UC Office of the President
Recent Work
Title
The effect of the price of gasoline on the urban economy: From route choice to
general equilibrium
Permalink
https://escholarship.org/uc/item/0b23p64c
Journal
Transportation Research Part A: Policy and Practice, 46(6)
Authors
Anas, Alex
Hiramatsu, Tomoru
Publication Date
2012-07-01
Peer reviewed
eScholarship.org Powered by the California Digital Library
University of California
The effect of the price of gasoline on the urban economy: From route choice
to general equilibrium
Alex Anas
⇑
, Tomoru Hiramatsu
1
State University of New York at Buffalo, Department of Economics, 415 Fronczak Hall, Amherst, NY 14260, United States
article info
Keywords:
Gasoline price
Urban structure
Travel
abstract
RELU-TRAN2, a spatial computable general equilibrium (CGE) model of the Chicago MSA is
used to understand how gasoline use, car-VMT, on-the-road fuel intensity, trips and loca-
tion patterns, housing, labor and product markets respond to a gas price increase. We find a
long-run elasticity of gasoline demand (with congestion endogenous) of 0.081, keeping
constant car prices and the TFI (technological fuel intensity) of car types but allowing con-
sumers to choose from car types. 43% of this long run elasticity is from switching to transit;
15% from trip, car-type and location choice; 38% from price, wage and rent equilibration,
and 4% from building stock changes. 79% of the long run elasticity is from changes in
car-VMT (the extensive margin) and 21% from savings in gasoline per mile (the intensive
margin); with 83% of this intensive margin from changes in congestion and 17% from
the substitution in favor of lower TFI. An exogenous trend-line improvement of the TFI
of the car-types available for choice raises the long-run response to a percent increase in
the gas price from 0.081 to 0.251. Thus, only 1/3 of the long-run response to the gas
price stems from consumer choices and 2/3 from progress in fuel intensity. From 2000
to 2007, real gas prices rose 53.7%, the average car fuel intensity improved 2.7% and car
prices fell 20%. The model predicts that from these changes alone, keeping constant popu-
lation, income, etc. aggregate gasoline use in this period would have fallen by 5.2%.
Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction
How does the urban economy respond to an increase in the price of gasoline? Since the mid 1970s econometric studies
have measured the price elasticity of the demand for gasoline, using state, national and international cross-sectional or time
series data. These studies tell us how a change in the gasoline price would affect total vehicle miles traveled (VMT), changes
in the stock of vehicles owned and the average fuel economy of cars being operated. From these studies, we know the prob-
able ranges of the price elasticity of the aggregate demand for gasoline, and the ‘‘rebound effect’’, the propensity to drive
more as the fuel economy of cars improves in response to higher gasoline prices.
Small and Van Dender (2007) estimate that both the price elasticity of the demand for gasoline and the rebound effect
declined over time (except, perhaps, in recent years not yet studied). Hughes et al. (2008) agree on the decline of the elas-
ticity but not necessarily on the decline of the rebound effect, although differences between the two studies appear to be
explainable by differences in specification. One reason for the declining elasticity is the fact that since the oil embargo of
1975, incomes have risen but gasoline prices have remained stable or declining (except for recent years). Another reason
is that CAFE standards and the fuel economy of cars on the market have improved.
0965-8564/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.tra.2012.02.010
⇑
Corresponding author. Tel.: +1 716 645 8663; fax: +1 801 749 7805.
E-mail addresses: alexanas@buffalo.edu (A. Anas), th29@buffalo.edu (T. Hiramatsu).
1
Tel: +1 716 645 8663.
Transportation Research Part A 46 (2012) 855–873
Contents lists available at SciVerse ScienceDirect
Transportation Research Part A
journal homepage: www.elsevier.com/locate/tra
In this paper, we will evaluate the effect of a higher gasoline price in a computable general equilibrium (CGE) model of a
spatially disaggregated urban economy (the Chicago MSA) as it responds from the very short run of travel route adjustments
to the long run of location and building stock changes. The structural model treats explicitly aspects that are suppressed in
reduced form econometric modeling. This advantage of CGE models sheds more light onto our understanding of how the
gasoline price affects urban form and structure.
1.1. Econometric studies and CGE modeling
We consider some limitations of reduced-form econometric specifications and explain how our CGE model, RELU-TRAN2,
attempts to compensate for them:
(a) Endogenous congestion: Road congestion indirectly affects gasoline consumption but is hard to treat well in an econo-
metric model. Efforts to capture the congestion effect are few and have relied on very rough aggregate proxies of con-
gestion such as the level of urbanization in a state or the ratio of adults to lane miles of highways ( Small and Van
Dender, 2007); or the average metropolitan-wide congestion delay indices of the Texas Transportation Institute
(Hymel et al., 2010). In our spatially disaggregated CGE model of the Chicago MSA, route choices on a spatial road net-
work are modeled explicitly for different incomes and values of time, taking into account monetary cost (e.g. gasoline)
as well as travel time, as these times and costs are endogenized by congestion.
(b) Effects of adjustments in urban markets: Labor, housing, and land markets are changed by travel behavior and in turn
affect travel behavior. But the indirect changes in gasoline consumption from endogenous changes in wages, rents and
goods prices, in location decisions and in building stocks are not explicitly treated in econometric studies. Our CGE
model treats travel and car use decisions for commuting and for discretionary (non-work) trips and also treats labor
supply, location of work and residence, location of firms, housing type and market adjustments in rents, wages and
retail prices, the asset prices of buildings and the adjustment of building stocks. These adjustments are simulated
in stages. Thus, we are able to trace the change in elasticity from the very short run to the long run, decomposing
the effect of each stage on the long run elasticity.
(c) Sorting out and decomposing various rebound effects: In the econometric literature, the commonly held definition of the
‘‘rebound effect’’ seems to be the increase in the use of an appliance when its fuel intensity falls. In the case of cars
most authors have measured use of the car by VMT (vehicle miles traveled), although TRIPS (number of trips made),
HOURS (total hours of travel) or GAS (gasoline burned) would all also be equally valid measures of use in the extensive
margin. A car’s use in the intensive margin can be measured by MPG (miles per gallon), by MPH or speed (miles per
hour), by GPM (gallons per mile) or by monetary cost (dollars per mile). Our CGE model predicts all of these indicators
of car use and the trade-offs among them that consumers make when they allocate their time and income between
travel and other activities and when they choose a bundle of trips to maximize utility. In the literature, GPM is nor-
mally used as a measure of fuel intensity. But we distinguish between the TFI (technological fuel intensity) of a car’s
engine and its on-the-road fuel intensity which is determined by the car’s speed under congested conditions given its
TFI. The TFI of cars actually driven improves by technological progress in the car industry or by consumer choice of
more fuel efficient cars. Then, as the price of gasoline rises, TFI and/or speed improve indirectly and the monetary cost
of driving a mile rebounds. From this, rebounds occur not only in VMT, but also in GAS, HOURS and TRIPS, taking back
from the initial reductions in these variables induced by the higher gasoline price.
(d) Changes in fuel intensity: A reduction over time in average fuel intensity is a well-observed trend. Fig. 1 plots the
national trends for 1980–2009.
2
Improvement comes in part from consumers choosing more fuel efficient vehicles in
response to a higher gas price. This would raise the demand for such cars causing imperfectly competitive car-makers
to produce more of them while marking-up prices. At the same time, in the used car market, fuel intensity is higher
on average and the relative prices of used cars would fall, offsetting in part the adoption of the more fuel efficient vehi-
cles. Changes in the supply of vehicles by fuel intensity could also be driven in some measure by CAFE standards. Bento
et al. (2009) have attempted to model the effects of these standards in a national model with endogenous car production,
but their model does not include urban structure and markets.
Our CGE model is focused on metropolitan structure. It treats as endogenous consumer choices among car-types but does
not treat car-production as endogenous. In the model, five abstract car types shown in Fig. 2 are available to consumers, and
differ by their TFI. Each higher curve in Fig. 2 represents a car type of higher TFI. Each curve also captures that for a given car
type on-the-road fuel intensity falls with car speed, making a relatively flat bottom and rising at high speeds.
3
We assume
that higher TFI cars are larger, more comfortable, safer but also more expensive to own. The choice of a car type trades off higher
ownership and gasoline cost for car size, comfort and safety. Average on-the-road fuel intensity is determined by the distribu-
tion of the consumers among the car types and by the traffic congestion which determines speed. In the legend of Fig. 2, the
2
The MPG data for Fig. 1 was not available post 2007.
3
The empirics of the curves in Fig. 2 will be discussed in Section 3.
856 A. Anas, T. Hiramatsu / Transportation Research Part A 46 (2012) 855–873
gallon per mile (GPM) of the car types is compared at a speed of 45 mph to show how GPM increases (MPG decreases) by the car
type’s TFI which is the height of the curves. The supply of cars of any of the five fuel intensities to the Chicago MSA is assumed
perfectly elastic. This is justified, at least in part, by the fact that any metropolitan area is small in the national car market and
that consumers can choose TFI by mixing new and used cars. Trend line progress in car TFI is treated in the CGE model by exog-
enously shifting (downward) the curves shown in Fig. 2.
1.2. How the simulations are structured
We design a series of nested simulations in which relevant processes of the CGE model are solved while higher level pro-
cesses are turned off. The consumer in the CGE model completes a hierarchically ordered sequence of choices in response to
an increase in the price of gasoline. Short term adjustments are completed fairly quickly. The fastest is changing one’s travel
route on each car trip, changing the mode of travel of a trip to or from auto to public transit and non-motorized, changing the
number and length of non-work trips, and changing one’s car type by selecting one of the curves in Fig. 2. Changes in location
of residence and/or workplace are longer term as are market clearing adjustments in wages, rents, prices and building stocks.
At every stage in the hierarchy, we endogenously calculate congestion and thus speed on the roads of the network. The
endogenous congestion requires a re-statement of the standard microeconomic textbook definition of the demand function
for gasoline. In the standard definition aggregate gasoline demand falls with the per gallon price, keeping constant all other
prices, consumer incomes and travel times. The demand for gasoline with endogenous congestion is the aggregate gasoline
GPM [MPG] at 45 MPH:
0
0.02
0.04
0.06
0.08
0.1
0.12
5 101520253035404550556065707580
Speed (MPH)
0.025 [40.0]
0.029 [34.5]
0.034 [29.4]
0.038 [26.3]
0.042 [23.8]
Fig. 2. Technological fuel intensity (TFI) of the five car types in RELU-TRAN2.
0
5
10
15
20
25
30
35
1980 1985 1990 1995 2000 2005 2009
Average U.S. passenger car MPG New passenger car MPG CAFE standards for passenger cars
Fig. 1. Time trends of miles per gallon (1980–2009). Source: Bureau of Transportation Statistics.
A. Anas, T. Hiramatsu / Transportation Research Part A 46 (2012) 855–873
857
demanded keeping all other prices and consumer incomes constant, when travel times are re-equilibrated via congestion, as
the per gallon price changes. Thus, the textbook aggregate demand for gasoline would be d(p|t, X), where p is the per-gallon
price, t is (average) travel time and X is other variables (variables after the vertical bar being kept constant). The demand
with endogenous congestion is DðpjXÞdðpjtðpÞ; XÞ, where t(p) is the congested travel time re-equilibrated as the gasoline
price p changes. We calculate the shortest-run demand for gasoline (with endogenous congestion) if consumers adjusted
only their route choices. The demand curve for gasoline becomes more elastic as consumers adjust route-and-mode choices,
route-mode-and-non-work trips and then car TFI and location choices and so on, congested travel times being endogenously
determined for each such adjustment.
1.3. Summary and results
The econometric literature on the price elasticity of gasoline is reviewed in Section 2.1. In Section 2.2 a theoretical anal-
ysis is presented, as a short-hand for the CGE model. The purpose of this is to clarify the main channels of causation in the
demand for gasoline. The analysis combines the extensive margin of VMT with the intensive margin of GPM, and separates
the direct effect of the gasoline price from the rebound effects. We show that there is a rebound in GPM and in the gasoline
cost per mile. These rebounds stem from two indirect effects of the higher gas price: the consumer’s choice of a lower TFI car-
type (a lower curve in Fig. 2) and the reduction in congestion as fewer miles are driven on aggregate.
In Section 3, the CGE model’s specification is described, with most of the attention on the consumer’s utility maximization
and the choice structure. In Section 4, we describe the data sources used to calibrate the CGE model, how the calibrated mod-
el was fitted to the data to validate it, and how well the elasticity measures and the value of time used in the calibrated mod-
el agree with values from the literatures. Section 5 is the key section. There we report on the structured simulations with
endogenous congestion and the composition in stages of the price elasticity of the demand for gasoline. The results are dis-
cussed in the context of the econometric studies and show that our long run price elasticity agrees well with those of the
most recent. The advantage of our approach, however, is that we can decompose the price elasticity by stage of adjustment,
and by the extensive margin (VMT) versus the intensive margin (gallons per mile, GPM), quantifying direct and rebound
effects.
Our long run price elasticity of gasoline is 0.081, keeping constant the TFIs of the model’s five car-types and the car
prices. Decomposing this long-run elasticity by stage of adjustment, about 43% is due to switches from cars to transit;
15% due to changes in trips made and in car-type choices, and job and residence locations. Another 38% is induced by
changes in rents, wages and prices in the urban markets. About 4% is induced by long run adjustments in building stocks.
Looking at the composition of the elasticity from a different angle, the extensive margin of car miles traveled (VMT) accounts
for 79%, while the intensive margin of gallons per mile (GPM) contributes 21%, about 17% of which is from congestion
improvement and about 4% from the substitution of lower TFI by consumers, constant the cost of owning the car-types
and their TFIs.
Our understanding of the long run response to a gasoline price increase is extended by two more simulations of the CGE
model. In the first simulation, the long run response of the demand for gasoline to its price, in the presence of a 2% trend line
improvement in the TFI of cars on the market (roughly similar to what has been experienced over a recent 5 year period) and
assumed to occur concurrently with the long run adjustments is 0.251 per percent increase in the gas price. Then, since the
long-run elasticity (constant TFI) was 0.081, 2/3rd of the long run response is caused by the modest trend-line improve-
ment in TFI and only 1/3rd by the endogenous choice of TFI by consumers.
In the second simulation, we model the impact of changes in the period 2000–2007. Over this period, the real annual
average price of gasoline in the Chicago MSA increased by 53.7% while the TFI of cars on the market improved by 2.7% on
average, car acquisition costs decreasing from 19.2% for fuel efficient cars to 23% for the less fuel efficient cars. We report
the separate and cumulative long run effects from these changes on gasoline consumption, VMT, speed, gallons per mile
and number of work and non-work trips. The gas price increase by itself causes a 4.21% decrease in gasoline use, but the
TFI improvement extends this decrease in gasoline use by half as much to 6.34%. But the decrease in car prices takes back
about 18% of the combined effect of the TFI and the gas price. The overall effect of all three changes together is a 5.2% drop in
gasoline use keeping constant other trends such as population and income.
In Section 5, we also report on how consumers in different commuting arrangements in the model respond to the gas
price increase by switching arrangement. Those car commuters most impacted by the gas price would be suburban residents
working in the central city and those least impacted suburb-to-suburb car commuters. Conversely, those hurt most by the
gasoline price increase benefit the most from a change in TFI. The result is that a gas price increase causes switches to sub-
urb-to-suburb commuting and could result in more suburbanization of car commuters. But this is offset as other suburban
commuters switch to transit and do so by moving to a residence in the central city, where transit is more accessible.
In Section 5, sensitivity analyses of the model’s elasticities with respect to key parameters of the CGE model are also re-
ported. We focus on increasing or decreasing idiosyncratic consumer taste dispersion to see how the elasticity of GAS, VMT
and GPM with respect to the gas price changes. Less (more) idiosyncratic heterogeneity increases (decreases) the elasticity.
In the short run a 20% change in taste dispersion causes a just less than 20% change in the elasticity, but in the long run as
more choice margins become available, the elasticity converges back towards its original value. That is the long run elasticity
is about three times as robust as the short run elasticity in the face of perturbations in consumer heterogeneity. Section 6
concludes.
858 A. Anas, T. Hiramatsu / Transportation Research Part A 46 (2012) 855–873
