this interpretation, as the introduction of AVs in a real-world
G(V (k)) = .822 + 1.833V (k) − .0486V (k)2 + .000651V (k)3 city will take decades to be fully implemented, and potentially cen-
turies to have complete effects on the urban spatial structure.16 Rather,
− .00000372V (k)4 (13)
these models are meant to represent a highly stylized world where AVs
This function gives about 14 miles per gallon at 10 miles per hour, are introduced and the city fully and immediately adjusts to its new
up to a maximum of 29 miles per gallon at 50 miles per hour, and equilibrium.
then falling to about 25 miles per gallon at 70 miles per hour. Under To implement the “closed city” (i.e. constant population) rendition
the assumptions that each worker in the city owns the same vehicle as of the model, first, the land price at the edge of the city must be equal to
the average vehicle in the U.S. fleet, this function gives an appropriate the agricultural land rent pL (k) = paL , and second, the number of workers
representation of commuting fuel used in the simulation. in the city must be equal to the number of jobs available 𝜖 N = E. The
For V(M(k)), we use the “Bureau of Public Roads” congestion func- simulation is initialized by assuming a house price at the edge of the
tion used by Muth (1975), where velocity is related to the ratio of cars CBD. If either of these equilibrium conditions is not met, the simulation
⃖⃖⃗(k) = ∫ 𝜖 2𝜋𝜃hous (k)kD(𝜅)d𝜅 , to the
travelling through the annulus, N
k is re-initialized with a higher or lower house price and solved again
k
⃖⃖⃗(k)∕R(k). The parameters a, b, and c are until subsequent iterations achieve an equilibrium solution. Note that
road area, R(k), or M (k) = N
in this city, equilibrium utility can rise or fall.
calibrated congestion parameters.
In the “open city” rendition, population adjusts in order to achieve a
1 reservation utility level. The solution method involves first establishing
V (k ) = (14)
a + bM (k)c a closed city solution, then if the utility is above (below) the reservation,
The term tw (k) in equation (12) represents the time-cost of commut- increasing (decreasing) the population and then re-solving the closed
ing. The expression 𝜏 w relates the time cost of commuting to the wage city, assuming full employment. In equilibrium, utility is equal to the
in a multiplicative fashion as a fraction 𝜏 , of the wage rate, w. The total reservation and the population may be different than that which was
time cost spent per commute is the integral of the time spent travelling initialized at the beginning of the solution algorithm.
through each annulus. The parameter 𝜏 is assumed to be 0.5 by Bertaud
and Brueckner (2005), and we follow this assumption in our baseline 3.4. Energy consumption
city. In the AV scenarios, we reduce this to 0.3 to reflect enhanced
leisure activities available while commuting, though we also consider After solving the model, energy consumption can be calculated
other values in an Appendix. recursively following Larson and Zhao (2017). Household energy con-
sumption at radius k, E(k), can be categorized into three main types:
3.3. Model solution electricity in the dwelling, ED (k), gasoline while commuting, EC (k),
and numeraire, which embodies all other forms of consumption, EN (k).
The solution method follows Muth (1975), Arnott and MacKinnon All energy is measured in terms of British thermal units (BTUs) and
(1977), Altmann and DeSalvo (1981), and McDonald (2009). The sys-

15 16
Larson et al. (2012) estimate this function using data gathered by West et Transportation networks are remarkably path-dependant. Historic structure
al. (1999) for an average vehicle in the U.S. fleet. designations also encourage path dependance.

5
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

includes energy consumed in production and transmission.17 assumed to have the same energy content as the numeraire good for
C D N purposes of computing energy consumption. The sum of these com-
E (k ) = E (k ) + E (k ) + E (k ) (16)
ponents is calculated as a residual from earnings net of gasoline and
Energy used in commuting by a household living in annulus k who electricity expenditures, w − pg EC (k)∕Eg − pe ED (k)∕Ee . The values
commutes to the CBD is given by pg EC (k)∕Eg and pe ED (k)∕Ee are the expenditures on gasoline and elec-
tricity, respectively, following Equations (17) and (18).
k
1 This numeraire expenditure value is multiplied by an (inverse)
EC (k) = Eg d𝜅 (17)
∫0 G(V (M (𝜅))) energy intensity parameter to calculate numeraire energy consumption.
where G(V(M)) is described in Equation (13), and Eg is the energy ( )
EN (k) = EN w − pg EC (k)∕Eg − pe ED (k)∕Ee (19)
embodied in a gallon of gasoline in BTUs, described as follows.18 The
base energy content of a gallon of 100% petroleum-based gasoline is The energy embodied in $1 of numeraire consumption is estimated
125,000 BTUs. According to the Federal Register (2000) published by to be EN = 7, 470 BTUs, which is the average energy intensity of
the Energy Information Administration, 1 gallon of gasoline requires an the U.S. economy (Energy Information Administration, 2011). Energy
additional 25,602 BTUs to be expended in the process of production and intensity is used for this measure because it implicitly includes all
distribution. Thus, Eg = 150, 602 BTUs of total energy are embodied energy in the raw materials, intermediate input production, final pro-
in final consumption. Eg is multiplied by the amount of fuel consumed duction, and transportation of the goods and services.
in gallons to arrive at the value for commuting energy consumption. Finally total energy consumption in the city is the integral of E(k)
Dwelling energy consumption is determined by three major factors: multiplied by the housing unit density D(k) over the city area.
the income of the household, the square feet of interior space, and
structure type. Larson et al. (2012) also estimate residential energy
k [ ]
E= 2𝜋 D(k) EC (k) + ED (k) + EN (k) dk (20)
demand parameters using the 2005 Residential Energy Consumption ∫kCBD
Survey (RECS). They find the partial elasticity of household energy
consumption with respect to interior space is 0.23 and the estimated
4. Baseline calibration and counterfactual cities
income elasticity is 0.07. Compared with the energy consumption in
single-family detached units, single-family attached dwellings consume
4.1. Baseline city
7% less energy and multi-family units consume 31% less energy. In the
simulation, the structure type, s(q(k)), is determined by the structure
We calibrate the model with respect to the characteristics of an aver-
to land ratio, q, defined as the ratio of housing square footage over lot
age present-day U.S. city. Cities are selected based on three filters which
size, denoted q = H∕L, which varies over k. The critical values of q for
are set to capture cities that have low regulation, few topographic inter-
each structure type are calibrated. The structure type is single-family
ruptions, and are of moderate size. The regulation filter is based on the
detached if q ∈ [0, 0.6], single-family attached if q ∈ (0.6, 0.7], 2–4
Wharton Residential Land Use Regulatory Index (Gyourko et al., 2008),
unit multifamily if q ∈ (0.7, 0.8] and 5 + unit multifamily when q is
and includes cities with an index value of less than zero. The topog-
above 0.8.
raphy filter includes cities with over 90% of nearby area topographi-
For simplicity, it is assumed all energy consumed in the dwelling is
cally available for development (Saiz, 2010). Finally, the city size filter
electricity.19 Each kilowatt hour of electricity consists of 3412 BTUs of
includes cities with between 300,000 and 700,000 housing units in the
energy. As with gasoline, there is also energy embodied in production
principal cities. The principal cities of Charlotte, Indianapolis, Kansas
and distribution. The total energy consumed in the production, distri-
City, and San Antonio pass these criteria, and our simulation is cali-
bution and final dwelling energy use can be calculated by dividing final
brated to a simple average of values in these cities.20
dwelling energy use by electricity efficiency parameter 0.303, giving Ee ,
Parameters are either taken from the existing literature or cali-
which is the product of the efficiency parameter for fossil fuel electric-
brated to give reasonable simulation outputs. These parameter values
ity production 0.328 and efficiency parameter for electricity transmis-
are shown in Table 1. Once calibrated, the housing and utility parame-
sion 0.924 (Federal Register, 2000). This gives dwelling electricity as a
ters are close to those found in Altmann and DeSalvo (1981). Altmann
function of dwelling size, with multiplicative shift factors that depend
and DeSalvo (1981) employs elasticities of substitution between struc-
on income, the energy price, and structure type.
ture and land inputs in the housing production function, and housing
[ ]
ED (k) = Ee exp 𝛾1 + 𝛾2 ln w + 𝛾3 ln pe + 𝛾4 ln h(k) + s(q(k))′ Γ (18) and the numeraire consumption good, of 0.75 in both equations. Land
shares to housing and roads are similar to Muth (1975), as well as the
In the expression above, the partial effects of covariates are represented speed parameters in the congestion function. Fixed and marginal com-
by 𝛾 s for income, energy prices and interior square feet, and the vector muting costs are from the American Automobile Association. The time
Γ for different structure types. cost of commuting for drivers is from Bertaud and Brueckner (2005),
Calculation of numeraire energy consumption begins with and set to 50% of the wage. The reservation agricultural rental price
numeraire good consumption. Added to this are expenditures on non- per acre per year is $500, which corresponds to $10,000 per acre at a
gasoline commuting costs and non-energy dwelling costs, which are 5% capitalization rate. Parking space per car is set to 300 square feet of
land area, and the annual parking fee per car in CBD is set to $1200.
17
The resulting baseline city is remarkably similar to the four-city
Different types of energy consumption may carry with them different types
composite. The major difference is the city we simulate is slightly
of externalities, and these are not considered. For instance, fossil fuels burned
denser and geographically smaller.21 Gradients within the baseline city
miles away from a city in a power plant may produce less particulate matter and
volatile organic compounds that harm households versus those burned within
the city in the form of gasoline. The simulation model in this paper does not
20
consider these nor other local environment or climate-related externalities. Suburbs are not included in the tabulation because the simulation is focused
18
While AV cars in the future may be electric and thus have different power- on areas nearest to the center of cities where gradients are closest to being
trains, these are not considered. The primary goal of the paper is to establish monotonic.
21
the effects of AVs, not electric vehicles. This is a well-known characteristic of cities simulated with a single income
19
Electricity-only consumption is associated with lower per-household energy group (Muth, 1975), and previous research has shown this characteristic—and
use compared to homes with natural gas, wood, or oil, according to the RECS. small perturbations in other simulation parameters—to have negligible effects
Therefore, estimates in this paper serve as the lower bound of energy consumed on differences between baseline and counterfactual cities (Larson and Zhao,
in the home. 2017).

6
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Table 1
Baseline simulation parameters.
Parameter Value Description Source
Central Business District
kC BD 1 Radius of the CBD (miles) Assumed
𝜃 park 0.25 fraction of CBD land used for parking Shoup (2005)
𝜃 prod 0.55 fraction of CBD land used for commercial production Assumed
𝜃 road 0.2 fraction of CBD land used for roads Muth (1975)
𝜈1 0.7 Labor share parameter in CBD production function Assumed
𝜈2 0.1 Land share parameter in CBD production function Assumed
pparkCBD $1200 Annual parking fee in CBD Various online sources
N 450,000 Households American Community Survey
W 49,867 Earnings per household per year American Community Survey
Residential District
𝜃 road 0.2 fraction of residential land devoted to roads Muth (1975)
𝜃 oth 0.45 fraction of residential land devoted to other use Muth (1975)
1∕(1 − 𝜌) 0.75 Elasticity of substitution in the housing production function Altmann and DeSalvo (1981)
𝛼1 1 Structure share parameter in housing production function Muth (1975), Altmann and DeSalvo (1981)
𝛼2 0.03 Land share parameter in housing production function Muth (1975), Altmann and DeSalvo (1981)
A 0.105 Housing production technology parameter Calibrated
1∕(1 − 𝜂) 0.75 Elasticity of substitution in the utility function Altmann and DeSalvo (1981)
𝛽1 1 Numeraire share parameter in utility function Numeraire
𝛽2 0.2863 Housing share parameter in utility function Calibrated
𝜏 0.5 Time-cost of commuting (fraction of wage) Bertaud and Brueckner (2005)
s 300 parking area per car (sq. ft) Authors’ measurement
paL 500 Reservation agricultural price per acre of land Bertaud and Brueckner (2005)
pg 3.5 Gasoline price per gallon Energy Information Administration
tf 2123 Fixed cost of commuting American Automobile Association
m 0.222 Dollars per mile of depreciation American Automobile Association
G(V(k)): constant 0.822 coefficient in gasoline-speed equation Estimated by Larson et al. (2012)
G(V(k)): 𝛽 V (k) 1.833 coefficient in gasoline-speed equation Estimated by Larson et al. (2012)
G(V(k)): 𝛽V (k)2 −0.048 coefficient in gasoline-speed equation Estimated by Larson et al. (2012)
G(V(k)): 𝛽V (k)3 0.000651 coefficient in gasoline-speed equation Estimated by Larson et al. (2012)
G(V(k)): 𝛽V (k)4 −3.7E-06 coefficient in gasoline-speed equation Estimated by Larson et al. (2012)
vlow 5 Minimum commuting speed Assumed
vhigh 45 Maximum commuting speed Assumed
c 1.75 Parameter in speed function Calibrated
Eg 150.6 Energy per gallon of gasoline (125 × 1/0.83 efficiency) Energy Information Administration
q0 0.8 5 + unit building Calibrated
q1 0.7 2–4 unit building Calibrated
q2 0.6 sf. attached Calibrated
Ee 0.303 Production and transmission efficiency for electricity Energy Information Administration
EN 7470 BTUs per dollar of GDP Energy Information Administration

Table 2
Simulation calibration.
City Charlotte Indianapolis Kansas City San Antonio Average Simulation Baseline
CBSA Code 16740 26900 28140 41700
Lot Size (acre) – Occupied Unitsa 0.36 0.31 0.25 0.20 0.28 0.16
Unit (square feet) – Occupied Unitsa 1694 1668 1655 1382 1599 1520
Area (sq. miles)b 444 409 515 505 468 290
Radius (assuming circle)b 11.9 11.4 12.8 12.7 12.2 9.6
Wharton Regulatory Index (Gyourko et al., 2008) −0.53 −0.74 −0.79 −0.21 −0.57 –
Unavailable Land (Saiz, 2010) 5% 1% 6% 3% 4% 0%
Median Incomeb 50,702 46,970 49,001 43,586 47,565 49,868
Total Occupied Unitsb 412,445 410,594 360,109 547,627 432,694 450,000
Time to workb 25.1 23.8 22.3 24.6 23.9 20.1
Fraction housed in 1 unit structuresb 71% 71% 70% 54% 66% 64%
Fraction housed in 2–4 unit structuresb 12% 12% 15% 14% 13% 17%
Fraction housed in 5 + unit structuresb 16% 17% 15% 32% 20% 19%
Energy consumed in dwelling, per capita (mmBTUs)c – – – – 49.8 89.6
CBD Parking Cost per Month 120 110 70 100 100 100
a
Source for actual values: AHS (2011).
b
Source for actual values: ACS (2010).
c
Source for actual values: RECS (2009) households with 100% electricity consumption.

7
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

are shown in Fig. 1, including downward sloping house price and struc- reallocate the center-city land endowment from parking to goods pro-
ture density, and upward sloping housing consumption and energy use duction, or 𝜃 park from 0.25 to 0 and 𝜃 prod from 0.55 to 0.8. Second,
gradients.22 because there is no center-city parking, the vehicle must make a return
trip to the place of residence to park during the day, then go back to the
4.2. Counterfactual cities CBD to pick up the worker for the commute home. This raises t(k), the
k 1
pecuniary costs of commuting, from t (k) = 2(mk + pg ∫0 G(V (M (𝜅)))
d𝜅)
To model the effects of autonomous vehicles, we alter various k 1
to t (k) = 4(mk + pg ∫0 G(V (M d𝜅). Finally, the total parking cost is
(𝜅)))
parameters and functions from the baseline model. The numerical solu-
reduced by eliminating CBD parking. The parking cost falls from
tions to these new models give counterfactual cities that can then be
tp (k) = c𝜖(pparkCBD + r(k)) in the baseline and City 1 to tp (k) = c𝜖 r(k),
compared to the baseline or other counterfactual cities.
which is simply the cost of parking in the residential district.
We consider three different parameterizations, each representing
different technological assumptions regarding implementation of AVs
within a city. It is useful to consider a range of models due to the range 4.2.3. City 3: autonomous taxis
of technological and public policy possibilities, including variation in One of the main drawbacks of eliminating CBD parking is the two
commuting costs, car ownership, parking reallocation, and means of “wasted” trips spent by the empty vehicle returning to the home parking
adjustment to a new urban equilibrium. In this section, we describe the spot during the day and then back to the CBD to pick up the commuter
each of these counterfactual cities in turn, before moving onto results for the trip home. This problem is exacerbated the further the parking
in the following sections. spot is from the CBD. In this city, we wish to implement a technological
alternative that uses a fleet of autonomous taxis rather than owner-
occupied automobiles.
4.2.1. City 1: basic autonomous vehicles
In this city, we introduce a perfectly competitive firm that owns and
In our first counterfactual city, we model the introduction of AVs in
operates a fleet of fully-autonomous vehicles. Commuters pay a hailing
the most direct manner possible; we reduce the time-cost of commuting
fee and then marginal costs for units of time and distance, as is standard
and make no other changes. There are no coincident technologies such
for current taxi services, including traditional taxi cabs, Uber, and Lyft.
as car sharing or electric vehicles, and there is no change in demand for
Commuters still dislike commuting, with a time cost fraction 𝜏 = 0.3
parking anywhere in the city. This is implemented by reducing the time
of the wage rate.
cost fraction, 𝜏 , in Equation (12). This is originally set at 0.5, or 50% of
This scenario is operationalized by making the following additional
the wage rate for the baseline city. For the AV city, we set 𝜏 = 0.3 to
changes to the model. First, there is no land used for parking any-
reflect a lower disamenity of time spent commuting. This new value for
where in the city, including in both the residential and central business
𝜏 is speculative, so we also consider alternative parameterizations of 𝜏 ,
districts.25 Second, the pecuniary fixed and marginal commuting cost
ranging from 0.1 to 0.4, shown in the Appendix.
functions are both substantially different, and are instead based on a
It is known a reduction in the marginal cost of distance will flatten
taxi-fare model. Accordingly we re-specify tf = 𝜑0 , tp (k) = 0, and
the house price, land price, and density gradients, and increase housing k
consumption, and energy consumption, all else equal (see Coulson and t (k) = 𝜑1 k + 𝜑2 ∫0 V (M1(𝜅)) d𝜅 , where 𝜑0 is the fixed hailing fee, 𝜑1 is
Engle, 1987, Mieszkowski and Mills, 1993, Brueckner et al., 2001, and the cost per mile, 𝜑2 is the cost per unit of time.
Larson and Zhao, 2017, for instance).23 The purpose of this model is The values for 𝜑0 , 𝜑1 , and 𝜑2 are based on current Uber rates
to capture this straightforward transportation cost effect before moving for Washington, DC (as of March 2018) and from a report by UBS
on to more complex models. (2017). Uber rates are chosen to give a reasonable parameterization
under a business model where their stated goal is to eventually shift
4.2.2. City 2: No CBD commuter parking to autonomous taxis.26 The UBS report estimates that current taxi rates
will fall by about 54% for autonomous taxis due to market efficiencies,
In the next counterfactual city, we introduce commuters who have
labor savings, and other cost reductions due to the widespread adop-
autonomous vehicles that return to home to park during the work day.
tion. This results in 𝜑0 = $1.36 per trip fixed costs, 𝜑1 = $0.50 per
This involves four one-way trips per day for the vehicle, with people
mile, and 𝜑2 = $0.078 per minute.27
riding in the vehicles for two of them. Because commuter parking no
This AV taxi cost structure steepens the bid-rent curve by shifting the
longer occurs in the CBD, this land is instead used by goods producers.24
cost share from fixed to marginal costs of distance and time. To illus-
To implement this technological alternative in terms of parameters
trate, suppose an 8 mile trip that takes 25 min at a constant speed (19.2
and functions in the model, we set 𝜏 = 0.3 as in the basic AV city,
mph). The AV taxi trip cost is $7.31, with $1.36 fixed costs and $5.95
but in addition, we make three other changes to the model. First, we
variable costs. In contrast, for owner-occupied vehicles (both owner and
autonomously driven), the total trip cost is $7.28, with an average fixed
cost of $4.25 and variable cost of $3.04.28 In this case, there is little dif-
22
For a complete set of gradients for the baseline city, see Appendix Figs. A.1
ference in the total pecuniary commuting costs. However, because the
and A.2. Dwelling and total energy use gradients have discrete jumps at struc- cost share changes from fixed to variable commuting costs, the house
ture type transitions from multifamily, to single-family attached, to single- price gradient will flatten less than in City 1.
family detached.
23
Analytic predictions of a fall in transportation costs in the standard urban
model are available in Brueckner (1987). These predictions encompass the 25
Parking for the autonomous vehicles is not specified in the model, but in
effects of a simple model with a reduction in commuting costs for any reason, practice an autonomous fleet of taxis could be housed on streets and/or in
including AVs. Unfortunately, analytic solutions of models with traffic conges- suburban parking lots outside of the city when not in use.
tion layered onto other model complexities, such as we do here, are unavailable 26
Uber has invested substantial resources experimenting with autonomous
(this also includes models of endogenous land use such as Fujita and Ogawa, taxis, see https://www.wired.com/story/bet-uber-bet-self-driving/.
27
1982). Congestion is a first-order concern in our model due to the additional These autonomous taxi rates are based on current Uber rates of $2.95 per
traffic AVs may cause in different regions of the city. We therefore leave ana- pickup, $1.08 per mile, and $0.17 per minute, with each parameter multiplied
lytic solutions of the effects of simple AVs under linear transportation costs to by 0.46. To explore the robustness of the model solution to these parameters,
these prior works, and proceed with numerical solution methods in this paper. we also consider other values in the Appendix.
24 28
We have also considered reallocation to residential uses. However, we have This assumes all fixed costs are divided by the total number of trips, and the
found goods production to dominate residential housing production in the entire 25 min trip is taken at a constant speed, or 19.2 miles per hour, which
model. gives 22.2 miles per gallon.

8
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Fig. 1. The baseline model and effects of autonomous vehicles on urban gradients.

5. Autonomous vehicles and sprawl lation is designed to give the maximum sprawl for an owner-occupied
autonomous vehicle because it includes only the reduced transportation
The first question posed in the introduction relates to sprawl. The costs.29
major unresolved issue is whether autonomous vehicles will cause cities The first thing we wish to examine is the change to the bid-rent
to become more dense because of the land saved from reduced park- curve for housing with the implementation of AVs, shown in Fig. 2. In
ing demand, or sprawl because of the lower transportation costs. In the closed city model (with a constant population), the bid-rent curve
attempting to answer this question, we consider six models; the three rotates to maintain the iso-utility condition. This reflects a lower will-
counterfactual cities, with both open and closed city model closing con- ingness to pay in areas near the CBD and a greater willingness to pay
ditions. in areas far from the CBD. Accordingly, the city sprawls. In the open
city model, the same rotation occurs, but to reach an inter-city equi-
librium, in-migration occurs, shifting the bid-rent curve up. This shift
5.1. City 1 results; basic autonomous vehicles occurs because there are greater levels of congestion at every radius.
Table 3 column 2 shows tabulations of results for the closed city,
We first compare the basic AV model solutions to the baseline model
solution. This is the simplest counterfactual scenario we consider, and
involves a 100% adoption of AVs for commuters, but no changes to
29
any other aspect of the model, including parking behavior. This simu- This is analogous to the main AV simulation in Rappaport (2016).

9
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Fig. 2. Direct effect of autonomous vehicles on bid-rent curves.

and Fig. 1 shows some important gradients.30 The city sprawls as pre- AV simulation and reflects the greater utility in the closed city. This
dicted, with the city radius expanding from 9.6 to 10.7 miles. The city indicates households are better off under an AV scenario where CBD
has higher welfare, with household utility rising from 2549 units to parking is eliminated, even though cars must now travel home during
2626 units. When the model is solved using the open city model clos- the day to park. Because of this extra marginal cost of distance, the resi-
ing conditions, migration occurs because the city utility is higher in dential bid-rent curve steepens substantially, and the city radius shrinks
the closed city than in the baseline. In this case, as shown in Table 4 from 9.6 miles in the baseline to 9.4 miles despite the much larger pop-
column 2, the population, measured by the total number of occupied ulation. Under this model, both the open and closed city results suggest
housing units, increases from 450,000 in the baseline to about 565,000, owner occupied AVs make the city smaller.
or about 26%. The city radius also expands a further 0.6 miles to 11.3
miles. Clearly, AVs in this most basic scenario cause sprawl.31 5.3. City 3 results; autonomous taxis

5.2. City 2 results; No CBD commuter parking Finally, we examine commuter vehicles that are autonomous taxis
instead of owner-occupied vehicles. This frees both CBD land for addi-
Next, we consider the case where the AV cars are still owner- tional production, and residential land for additional housing. In the
occupied, but center-city commuter parking is no longer available. closed city, the city radius falls from 9.6 miles in the baseline to 9.2
In this simulation, CBD land used in previous simulations for park- miles, as seen in Table 3 column 4. While the bid-rent curve flattens in
ing is now allocated to goods production. Commuter vehicles travel Fig. 1, more land is available for housing so the city radius still shrinks.
autonomously (and empty) back to the housing unit during the day to In the open city, as shown in Table 4 column 4, the population is higher
park, then return to the CBD in the evening to pick up the worker for than in any previous open-city model, with 1.16 million households, an
the return trip home. increase of about 150%. Even with this dramatic increase in the popu-
In the closed city rendition of the model, shown in Table 3 column lation, the city only sprawls an additional 0.3 miles relative to the base-
3, the city radius falls from 9.6 miles in the baseline to 8.8 miles. This line. This is reflected by a large increase in density, from about 1600
occurs because the bid-rent curve steepens substantially, as depicted in households per square mile in the baseline to 3800 in the open-city sce-
Fig. 1. In the open city model, shown in Table 4 column 3, the number nario. Accordingly, while about 19% of all households occupy units in
of households increases by 65%–743,000, which is larger than the Basic 5 + unit structures, this changes to about 66% in the AV scenario with
autonomous taxis in an open city.
To summarize the results: 1) AVs increase welfare in all models con-
30
For a complete set of gradients for both open and closed cities, see
sidered, as reflected directly by utility increases for closed city solu-
Appendix Figs. A.1–A.4. tions, and population increases for open cities; 2) when AVs are imple-
31
Other gradients behave predictably, including a rotation in the land and mented alone with no changes to parking behavior, additional sprawl is
house price gradients, increase in commuting times, and a decrease in structure the result; 3) when technology permits parking land reallocation from
density per unit of built-up land area. parking to goods production, sprawl decreases slightly; 4) the introduc-

10
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Table 3
Baseline and closed city model solutions.
Model [1] [2] [3] [4]
City Name Baseline Basic AV (1) No CBD Parking (2) Autonomous Taxis (3)
Assumptions
Day Parking Land CBD CBD Home None
Night Parking Land Home Home Home None
Time-Cost of Commuting (x Wage) 0.50 0.30 0.30 0.30
Urban Form
Total Occupied Units 450,000 450,000 450,000 450,000
Lot Size (acres, detached avg) 0.15 0.17 0.16 0.21
Unit Size (Sq. ft. avg.) 1520 1578 1656 2059
Residential Struct./Land ratio (CBD) 1.40 1.06 1.73 1.73
Residential Density (hh per sq. mile) 1571 1262 1874 1713
Time to work (avg) 20.1 21.0 19.4 19.6
Fraction housed in 1 unit structures 65% 78% 48% 49%
Fraction housed in 2–4 unit structures 17% 15% 16% 15%
Fraction housed in 5 + unit structures 19% 7% 37% 35%

House Prices
House Price per Sq. Ft. (CBD) 15.17 14.72 15.21 12.33
House Price per Sq. Ft. (avg) 13.88 13.66 13.90 11.33

Land Use
City Radius (assuming circle) 9.6 10.7 8.8 9.2
City Area (sq. miles) 289.5 359.7 243.3 265.9
Area (Production) 1.7 1.7 2.5 2.5
Area (Residential) 80.6 101.0 68.7 92.0
Area (CBD Parking) 0.8 0.8 0.0 0.0
Area (Residential Parking) 19.6 23.8 15.3 0.0

Expenditure Accounting
Income per household $ 49,868 $ 49,868 $ 52,259 $ 52,259
Numeraire Expenditure $ 23,472 $ 22,898 $ 24,312 $ 25,195
Housing Services Expenditure $ 21,098 $ 21,542 $ 23,021 $ 23,331
Commuting Expenditure $ 5298 $ 5427 $ 4926 $ 3734

Energy Consumption per Household (million BTUs)


Total 524.2 530.0 563.7 543.5
Commuting 24.7 31.9 72.5 56.0
Dwelling 142.6 144.8 128.5 126.1
Numeraire 356.9 353.3 362.7 361.5

Welfare Accounting
Utility 2549 2626 2777 3194

tion of autonomous taxi technology has an ambiguous sprawl effect, land that is repurposed from parking to production increases the wage
with the open city showing an increase due to in-migration and the rate through an increase in the land endowment and the marginal prod-
closed city showing a decrease.32 uct of labor. Second, the production function has increasing returns in
These results imply the effects of AVs on the urban form of the city labor inputs, so in the open-city renditions of the model, population and
depend on the manner in which AV technology is implemented. The incomes are endogenous and positively related.33
area of the land used by the city depends on the ability of the city
to adjust land use patterns to changes in demand for parking, and the 6.1. City 1 results; basic autonomous vehicles
extent to which cities grow in response to the introduction AV cars.
This includes cities for a given size class, as given by the closed-city Expenditure substitution is best illustrated by closed-city version of
models, and for cities with in-migration in response to AVs, as given by the Basic AV model (Table 3, column 2). In this model, income is the
the open-city models. same as the baseline model because population and the land endow-
ment are identical. Effects on energy consumption are therefore due to
6. Autonomous vehicles and energy consumption substitution prompted by changes in relative prices and a small income
effect from a change in the average price level for the expenditure bas-
The second question we wish to consider, using the solutions in ket. In this model, household energy consumed while commuting rises
Tables 3 and 4, is the issue of energy consumption and whether from about 25 million BTUs (mmBTU) annually to about 27 million
autonomous vehicles save energy. One factor influencing the result BTUs. This occurs due to the suburbanization of households and the
is how the introduction of AVs affects the relative prices of housing, associated longer average commute, which rises from about 20 min to
commuting, and the numeraire good, and how these changing relative
prices affect energy consumption vis a vis expenditures. Additionally,
there are two channels in the model that increase incomes. First, CBD 33
It should be noted from the outset that we only consider energy consumed
by households through their dwellings, commuting, and numeraire consump-
tion. However, this still provides a relatively complete picture of energy con-
32
These results are robust to alternative parameterizations of 𝜏 found in the sumption through energy intensity multipliers described in Section 3.4, which
Appendix, including the ambiguity of the sprawl effect under the autonomous account for all energy embodied in all stages of production, transportation, and
taxi scenario. consumption.

11
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Table 4
Baseline and open city model solutions.
Model [1] [2] [3] [4]
City Name Baseline Basic AV (1) No CBD Parking (2) Autonomous Taxis (3)
Assumptions
Day Parking Land CBD CBD Home None
Night Parking Land Home Home Home None
Time-Cost of Commuting (x Wage) 0.50 0.30 0.30 0.30
Urban Form
Total Occupied Units 450,000 564,855 743,483 1,162,530
Lot Size (acres, detached avg) 0.15 0.16 0.14 0.16
Unit Size (Sq. ft. avg.) 1520 1524 1496 1612
Residential Struct./Land ratio (CBD) 1.40 1.30 2.80 5.86
Residential Density (hh per sq. mile) 1571 1419 2709 3815
Time to work (avg) 20.1 28.1 38.9 77.3
Fraction housed in 1 unit structures 65% 70% 34% 24%
Fraction housed in 2–4 unit structures 17% 17% 12% 10%
Fraction housed in 5 + unit structures 19% 13% 54% 66%

House Prices
House Price per Sq. Ft. (CBD) 15.17 15.08 16.35 14.24
House Price per Sq. Ft. (avg) 13.88 13.84 14.50 11.90

Land Use
City Radius (assuming circle) 9.6 11.3 9.4 9.9
City Area (sq. miles) 289.5 401.1 277.6 307.9
Area (Production) 1.7 1.7 2.5 2.5
Area (Residential) 80.6 112.1 77.1 106.7
Area (CBD Parking) 0.8 0.8 0.0 0.0
Area (Residential Parking) 19.6 27.3 19.0 0.0

Expenditure Accounting
Income per household $ 49,868 $ 49,585 $ 51,607 $ 51,034
Numeraire Expenditure $ 23,472 $ 22,926 $ 24,318 $ 24,540
Housing Services Expenditure $ 21,098 $ 21,085 $ 21,701 $ 19,179
Commuting Expenditure $ 5298 $ 5574 $ 5589 $ 7315

Energy Consumption per Household (million BTUs)


Total 524.2 531.4 556.2 544.5
Commuting 24.7 26.7 46.3 23.5
Dwelling 142.6 148.6 138.5 146.3
Numeraire 356.9 356.0 371.3 374.6

Welfare Accounting
Utility 2549 2549 2548 2549

21 min per trip. Due to greater suburbanization, congestion at each the introduction of AVs, or about 1.3%.34
radius increases, reducing the speed and fuel economy of each vehicle The effect on energy consumption in the open city version of the
as it passes through each successive annulus on the way to the CBD. model (Table 4, column 2) is slightly lower than in the closed city
The greater suburbanization has an additional effect on housing con- model, though there are a variety of mostly offsetting factors that
sumption. Because the average rental price per unit of housing in the must be considered. In addition to the substitution and income effects
city falls, from about $13.9 per year in the baseline model to $13.7 increasing energy consumption, there is also additional population due
in the Basic AV model, average housing consumption increases from to the regional equilibrium condition. This increases productivity due
about 1520 square feet per household to 1580 per household. Also, to agglomeration economies. However, in order to achieve the regional
the average structure type changes, from 64% single-family detached utility, population rises to the point where the wage rate is actually
in the baseline to 78% in the AV scenario. Due to both of these factors, below the baseline—even with agglomeration—due to the benefits of
dwelling energy consumption rises from 143 mmBTUs to 149 mmB- AVs. Finally, due to the greater population and higher transportation
TUs, nearly 3 times the direct effect from commuting alone (2 = 27-25 and housing costs at each radius, housing consumption falls and struc-
mmBTUs). ture density rises. Thus, there are both positive and negative effects
Because house prices fall relative to the numeraire good, the substi- of AVs on household energy consumption when population is endoge-
tution effect causes a reduction in numeraire good consumption and nous. Overall, commuting energy consumption rises by more than in
its associated energy consumption. The income effect slightly coun- the closed-city model, housing energy consumption rises by less, and
ters this, but on a net basis, the effect of the fall in house prices is a numeraire energy consumption falls by more. The net effect is a slightly
reduction in numeraire goods consumption, from 357 mmBTUs to 356 lower increase in household energy consumption, from 524 mmBTU in
mmBTUs. Overall, when considering all three consumption categories, the baseline, to 530 mmBTU in the open city vs 531 mmBTU in the
annual household energy consumption rises by about 7 mmBTUs after closed city.

34
It should be noted that this AV effect is purely a commuting effect. The
reduction in marginal transportation effects may change non-commuting travel
behavior as well, including shopping, school, etc. Thus, our model gives, per-
haps, a lower bound because we do not consider other means of adjustment.

12
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

6.2. City 2 results; No CBD commuter parking population of the open city, energy consumed commuting increases
to 56 mmBTU. In autonomous taxis, commuters are willing to bear
The city with no CBD commuter parking also has increased house- extremely long commutes in the open city, with the average commute
hold energy consumption for both closed-city and open-city versions stretching to 77 min per one-way trip, relative to 20 min in the baseline.
(Table 3, column 3 and Table 4, column 3, respectively). After driving As the city area increases only modestly, this increase in commuting
the commuter to the CBD, the vehicle returns to the place of residence time can be attributed to additional traffic congestion.
to park for the day. Then before the end of the work day, returns to the Numeraire energy consumption rises substantially in both scenarios,
CBD to pick up the worker for the return trip. These additional trips from 356 mmBTU in the baseline to 575 mmBTUs (closed city) and 362
add substantial energy consumption and steepen the bid-rent curve. mmBTUs (open city), respectively. This is due to an effect of rising
Additionally, the land that used to be CBD parking in the baseline incomes that is partially offset by an increase in the relative price of
model is now allocated to goods production, increasing earnings from the numeraire good. Overall, total energy consumption per household
about $49,900 per year to about $52,300 per year (closed city) and increases from 524 mmBTUs in the baseline to 545 in the closed city
$51,600 (open city). This additional income is spent on expenditures and 544 mmBTU in the open city, for increase in energy consumption
which embody energy. of 3.9% and 3.7%, respectively.
This model shows reduced energy consumption in dwellings. The Across all simulations, energy consumption increases. This is due
price of housing increases from $13.88 per square foot in the baseline to greater labor productivity, and depending on the model, greater
to $13.90 (closed city) and $14.50 (open city). When combined with the housing consumption, a larger number of trips, and/or more congested
additional income, consumption increases from about 1520 square feet roads. While autonomous vehicles may reduce energy consumption in
in the baseline to 1660 in the closed city, but falls to 1500 in the open some sectors (commuting, housing, or numeraire consumption), when
city. These dwellings exist in more efficient multi-unit structures, caus- summed, the model result is an increase in total energy consumption in
ing a drop in dwelling energy consumption overall, with 143 mmBTU every case.
in the baseline falling to 139 mmBTU (closed city) and 129 mmBTU
(open city). 7. Autonomous vehicles and housing affordability
Offsetting this reduction, the model result is an increase in energy
consumption in both commuting and numeraire good consumption. The final issue we address with our series of models is the effect of
While relative prices change slightly, the additional income creates autonomous vehicles on housing affordability. By reducing commuting
substantial additional demand for the numeraire good, with embod- costs, AVs increase CBD accessibility; and by reducing the land nec-
ied energy consumption rising from 357 mmBTU in the baseline to 371 essary for parking, AVs increase the land available for housing con-
mmBTU (closed city) and 363 mmBTU (open city). Due to the addi- struction in currently built-up areas. Both effects increase housing sup-
tional vehicle trips to park at the place of residence, commuting energy ply, making housing more affordable in partial equilibrium. However,
consumption rises from 25 mmBTU to 46 mmBTU (closed city) and when moving to a general equilibrium analysis, the models we have
72.5 mmBTU (open city). In the closed city, the effect of doubling the developed suggest that AVs increase housing affordability in many but
number of trips is less than double of that on energy because of the not all cases. In this section, the key variable of interest is the average
reduction in sprawl, whereas in the open city, the effect is more than house price per square foot across the city. Lower values relative to the
twice the baseline because of the additional population and congestion. baseline are interpreted as an increase in housing affordability, even if
Overall, in this model, energy consumption summed over the three cat- the average unit size rises making the total expenditure on each housing
egories increases, from 524 mmBTU in the baseline to 556 mmBTU unit higher.
(closed city) and 564 mmBTU (open city), for a 6.1% and 7.6% increase,
respectively.
7.1. City 1 results; basic autonomous vehicles

6.3. City 3 results; autonomous taxis As with the other sections, the model that gives the pure trans-
portation cost effect is the Basic AV simulation under the closed-city
The city with autonomous taxis has energy consumption patterns model closing condition. This model holds population, income, and
that fall in between the Basic AV model and the scenario with no CBD CBD parking land use constant, and therefore focuses exclusively on
commuter parking. For autonomous taxis relative to owner occupied the accessibility effects of AVs on affordability. Because transportation
vehicles, marginal pecuniary commuting costs are higher and fixed costs fall, the price falls from an average price of $13.9 in the base-
commuting costs are lower. This rotates the bid-rent curve relative to line model to $13.7 in the Basic AV model. For a city that introduces
the previous models. Because parking land in the residential district is AVs and then grows in a response, we turn to the open-city rendition
no longer required for commuters, the land available for housing con- of the model. In this city, the population increases, but housing still
struction increases substantially. This effect reduces the price of land, becomes slightly more affordable, with average house prices falling
which has a host of effects on patterns of housing consumption in the from $13.9 (baseline) to $13.8. Overall, AVs and the associated trans-
closed city. A lower land price reduces the price of housing, which both portation cost reduction increases housing affordability, even when it
reduces the amount of structure produced on each unit of land and causes an increase to the city size.
increases housing consumption. But because there is more land avail-
able, population per annulus actually rises, implying a larger number
7.2. City 2 results; No CBD commuter parking
of lower-density homes per square mile. Combined with the rotation in
the bid-rent curve, the autonomous taxi model has dramatically higher
With owner-occupied vehicles, when CBD parking land is reallo-
population density in both cities, increasing from 1571 housing units
cated to goods production and the vehicle makes the additional two
per square mile in baseline model to 1713 (closed city) and 3814 (open
trips each day, the effect on housing affordability is ambiguous. In the
city). The housing unit size effect dominates in the closed city, increas-
closed city, the house prices rise slightly ($0.02 per square foot) but
ing energy consumption from 143 mmBTUs in the baseline to 146 mmB-
incomes rise by 4.8%, from about $49,900 to $52,300. Because incomes
TUs, while the density effect dominates in the open city, decreasing
rise faster than housing costs, we conclude that housing is more afford-
energy consumption to 126 mmBTUs.
able in this city than in the baseline. On the other hand, in the open city,
Commuting energy consumption falls in the closed city model to
the population is much larger, with lower incomes and higher house
24 mmBTU from the baseline value of 25 mmBTU because the city
prices. In this case, housing is less affordable. We therefore conclude
becomes more compact with a constant population. With the higher

13
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

that the effect of this scenario on housing affordability is ambiguous, basis, it always rises. This is the case for cities within a size class (i.e.
a finding reinforced by the alternative parameterizations shown in the holding population constant) or in a regional equilibrium where the city
Appendix. population increases in response to the introduction of AV cars.
Finally, housing becomes more affordable in most of the model
cities. The mechanism for the increase in housing affordability is an
7.3. City 3 results; autonomous taxis
increase in the effective land supply brought about by 1) greater acces-
sibility of land far from the CBD due to falling transportation costs, and
Autonomous taxis dramatically increase housing affordability.
2) more available land within the city due to lower demand for park-
Because AV cars both make accessible more land area proximate to
ing. Lower commuting and housing costs give rise to substantial utility
the city and increase the land available to housing in the residential
increases in all AV scenarios considered.
district, the land price falls. Because the land price falls, so too does the
The approach we have taken in this paper is different than prior
house price. In the closed city, the average house price falls from $13.9
work in several ways. First, engineering studies do not tend to incorpo-
per square foot to $11.3 in the closed city model and $11.9 in the open
rate the dynamic response of households and firms to the fall in trans-
city model. It is remarkable that the open city has 250% the popula-
portation costs and reduction in parking demand, and thus overstate the
tion of the baseline city, yet is physically smaller with lower housing
effects of AVs on reductions to energy consumption. Second, empirical
costs.
studies are hampered by the fact that data do not exist on the adop-
Overall, in 5 of the 6 models considered, housing becomes more
tion of AVs—AVs are, after all, a nascent technology. Finally, previous
affordable, while in 1 of the 6, it becomes more expensive. Generally,
models of the city tend to present a single view of how a city responds
house prices fall because more land area in the city is accessible to
to the introduction of AVs. In the present paper, we consider a variety
development, either through a reduction in commuting costs or because
of dimensions across which the city may or may not adjust, including
land for parking is reclaimed for other uses. In the model where house
changes to population, reduction to demand for center-city and resi-
prices rise, the cause is the dramatic influx of people into the city. In
dential parking, and whether commuter vehicles are owner-occupied.
all models, however, the introduction of AVs results in greater utility,
In sum, our paper is the first to consider these combined margins for
as demonstrated through the higher utility level in the closed cities and
adjustment in a general equilibrium framework.
the higher population in the open cities.
Despite the progress made here, there are several ways in which the
model could be improved in future research. We only consider auto-
8. Conclusion mobile commuters. In reality, workers have a rich variety of modal
choice available, including walking, subway, light rail, buses, or tele-
In this paper, we have presented a rendition of the canonical mono- work (Borck and Brueckner, 2018). The open-city models assume a
centricy city model and solved this model under assumptions that emu- fixed regional utility. In reality, other cities’ utility will also rise were
late the introduction of autonomous vehicles (AVs). Long-run solutions AVs to be introduced in those cities as well. Even in cities that do
to this model are then used to address three salient questions: will not face congestion, there will still be benefits, even if they are small.
AVs cause cities to sprawl or become more compact; will AVs increase Modeling a true system of cities, similar to Rappaport (2016) would
or decrease household energy consumption; and, will AVs make hous- be an excellent next step. Finally, other coincident technologies could
ing more expensive or more affordable? Three alternative technologies be considered, including electric cars, and there may be interactions
under which AVs may operate are considered. with regulations such as historic designations (density cannot increase
AV cars have an uncertain effect on the urban form of the city. The optimally) or greenbelts (density is forced). As this research shows, the
direct effect of AVs clearly increases sprawl, and this effect dominates monocentric city model is a flexible platform on which to examine such
in some models. But the indirect effects of reduced land for parking act issues.
as forces making the city more compact. The net result is dependent on
the particular implementation of AVs in the city, whether parking land Acknowledgments
can be reclaimed for other uses, and if AVs induce in-migration.
Predictions involving energy consumption are much clearer. In all The authors would like to thank seminar participants at the AREUEA
model cities, energy consumption is higher compared to the current national meetings and the University of Louisville, Tony Yezer, Jan
baseline city. Energy consumption embodied in commuting, housing Brueckner, Jordan Rappaport, our editor Jeff Lin, and anonymous ref-
consumption, or numerare consumption may rise or fall, but on a net erees for valuable advice and feedback.

14
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Appendix A.

A.1. Appendix figures

Fig. A.1 Autonomous Vehicles and Urban Gradients (Closed City).

15
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Fig. A.2 Autonomous Vehicles and Energy Gradients (Closed City).

16
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Fig. A.3 Autonomous Vehicles and Urban Gradients (Open City).

17
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Fig. A.4 Autonomous Vehicles and Energy Gradients (Open City).

18
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

A.2. Appendix tables

Table A.1
CBD radius (kCBD ) robustness.
CBD Radius House Price City Radius Energy Per Capita
Baseline 0.75 Miles 13.85 9.75 522.49
1 Mile∗ 13.90 9.60 524.16
1.5 Miles 13.98 9.60 527.17

%Δ relative to Baseline at same CBD Radius


Basic AV 0.75 Miles −1.72% 9.23% 1.35%
1 Mile∗ −1.92% 11.46% 1.39%
1.5 Miles −2.16% 11.46% 1.45%

No CBD Parking 0.75 Miles 0.28% −8.21% 5.68%


1 Mile∗ 0.15% −8.33% 6.11%
1.5 Miles −0.17% −7.29% 7.03%

Autonomous Taxis 0.75 Miles −18.29% −5.13% 3.87%


1 Mile∗ −18.58% −4.17% 3.89%
1.5 Miles −19.05% −5.21% 3.98%
Note: ∗ indicates original parameter value in Tables 2–4. Each model is solved under closed-city assump-
tions.

Table A.2
Time-cost of commuting (𝜏 ) robustness.
Time-Cost Fraction House Price City Radius Energy Per Capita
Baseline 0.50∗ 13.90 9.60 524.16

%Δ relative to Baseline
Basic AV 0.10 −3.86% 23.96% 2.69%
0.20 −2.86% 16.67% 2.08%
0.30∗ −1.92% 11.46% 1.39%
0.40 −0.98% 6.25% 0.68%

No CBD Parking 0.10 −1.50% 0.00% 7.43%


0.20 −0.51% −7.29% 6.63%
0.30∗ 0.15% −8.33% 6.11%
0.40 0.87% −10.42% 5.58%

Autonomous Taxis 0.10 −19.46% 3.13% 5.13%


0.20 −19.03% 0.00% 4.53%
0.30∗ −18.58% −4.17% 3.89%
0.40 −18.13% −8.33% 3.38%
Note: ∗ indicates original parameter value in Tables 2–4. Each model is solved under closed-city assumptions.

Table A.3
Autonomous taxi (𝜑0 , 𝜑1 , 𝜑2 ) robustness.
Pecuniary Costs House Price City Radius Energy Per Capita
Baseline See Text∗ 13.90 9.60 524.16

%Δ relative to Baseline
Autonomous Taxis −20% −19.44% 3.13% 5.77%
See Text∗ −18.58% −4.17% 3.89%
+20% −17.76% −11.46% 3.65%
Note: ∗ indicates original parameter value in Tables 2–4. Each model is solved under closed-city assumptions.

Appendix B. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.regsciurbeco.2019.103484.

References American Housing Survey, 2011. Public Use File. U.S. Department of Housing and Urban
Development, 2013, Washington, DC.
Albright, J., Bell, A., Schneider, J., Nyce, C., 2015. Automobile Insurance in the Era of Anderson, J.M., Nidhi, K., Stanley, K.D., Paul, S., Constantine, S., Oluwatola, T.A., 2016.
Autonomous Vehicles. Technical report. KPMG. Autonomous Vehicle Technology: A Guide for Policymakers. RAND Corporation,
Alonso, W., 1964. Location and Land Use: toward a General Theory of Land Rent. Santa Monica, Calif. RR-443-2-RC, https://www.rand.org/pubs/research_reports/
Harvard University Press. RR443-2.html.
Altmann, J.L., DeSalvo, J.S., 1981. Tests and extensions of the Mills-Muth simulation Arnott, R.J., MacKinnon, J.G., 1977. The effects of the property tax: a general
model of urban residential land use. J. Reg. Sci. 21 (1), 1–21. equilibrium simulation. J. Urban Econ. 4 (4), 389–407.
American Community Survey, 2010. 5-Year Estimates. U.S. Census Bureau, 2013, Bertaud, A., Brueckner, J.K., 2005. Analyzing building-height restrictions: predicted
Washington, CD. impacts and welfare costs. Reg. Sci. Urban Econ. 35 (2), 109–125.

19
W. Larson, W. Zhao Regional Science and Urban Economics 81 (2020) 103484

Bierstedt, J., Gooze, A., Gray, C., Peterman, J., Raykin, L., Walters, J., 2014. Effects of Larson, W., Zhao, W., 2017. Telework: urban form, energy consumption, and greenhouse
Next-generation Vehicles on Travel Demand and Highway Capacity. FP Think gas implications. Econ. Inq. (page forthcoming).
Working Group. . Lesne, D., Hummel, P., Radinger, J., Takahashi, K., 2017. How Disruptive Will a Mass
Blincoe, L., Seay, A., Zaloshnja, E., Miller, T., Romano, E., Luchter, S., Spicer, R., 2002. Adoption of Robotaxis Be?. Technical report. UBS.
The Economic Impact of Motor Vehicle Crashes, 2000. Technical report. National Lucas, R.E., Rossi-Hansberg, E., 2002. On the internal structure of cities. Econometrica
Highway Traffic Safety Administration. 70 (4), 1445–1476.
Borck, R., 2016. Will skyscrapers save the planet? building height limits and urban McDonald, J.F., 2009. Calibration of a monocentric city model with mixed land use and
greenhouse gas emissions. Reg. Sci. Urban Econ. 58, 13–25. congestion. Reg. Sci. Urban Econ. 39 (1), 90–96.
Borck, R., Brueckner, J.K., 2018. Optimal energy taxation in cities. J. Assoc. Environ. Meyer, J., Kain, J., Wohl, M., 1965. The Urban Transportation Problem. Harvard
Resour. Econ. 5 (2), 481–516. University Press.
Brown, A., Gonder, J., Repac, B., 2014. An analysis of possible energy impacts of Mieszkowski, P., Mills, E.S., 1993. The causes of metropolitan suburbanization. J. Econ.
automated vehicle. In: Road Vehicle Automation. Springer, pp. 137–153. Perspect. 7 (3), 135–147.
Brueckner, J.K., 1987. The structure of urban equilibria: a unified treatment of the Muth Mills, E.S., 1967. An aggregative model of resource allocation in a metropolitan area.
- Mills model. Handb. Reg. Urban Econ. 2, 821–845. Am. Econ. Rev. 57 (2), 197–210.
Brueckner, J.K., Franco, S.F., 2017. Parking and urban form. J. Econ. Geogr. 17 (1), Muth, R., 1969. Cities and Housing: the Spatial Pattern of Urban Residential Land Use.
95–127. University of Chicago Press.
Brueckner, J.K., Franco, S.F., 2018. Employer-paid parking, mode choice, and Muth, R.F., 1975. Numerical solution of urban residential land-use models. J. Urban
suburbanization. J. Urban Econ. 104, 35–46. Econ. 2 (4), 307–332.
Brueckner, J.K., Mills, E., Kremer, M., 2001. Urban Sprawl: Lessons from Urban National Highway Traffic Safety Administration, 2013. Preliminary Statement of Policy
Economics [with Comments]. Brookings-Wharton Papers on Urban Affairs, pp. Concerning Automated Vehicles.
65–97. Rappaport, J., 2016. Productivity, Congested Commuting, and Metro Size. Technical
Coulson, N., Engle, R.F., 1987. Transportation costs and the rent gradient. J. Urban report. .
Econ. 21 (3), 287–297. Residential Energy Consumption Survey, 2009. Public Use Microdata. U.S. Energy
Duranton, G., Puga, D., 2015. Urban land use. In: Gilles Duranton, J.V.H., Strange, W.C. Information Administration, 2011, Washington, DC.
(Eds.), Handbook of Regional and Urban Economics, Volume 5 of Handbook of Saiz, A., 2010. The geographic determinants of housing supply. Q. J. Econ. 125 (3),
Regional and Urban Economics, Elsevier, pp. 467–560. 1253–1296.
Duranton, G., Turner, M.A., 2011. The fundamental law of road congestion: evidence Santi, P., Resta, G., Szell, M., Sobolevsky, S., Strogatz, S.H., Ratti, C., 2014. Quantifying
from US cities. Am. Econ. Rev. 101 (6), 2616–2652. the benefits of vehicle pooling with shareability networks. Proc. Natl. Acad. Sci. 111
Fagnant, D.J., Kockelman, K., 2015. Preparing a nation for autonomous vehicles: (37), 13290–13294.
opportunities, barriers and policy recommendations. Transp. Res. A Policy Pract. 77, Shoup, D., 2005. The High Cost of Free Parking. Planners Press, American Planning
167–181. Association.
Fernandes, P., Nunes, U., 2012. Platooning with IVC-enabled autonomous vehicles: U.S. Department of Energy, 2000. Electric and hybrid vehicle research, development,
strategies to mitigate communication delays, improve safety and traffic flow. IEEE and demonstration program; petroleum-equivalent fuel economy calculation. Fed.
Trans. Intell. Transp. Syst. 13 (1), 91–106. Reg. 65 (113), 36986–36992.
Fujita, M., Ogawa, H., 1982. Multiple equilibria and structural transition of U.S. Energy Information Administration, 2011. Energy Consumption, Expenditures, and
non-monocentric urban configurations. Reg. Sci. Urban Econ. 12 (2), 161–196. Emissions Indicators Estimates, Selected Years, 1949-2011. Annual Energy Review
Glaeser, E.L., Gottlieb, J.D., 2009. The wealth of cities: agglomeration economies and 2011, p. 15.
spatial equilibrium in the United States. J. Econ. Lit. 47 (4), 983–1028. Wadud, Z., MacKenzie, D., Leiby, P., 2016. Help or hindrance? The travel, energy and
Greenblatt, J.B., Saxena, S., 2015. Autonomous taxis could greatly reduce carbon impacts of highly automated vehicles. Transp. Res. A Policy Pract. 86, 1–18.
greenhouse-gas emissions of US light-duty vehicles. Nat. Clim. Chang. 5 (9), West, B.H., McGill, R.N., Sluder, S., 1999. Development and verification of light-duty
860–863. modal emissions and fuel consumption values for traffic models. U.S. Department of
Gyourko, J., Saiz, A., Summers, A., 2008. A new measure of the local regulatory Transportation, Federal Highway Administration. Report FHWA-RD-99-068.
environment for housing markets: the Wharton Residential Land Use Regulatory Wheaton, W.C., 1998. Land use and density in cities with congestion. J. Urban Econ. 43
Index. Urban Stud. 45 (3), 693–729. (2), 258–272.
Larson, W., Liu, F., Yezer, A., 2012. Energy footprint of the city: effects of urban land Wheaton, W.C., 2004. Commuting, congestion, and employment dispersal in cities with
use and transportation policies. J. Urban Econ. 72 (2–3), 147–159. mixed land use. J. Urban Econ. 55 (3), 417–438.
Larson, W., Yezer, A., 2015. The energy implications of city size and density. J. Urban Zakharenko, R., 2016. Self-driving cars will change cities. Reg. Sci. Urban Econ. 61,
Econ. 90, 35–49. 26–37.

20

You might also like