Transport Policy 133 (2023) 234–241

Contents lists available at ScienceDirect

Transport Policy
journal homepage: www.elsevier.com/locate/tranpol

The elasticity of demand on urban highways: The case of Santiago

Marco Batarce a, *, Franco Basso b, c, Leonardo J. Basso c, d
a
Facultad de Economía y Gobierno, Universidad San Sebastián, Santiago, Chile
b
School of Industrial Engineering, Pontificia Universidad Católica de Valparaíso, Valparaíso, Chile
c
Instituto Sistemas Complejos de Ingeniería (ISCI), Chile
d
Civil Engineering Department, Universidad de Chile, Santiago, Chile

A B S T R A C T

This study examines the relevant flow variables on urban highways in Santiago de Chile. The objective is to contribute to the knowledge of the elasticity of the flow
on highways in less developed nations, bearing in mind that existing studies tend to focus on European countries and the United States. Using panel data models and
five-year information from Autopista Central in Santiago, we study the elasticity of tolls, gasoline price, diesel price, and economic activity. The results indicate that
economic growth is the primary variable that explains the increase in flows, which has brought a sustained rise in the country’s motorization rate. When dis­
tinguishing economic activity between mining and other sectors, it is observed that the relevant activity in Santiago is not mining. Additionally, the flow is inelastic to
the toll rate in the short term, while elasticity increases in the long term. The flow is inelastic with respect to fuel prices but less so than toll tariffs.

1. Introduction failure rates and classification errors of this technology are minimal
(<1%), since the concessionaire’s income depends on this infrastructure
Since the operation of urban highways in Santiago de Chile began in (Basso et al., 2018). The price paid by the user depends on the type of
the 2000s, demand for this type of infrastructure has increased steadily vehicle (light, heavy, heavy with more than two axles, or motorcycle),
over time. Although public opinion offers many arguments regarding the the number of AVI gantries through which it circulates, and the type of
reasons for this behavioral pattern, few studies have quantified the rate (low, high, and saturation) of the time slot in which the vehicle pass.
importance of different variables in this upturn in the case of Santiago. The information used in this research covers five years of operation
This study aims to examine the effect of those relevant variables on flow between 2013 and 2018.
patterns on urban highways by estimating elasticity for Santiago. The scientific literature reports numerous studies that have esti­
Although our findings are not directly generalizable, they do contribute mated different responses to changes in transportation prices. Oum et al.
to the knowledge base on freeway flow elasticity. This is particularly (2007) describe the characteristics of the different elasticities to be
relevant given the fact that studies on the elasticity of highway travel considered in the transportation area and exposes some deficiencies in
demand have focused on developed countries in Europe and the United the literature regarding their calculation. They also provide recom­
States. mendations to improve those deficiencies to obtain comparable and
Autopista Central, an urban highway located in Santiago, Chile, extensible results. Litman (2021) describes the main factors influencing
provided the information for this study. This highway has a total length transport, its characteristics, and its relationship with different modes.
of 62.28 km. and crosses the city in a north-south direction through two Additionally, it summarizes the results obtained by the literature,
high-speed roads: Route 5 and General Velásquez, as shown in Fig. 1. exposing and comparing the findings, segmented by nature, type of
The highway is privately operated, and users are charged for their use estimation, vehicles, place, and period in which they were developed.
through the information obtained from the interaction between gantries Other literature reviews also seek to compare elasticities obtained by
with Automatic Vehicle Identification (AVI) technology and devices numerous studies in the area of transportation. Generally speaking, the
installed in vehicles, called TAGs. By law, each vehicle that uses this results are similar. For example, average elasticities are similar in
highway must have this device equipped on the front windshield to magnitude and proportion in the short and long term (Dunkerley et al.
circulate on the road. The TAG emits a microwave signal each time the (2014); Goodwin et al. (2004); Graham & Glaister (2002)). Goodwin
vehicle crosses one of the AVI gates, allowing the vehicle to be identified et al. (2004) conducted a literature review considering 69 studies that
without impeding its normal circulation. Unlike other technologies, the use different estimation and data methods. In general terms, they point

* Corresponding author.
E-mail address: [email protected] (M. Batarce).

https://doi.org/10.1016/j.tranpol.2023.02.008
Received 5 September 2022; Received in revised form 2 February 2023; Accepted 5 February 2023
Available online 7 February 2023
0967-070X/© 2023 Elsevier Ltd. All rights reserved.
M. Batarce et al. Transport Policy 133 (2023) 234–241

out that in the face of a 10% increase in the price of gasoline, the volume between time blocks (peak hours), type of vehicle, and geographical
of traffic drops by 1% in the short term (1 year) and 3% in the long term areas. They obtain elasticities between − 0.069 and − 0.216, consistent
(5 years). As for gasoline consumption, this falls by about 2.5% in one with the literature. The authors show that the price elasticity is greater
year and about 6% in five years. On the other hand, with a 10% increase in private cars compared to other types of vehicles (commercial vehicles,
in income, vehicle numbers and gasoline consumed will increase by 4% public transport, taxis, motorcycles, etc.). Matas and Raymond (2003)
over a 1-year period and over 10% over the 5-year period, while traffic estimate the elasticity of demand for toll roads in Spain with data be­
volume grows by 2% and 5% in the short and long term, respectively. tween 1981 and 1998, using panel data and first differences in a log-log
Graham and Glaister (2002) and Dunkerley et al. (2014) develop model. They study the elasticities and their differences between four
reviews of scientific articles that determine the response of motorists to groups of toll gates from lower to higher elasticity in the face of rate
changes in the price of gasoline. Both focus their research mainly on variations. They obtain elasticities of between − 0.209 and − 0.828 for
Europe and North America. Graham and Glaister (2002) conclude that the short term and between − 0.33 and − 1.307 for the long term. They
fuel price elasticity sits between − 0.2 and − 0.3 in the short term and also estimate the response of demand to changes in GDP and oil prices,
between − 0.6 and − 0.8 in the long term. Dunkerley et al. (2014) indi­ obtaining short-term elasticities of 0.89 and − 0.336 and 1.405 and
cate that the elasticity of road use to changes in fuel prices is between − 0.531 in the long term for GDP and fuel prices, respectively.
− 0.1 and − 0.5. In Chile, studies have addressed the demand elasticity for inter-urban
In Spain, Bakhat et al. (2017) estimate the price elasticity of trans­ and urban highways. Saens and Lobos (2013) estimate the price elas­
port fuels – differentiating those periods with and without economic ticity of demand in the use of inter-urban highways through time series
crisis and using data between 1999 and 2015 – through data panel and SUR models, using monthly data from 2004 to 2007 corresponding
models. The authors found that, during crisis periods, the elasticity is to 21 toll gates between La Serena and Puerto Montt. The authors find
lower both in the short and long term for gasoline and diesel. The results that the price elasticity of demand for highway use is − 0.167 for cars
showed that short-term elasticity is − 0.318 for diesel and − 0.248 for and − 0.051 for trucks, while the price elasticity of fuel is − 0.35 for cars
gasoline without a crisis. While during a crisis, they reach − 0.311 and and − 0.118 for trucks. Income elasticity reaches 1.515 and 1.376 for
− 0.244. The long-term elasticities for diesel and gasoline are − 0.591 cars and trucks, respectively.
and − 0.706 outside of a crisis and − 0.578 and − 0.694 within one. As for urban highways, de Grange et al. (2015) estimate the price
The effect of the price of tolls, electronic tolls, or tariffs has also been elasticity of demand for urban roads in Santiago. Using data from 2009
addressed in the literature. Olszewski and Xie (2002) estimate elastici­ to 2011 and log-linear models, the authors estimate differentiated
ties for toll roads in Singapore between 1998 and 1999, differentiating elasticities for each highway and conclude that those highways with

Fig. 1. Autopista central, Santiago, Chile (based on Basso et al. (2021)).

235
M. Batarce et al. Transport Policy 133 (2023) 234–241

more alternative roads show a larger elasticity to the toll price. In Santiago since the industry is concentrated in other regions of Chile.
addition, they find that users are more sensitive to fuel price changes Table 2 shows the descriptive statistics of the variables used in the
than to toll rate changes. Toll price elasticities vary between − 0.047 and estimation, and Table 3 shows the correlation between them.
− 0.47 between highways, while the fuel price elasticity varies between The following graphs show the monthly evolution of aggregate
− 0.09 and − 0.52. Table 1 summarizes the results reported in the pre­ vehicle flows for the entire highway and the independent variables used
viously described literature. in the models. Figs. 2 and 3 show vehicle traffic by category. Category 1
For this study, we use the flow data aggregated monthly in three (cars and vans) shows a level of over 20 million passes per month
categories: light vehicles, two-axle trucks and buses, and heavy trucks. (Fig. 2), well above Category 2, which reaches levels of 3 million, and
The flow is recorded at each toll gate along the highway (see Fig. 1); Category 3, with levels of 1 million (see Fig. 3). Traffic growth is
therefore, we have information with a high level of spatial disaggrega­ observed in all categories.
tion. De Grange et al. (2015) conducted a study similar to ours, using The evolution of economic activity is shown in Fig. 4. It shows the
flow information aggregated per hour. Unlike that study, we include total IMACEC, mining IMACEC, and non-mining IMACEC. We can see
variables that measure economic activity, which turns out to be highly that these are highly volatile variables and increasingly so, as is the
significant. Its elasticity to the tariff is very short-term compared to the norm for product measurements. Another important aspect is that the
one we obtained. mining and non-mining IMACEC are not correlated, which justifies the
The methodology of this article consists of estimating econometric inclusion of both variables separately. On the other hand, the total
models with different specifications and the following variables: rates, IMACEC is correlated to the non-mining IMACEC, so using only the first
price of the different fuel types, and economic activity indices, dis­ of these indices can induce bias in the estimation by omitting a relevant
aggregated between mining and non-mining. We also consider inter­ variable, such as the mining IMACEC.
temporal substitution patterns. Additionally, we examined price Fig. 5 shows the evolution of fuel prices. No trend is observed, but a
variation over time, notably when a toll gate changed from a normal rate relatively high volatility. This volatility is due to variations in the in­
to a heavy-traffic or saturation rate. ternational diesel price and the dollar. The high variability of the series
The remainder of the study is organized as follows. Section 2 presents is a characteristic that improves the precision of the estimated param­
and describes the data collected for the analysis. Section 3 explains in eters. In addition, both series are highly correlated, although this is not a
detail the methodology and the specification used for the econometric problem since both variables are never used in the same model.
model, while Section 4 provides detailed estimation results. Finally, Fig. 6 shows the evolution of the base rate per off-peak kilometer for
Section 5 discusses and analyzes the results. each of the three categories. As expected, the behavior of the highway
tariff is entirely deterministic and increasing since it is established ac­
2. Data cording to the concession contract.

Vehicle flow data on the highway was provided by Autopista Central. 3. Methodology
The information corresponds to vehicular flow by electronic toll
collection gantry, category, and period for workdays between January If we say each section of the highway represents an individual
2013 and November 2018. The vehicle categories correspond to the observed over time, the data available for the estimation corresponds to
following. a panel; therefore, the estimation techniques consider the panel data
model approach. This approach supposes that each section is an indi­
• Category 1: Cars and vans vidual with particular characteristics and, therefore, the sections are
• Category 2: Buses and two-axle trucks heterogeneous. Two possible basic approaches to capture this hetero­
• Category 3: Trucks with more than two axles geneity are fixed effects and random effects. In both cases, heterogeneity
• Category 4: Unidentified vehicles is represented by a parameter that does not interact with other inde­
pendent variables of the model. In a linear-in-parameters model, like
We do not use data on Category 4 because it represents a low fraction those used in this work, this assumption means that each section of the
of the highway flow (less than 0.1%) and mixes unclassified vehicles highway has a different intercept (or constant term).
from the other categories. Random effects models assume that heterogeneity is independent of
There are 48 periods, dividing the day into 30-min intervals. The the other variables in the model and that the representing parameters
total monthly flow for a gantry corresponds to the sum of these periods. are independent and identically distributed random variables. With
We obtained pricing information from the annual reports published by these assumptions, the estimation method determines the parameters of
Autopista Central to inform the public of their rates. Three tariff levels the random effects distribution, which are generally two: mean and
depend on the flow level in the previous year: low, high, and saturation. variance. Fixed effects models do not assume independence of the in­
The high rate equals the low rate x2, while the saturation rate equals the dependent variables in the model and are, therefore, more general.
low rate x3. We created a database from these reports, showing the price However, these models require an estimation of the same number of
per gantry, category, and period for all the sample months. It was parameters as there are individuals (highway sections) in the sample,
essential to detect those cases in which a gantry passed from the low rate which creates the problem of incidental parameters, which is to say, the
to the high rate or saturation rate. number of parameters grows with the sample, which can lead to sta­
We use the prices of 95-octane gasoline (CL$/lt)and diesel (CL$/lt) tistical problems known as bias and inconsistency. The problem of
in the Santiago Metropolitan Region as independent variables. For this, incidental parameters is more serious when the panel is short (there are
we collected the fuel prices information from the National Energy few periods in the sample) or the models are not linear. Since this work
Commission. Also, to be used as an independent variable, we collected comprises a sample of 70 months and estimates linear models, using
the monthly economic activity index (IMACEC) reported by Chile’s fixed effects will not be a problem. There are two equivalent estimators
Central Bank. This index corresponds to the production volume at prices for a linear model with fixed effects: least-square dummy variables and
of the previous year, and the base level equal to 100 is the average of within-group estimator (Arellano, 2003). We use the latter estimator,
2013. The IMACEC was collected disaggregated by economic sector, which eliminates the fixed effects by demeaning the observations,
although we use partial aggregations to differentiate by economic because our primary interest is the coefficients representing elasticities.
sector. It seems particularly relevant to differentiate mining from other Therefore, the models estimated in this work are linear (in the param­
activities. Mining represents a significant portion of monthly economic eters) with fixed effects per gantry.
activity. Still, it should have a weak effect on the behavior of flows in When determining the level of aggregation of the dependent

236
M. Batarce et al. Transport Policy 133 (2023) 234–241

Table 1
Summary of elasticities reported in the literature.
Study Type Study Dependent Variable Elasticity regarding: Short-term Long-term

Revision of literature/Meta- Goodwin et al. (2004) Fuel consumption (total) Fuel prices per liter − 0.25 − 0.64
analysis Fuel consumption (per vehicle) − 0.08 − 1.1
Km-vehicle − 0.1 − 0.3
Stock of vehicles − 0.08 − 0.25
Graham and Glaister Fuel consumption Fuel prices per liter − 0.2 a − 0.3 − 0.6 a − 0.8
(2002)
Dunkerley et al. (2014) Flow Fuel prices − 0.1 a − 0.5
Income 0.5 a 1.4
Estimation of elasticities Taiebat et al. (2019) Miles per vehicle Time cost and fuel prices − 0.4
combined
Matas and Raymond Highway demand GDP 0.89 1.405
(2003) Price of petroleum − 0.336 − 0.531
Toll prices − 0.209 a − 0.330 a
− 0.828 − 1.307
Bakhat et al. (2017) Fuel consumption (Outside of a crisis/During Price of gasoline − 0.248/ − 0.591/
a crisis) − 0.244 − 0.578
Price of diesel − 0.318/ − 0.706/
− 0.311 − 0.694
Olszewski and Xie Flow Toll prices between − 0.069 y − 0.216
(2002)
Saens and Lobos (2013) Highway demand (cars/trucks) Toll prices − 0.17/-0.05
Income 1.52/1.38
Fuel prices − 0.35/-0.12
de Grange et al. (2015) Flow Toll prices (electronic) between − 0.047 y − 0.47
Fuel prices between − 0.09 y − 0.52

estimation. Second, the logarithmic specification makes it possible to

Table 2
directly obtain the elasticity of the flow with respect to the different
Summary of variables used in the estimation.
independent variables. Third, the logarithmic models performed statis­
Minimum Maximum Median Mean SD tically better in the preliminary estimates. In what follows then, the
Tariff per kilometer Cat. 1 48 74 56 59 8 specifications and results of the logarithmic models will be presented.
(CL$) Therefore, the dependent variable is the logarithm of the total
Tariff per kilometer Cat. 2 97 147 112 117 16
monthly flow by vehicle category and gantry, without distinguishing by
(CL$)
Tariff per kilometer Cat. 3 145 221 168 176 25
the day period. After preliminary estimations, we consider as indepen­
(CL$) dent variables the category’s rate per kilometer, the monthly fuel price
Price of 95-oct. fuel (CL$/lt) 673 916 767 778 62 (95-octane gasoline for category 1 and diesel for categories 2 and 3), and
Price of diesel (CL$/lt) 383 688 529 546 84 the monthly IMACEC separated by mining and non-mining sectors.1 This
IMACEC 90.6 120.6 103.9 104.5 5.7
distinction of economic activity is relevant because non-mining is the
IMACEC (Non-Mining) 91.2 121.6 104.0 105.0 6.1
IMACEC (Mining) 77.0 113.6 99.6 99.6 7.1 main activity in Santiago, and not doing so could bias its effect (the
preliminary results confirmed this). We control the impact of the year’s

Table 3
Correlation between independent variables.
Tariff per kilometer Cat. 1 (CL$) 1
Tariff per kilometer Cat. 2 (CL$) 0.945 1
Tariff per kilometer Cat. 3 (CL$) 0.908 0.995 1
Price of 95-oct. fuel (CL$/lt) − 0.119 0.045 0.092 1
Price of diesel (CL$/lt) − 0.187 0.004 0.061 0.925 1
IMACEC 0.476 0.404 0.374 − 0.235 − 0.288 1
IMACEC (Non-Mining) − 0.124 − 0.102 − 0.093 0.220 0.218 0.395 1
IMACEC (Mining) 0.509 0.432 0.399 − 0.274 − 0.329 0.994 0.290

variable, there is a compromise between granularity and estimation season on the flows by including eleven dummy variables to represent
power. The more specific (in space and time) the flow of vehicles you the months. In addition, we included the number of periods with a high
want to estimate, the less power the data will have. After various ana­ rate and the number of periods with a saturation rate. We add the lag of
lyses, we conclude that the best dependent variable to use corresponds the monthly flow (in log) to capture the persistence of the demand and
to the monthly flow per vehicle category and for each gantry. We reject the users’ lag in internalizing changes in the highway toll and fuel price.
disaggregating the dependent variable by the period of the day. Omitting the fixed effects per gantry, Equation (1) describes the model
The model’s variables can be expressed in level (linear model in the for each category:
variables) or logarithm (logarithmic model). The second option was
chosen for three reasons. First, and conceptually, the logarithmic form of
the models avoids problems of truncation of the dependent variable.
Namely, since the flow cannot take negative values, the random
component (or random error or error term) underlying the models is also
unable to do so. When using logarithmic models, this condition is 1
The base tariff of each gantry was also included, but the models based on
imposed on random components and avoids bias problems in the
this variable did not deliver satisfactory results.

237
M. Batarce et al. Transport Policy 133 (2023) 234–241

Fig. 2. Evolution of traffic on the highway. Fig. 6. Evolution of the base tariff outside of the peak for all categories
of vehicle.

ln Qcpt = α1 ln Rct + α2 ln Ctc + α3 ln Mt + α4 ln NMt + α5 Ppt + α6 Spt + ρ ln Qcp(t− 1)

∑
11
+ γ m dmi
i=1

(1)

where Qcpt represents the flow of category c through the gantry p in the
month t, Rct is the tariff per kilometer, Cct is the fuel price, Mt IMACEC of
mining sector and NMt of non-mining sector, Ppt and Spt represents the
number of periods of the day with a peak or saturation tariff, and dmj are
dummy variables that represent each month (except for April, used as a
reference) and equal 1 when i = m y 0 otherwise. The parameters of
interest are α1 , ..., α6 , and ρ because they correspond to those variables
Fig. 3. Evolution of categories 2 and 3 traffic. that change with time and are the most relevant for predicting flows or
the analysis of the impact of tariffs and economic growth, for example.
These parameters are differentiated by vehicle category, although they
do not differ in the notation used.
The lag of the monthly flow produces effects that propagate in time.
This means an increase in flow in one month will affect the following
months. How persistent this effect is depends on the value of the coef­
ficient associated with the lag, which we denote by ρ. The long-term
effect of the exogenous variables is then given by αi /(1 − ρ). In this
way, we can estimate the short-run and long-run elasticities.
It is worth noting that when using the lagged dependent variable as a
regressor in the within-group estimation approach, the error component
correlates with this variable. Consequently, the model’s parameters are
biased because of endogeneity. This bias reduces with the length of the
panel as it is equal to − (1 +ρ)/T (Nickell, 1981; Hahn and Kuersteiner,
Fig. 4. Evolution of the total IMACEC, mining IMACEC, and non- 2002), where ρ is the parameter of the lagged variable (see Equation
mining IMACEC (1)). In our models, the bias is around 0.02 because T is relatively large.
As shown below, this bias is smaller than the parameter’s standard error,
and we consider it unnecessary to correct. Moreover, we estimate the
model using indirect inference (Gourieroux et al., 1993; Gouriéroux
et al., 2010) to reduce endogeneity bias, resulting in almost the same
parameters (see Appendix A).

4. Results of the estimation

Table 4 shows the model estimation result for each vehicle category
according to Equation (1).2 All variables are introduced in logarithms,
which allow a straightforward interpretation of their parameters. The
parameters of the fixed and seasonal effects are omitted in this table
since, although they help in estimation, their values are not informative.
The goodness-of-fit indicator is high.
Among the economic variables that explain the evolution of vehicle
Fig. 5. Evolution of diesel and 95-octane gasoline prices.
flow, the only unexpected result is the statistical insignificance of the

2
The model estimation and the heteroskedasticity and autocorrelation tests
were carried out using the R package plm (Croissant and Millo, 2008).

238
M. Batarce et al. Transport Policy 133 (2023) 234–241

Table 4 Table 5
Estimation results. Summary of elasticities.
Category 1 Category 2 Category 3 Light vehicles Buses and trucks Trucks over 2 axles

Estimate t-test Estimate t-test Estimate t-test Short Long Short Long Short Long
term term term term term term
Tariff per − 0.8189 − 13.49 − 0.6768 − 3.42 − 0.6623 − 2.34
kilometer Tariff per − 0.82 − 1.29 − 0.68 − 0.96 − 0.66 − 1.15
(in log) kilometer
Price of − 0.2272 − 12.89 − 0.0871 − 3.16 − 0.0127 − 0.34 Fuel price − 0.23 − 0.36 − 0.09 − 0.12 − 0.01 − 0.02
gasoline/ Non-Mining 3.60 5.66 2.74 3.89 2.15 3.72
diesel (in IMACEC
log) Mining − 0.22 − 0.34 − 0.10 − 0.14 − 0.01 − 0.02
Non-Mining 3.5953 29.89 2.7432 5.36 2.1492 2.87 IMACEC
IMACEC (in
log)
Mining − 0.2152 − 15.25 − 0.0993 − 3.04 − 0.0113 − 0.24 be explained by the still significant differences in motorization rates. In
IMACEC (in other words, a marginal increase in economic activity (and so in income)
log)
Flow lag (in 0.3645 7.47 0.2956 7.11 0.4229 6.80
in Chile generates a much larger increase in the motorization rate, and
log) therefore in the flow, than in countries that already have high motori­
Peak periods − 0.0030 − 3.94 − 0.0041 − 3.21 − 0.0019 − 1.01 zation rates.
Saturation − 0.0098 − 3.41 − 0.0119 − 3.35 − 0.0136 − 5.52 The flow behavior in the short term is inelastic to the tariff per
periods
kilometer for all three vehicle categories, with an elasticity value
R2 of within- 0.683 0.547 0.490 ranging from 0.6 to 0.8 (in absolute value). This result is higher than that
group model
obtained in other studies in Chile, where the elasticity is less than 0.47
Overall 3020.4 (<2e- 3136.0 (<2e- 1999.4 (<2e-
significance 16) 16) 16)
(e.g., Saens and Lobos, 2013; de Grange et al., 2015). It is also lightly
F (p-value) above the results obtained for Spanish highways by Matas and Raymond
Observations 2041 2041 2041 (2003). From the point of view of highway concessionaires, increasing
the rate per kilometer will increase their revenue, even though demand
will decrease. In other words, the higher income due to the higher price
diesel price in trucks with more than two axles (category 3). The effect of
dominates the loss of income due to the lower flow.4 This result explains
mining IMACEC is low regarding the non-mining IMACEC. Moreover,
why urban highway concessionaires in Santiago annually increase their
the mining IMACEC effect is negative, showing the increase of economic
rates by the maximum their contracts allow. In the long term, demand
activity far from Santiago leads to reduce the use of the highway (and
becomes less inelastic. In the case of light vehicles and heavy trucks, it is
possibly the movements in the entire city). This reduction may be
barely above one, causing highway concessionaires to lose market
explained by increased labor traveling out of Santiago during growing
power. However, the demand loss due to tariff increment is compen­
mining activity.3
sated by the significant effect of economic growth in the long term.
As for inference, we estimate the standard errors using Arellano’s
In all categories, the flow shows inelastic behavior to the fuel price.
(1987) approach, which allows a fully general structure regarding het­
Indeed, for light vehicles, the elasticity is − 0.2; for trucks with two axles
eroskedasticity and serial correlation. This estimator is consistent when
and buses, it is − 0.09; for trucks with more than two axles, it is 0.01
the number of individuals grows and the number of periods remains
(although not statistically significant). These values, which show high
fixed. We use robust standard errors because serial correlation and
inelasticity in the short term, are within the ranges observed interna­
heteroskedasticity tests reject the hypothesis of uncorrelated and ho­
tionally for the short term (one or two years). For example, the meta-
moscedastic errors. We tested heteroskedasticity with Pesaran’s (2004,
analysis by Dunkerley et al. (2014) places this elasticity at around
2015) test for panel cross-sectional dependence. The null hypothesis of
− 0.2. We should note that the short-term effect we estimate is one
no cross-sectional dependence was rejected in all vehicle categories. We
month, and the long-term effect would correspond to one year or more.
used the Breusch-Godfrey test for serial correlation in panel models
In this way, the estimated fuel-price long-term impact could compare to
(Wooldridge, 2002). The null hypothesis of no serial correlation in
the short-term one of international studies. In any case, the long-term
idiosyncratic errors also was rejected for all categories.
elasticity estimated by us is consistent with those studies, and the flow
behavior is inelastic.
5. Analysis
The effect of the number of periods with peak and saturation tariffs is
significant, as expected, for the three categories of vehicles (except the
In the estimated models, the parameters directly represent the elas­
number of peak periods for heavy trucks). If the number of periods with
ticities. Table 5 summarizes the elasticities of the most relevant
tariffs higher than the base increases, the flow decreases. It decreases
variables.
more when the periods with the saturation tariff increase. However, the
The primary determinant of flow on the highway is economic
effect – although statistically significant – is minimal in all cases. Indeed,
growth, measured by the non-mining IMACEC. This variable presents a
heavy vehicles exhibit the largest elasticity to the number of periods
short-term elasticity value of 3.6 for the case of light vehicles, 2.7 for
with a saturation tariff, which only reaches − 0.014. In the case of light
buses and trucks, and 2.2 for heavy trucks. These elasticities are higher
vehicles, this same elasticity reaches just − 0.010. Thus, given that the
than those obtained by Matas and Raymond (2003) for Spanish high­
tariff increases from base to peak and from peak to saturation are high,
ways ranging from 0.9 to 1.4. In other developed countries, the elasticity
but the elasticity is small, increasing the number of periods with peak or
to income is also smaller than those we found, varying between 1 and
1.7 depending on the study. However, the income is not directly com­
parable with IMACEC or GDP. The long-term elasticities we found may 4
Indeed, if we consider that Q(P) is the flow (in total kilometers) according to
the tariff per kilometer P, revenue is given by Q(P) P. If we then add P, revenue
( )
variates in accordance with d[(Q(P)P]
dP = Q(P) 1 − 1ε , where ε is the price elasticity
3
With our data, we cannot verify this assertion, but there is anecdotal evi­ of demand. From this equation we can determine that revenue variation will be
dence of temporal migration to work in mining, where shifts are scheduled in positive while the Price elasticity of demand will be such that |ε| < 1, which is
periods of two weeks. precisely the result found in the short-term estimations.

239
M. Batarce et al. Transport Policy 133 (2023) 234–241

saturation tariffs has a lesser effect on the highway flow. This low in vehicle flows on urban highways, tenders must consider this effect on
elasticity indicates that tolls are a weak tool for managing demand on the auction mechanism and tariff policies. For instance, the auction may
highways. use a score combining the net present value of income (Engel et al.,
2001) with aspects related to risk, investment amount, financing
6. Conclusions structure, etc.

In this article, we present the results of estimating the elasticity of the Author statement
flow of an urban highway to various variables. We study the effect of
tariffs, fuel prices, and economic activity. These results contribute to the Marco Batarce: Conceptualization, Methodology, Data curation,
knowledge of the factors determining the demand for trips on urban Validation, Estimation, Formal analysis, Writing- Reviewing and Edit­
highways, particularly in underdeveloped countries. For example, a ing. Franco Basso: Conceptualization, Formal analysis, Data curation,
notable difference between the previous results, which focus on the US Writing- Reviewing and Editing, Visualization. Leonardo J. Basso:
and European countries, is that economic growth has twice the effect on Supervision, Conceptualization, Formal analysis, Writing- Reviewing
the flow in Chile than the studies reported thus far. We also detect that and Editing, Project administration, Funding acquisition.
economic activity has different effects by sector and that the results can
be severely biased if this differentiation is not included in the econo­ Data availability
metric model.
Another finding is that in the short term (one month), the flow is The authors do not have permission to share data.
inelastic to the per-kilometer tariff, but in the long term (one year), the
flow becomes almost perfectly elastic. This result appears to contradict Acknowledge
the behavior of urban highway concessionaires who, in the case of
Santiago, annually increase rates by the total permitted by the contracts. This study was partially funded by a grant from Instituto Sistemas
However, the flow growth is dominated by economic growth, which in Complejos de Ingeniería (ANID PIA/APOYO [AFB180003 and
turn implies an increase in the rate of motorization and car use. This AFB220003]), and a grant from Science, Technology, Knowledge, and
result is a warning for the design of concession contracts in developing Innovation Ministry of Chile (FONDECYT Project 11200167). We also
countries. To the extent that economic growth is the determining factor thank Autopista Central for sharing the data for this study.

APPENDIX

This appendix presents the models estimated with indirect inference (II), free of endogeneity bias (Gouriéroux et al., 2010). The bias introduced by
the within-group (WG) estimator is lower than − 0.02 in all cases. The standar errors of WG estmates (S.E. WG) show that the sampling error is greater
than the possible bias due to endogeneity.

Table A.1
Comparison of indirect inference and withing-group estimates

Variable Category 1 Category 2 Category 3

II WG S.E. WG II WG S.E. WG II WG S.E. WG

Flow lag 0.379 0.365 0.049 0.310 0.296 0.042 0.443 0.423 0.062
Tariff per kilometer − 0.827 − 0.819 0.061 − 0.670 − 0.677 0.198 − 0.660 − 0.662 0.283
Price of gasoline/diesel − 0.227 − 0.227 0.018 − 0.088 − 0.087 0.028 − 0.014 − 0.013 0.037
Non-Mining IMACEC 3.592 3.595 0.120 2.708 2.743 0.511 2.139 2.149 0.748
Mining IMACEC − 0.215 − 0.215 0.014 − 0.095 − 0.099 0.033 − 0.012 − 0.011 0.047
Peak periods − 0.003 − 0.003 0.001 − 0.004 − 0.004 0.001 − 0.002 − 0.002 0.002
Saturation periods − 0.010 − 0.010 0.003 − 0.012 − 0.012 0.004 − 0.013 − 0.014 0.002
April − 0.184 − 0.183 0.006 − 0.089 − 0.090 0.033 − 0.060 − 0.059 0.050
August 0.009 0.009 0.003 0.052 0.050 0.010 0.059 0.059 0.016
December − 0.413 − 0.413 0.020 − 0.318 − 0.322 0.080 − 0.266 − 0.268 0.116
January − 0.034 − 0.035 0.008 0.033 0.032 0.015 0.051 0.049 0.020
February 0.065 0.065 0.011 0.105 0.106 0.012 0.132 0.132 0.017
July 0.015 0.014 0.003 0.037 0.036 0.009 0.046 0.043 0.015
June − 0.025 − 0.026 0.003 0.006 0.005 0.009 0.005 0.005 0.016
March − 0.225 − 0.227 0.011 − 0.099 − 0.103 0.047 − 0.010 − 0.012 0.073
May − 0.105 − 0.105 0.005 − 0.043 − 0.045 0.023 − 0.027 − 0.027 0.035
November − 0.203 − 0.203 0.010 − 0.101 − 0.103 0.043 − 0.082 − 0.083 0.062
October − 0.008 − 0.008 0.006 0.044 0.042 0.018 0.057 0.055 0.029

References Basso, F., Basso, L.J., Bravo, F., Pezoa, R., 2018. Real-time crash prediction in an urban
expressway using disaggregated data. Transport. Res. C Emerg. Technol. 86,
202–219.
Arellano, M., 2003. Panel Data Econometrics. Oxford University Press.
Basso, F., Pezoa, R., Varas, M., Villalobos, M., 2021. A deep learning approach for real-
Arellano, M., 1987. Computing robust standard errors for within-groups estimators. Oxf.
time crash prediction using vehicle-by-vehicle data. Accid. Anal. Prev. 162, 106409.
Bull. Econ. Stat. 49 (4), 431–434.
Croissant, Y., Millo, G., 2008. Panel data econometrics in R: the plm package. J. Stat.
Bakhat, M., Labandeira, X., Labeaga, J.M., López-Otero, X., 2017. Elasticities of transport
Software 27 (2).
fuels at times of economic crisis: an empirical analysis for Spain. Energy Econ. 68,
de Grange, L., González, F., Troncoso, R., 2015. Estimates of price elasticity of demand
66–80. https://doi.org/10.1016/j.eneco.2017.10.019.
for urban freeway use with high-frequency control variables: the case of Santiago,

240
M. Batarce et al. Transport Policy 133 (2023) 234–241

Chile. Appl. Econ. 47 (22), 2326–2337. https://doi.org/10.1080/ Matas, A., Raymond, J.L., 2003. Demand elasticity on tolled motorways. J. Transport.
00036846.2015.1005825. Stat. 6 (2/3), 91–208.
Dunkerley, F., Rohr, C., Daly, A., 2014. Road traffic demand elasticities: a rapid evidence Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica 49 (6),
assessment. In: Road Traffic Demand Elasticities: A Rapid Evidence Assessment. 1417–1426.
RAND Corporation. https://doi.org/10.7249/RR888. Olszewski, P., Xie, L., 2002. Traffic demand elasticity with respect to road pricing – some
Engel, E.M., Fischer, R.D., Galetovic, A., 2001. Least-present-value-of-revenue auctions evidence from Singapore. Proceedings of the International Conference on Seamless and
and highway franchising. J. Polit. Econ. 109 (5), 993–1020. Sustainable Transport, November 217–228.
Gourieroux, C., Monfort, A., Renault, E., 1993. Indirect inference. J. Appl. Econom. 8 Oum, T.H., Waters II, W.G., Fu, X., 2007. Transport demand elasticities. J. Transport.
(S1), S85–S118. Eng. 127 (6), 239–255. https://doi.org/10.1108/9780857245670-012.
Gouriéroux, C., Phillips, P.C., Yu, J., 2010. Indirect inference for dynamic panel models. Pesaran, M.H., 2004. General Diagnostic Tests for Cross-Section Dependence in Panels.
J. Econom. 157 (1), 68–77. Institute for the Study of Labor (IZA). IZA Discussion Paper No. 1240).
Goodwin, P., Dargay, J., Hanly, M., 2004. Elasticities of road traffic and fuel Pesaran, M.H., 2015. Testing weak cross-sectional dependence in large panels. Econom.
consumption with respect to price and income: a review. Transport Rev. 24 (3), Rev. 34 (6–10), 1089–1117.
275–292. https://doi.org/10.1080/0144164042000181725. Saens, R., Lobos, G., 2013. Elasticidad precio de la demanda por autopistas interurbanas
Graham, D.J., Glaister, S., 2002. The demand for automobile fuel: a survey of elasticities. en Chile, vol. 79. Lecturas de Economía. https://doi.org/10.17533/udea.le.n79a5.
J. Transport Econ. Pol. 36 (1), 1–26. Taiebat, M., Stolper, S., Xu, M., 2019. Forecasting the impact of connected and
Hahn, J., Kuersteiner, G., 2002. Asymptotically unbiased inference for a dynamic panel automated vehicles on Energy use: a microeconomic study of induced travel and
model with fixed effects when both n and T are large. Econometrica 70 (4), Energy rebound. Appl. Energy 247, 297–308. https://doi.org/10.1016/j.
1639–1657. apenergy.2019.03.174.
Litman, T.A., 2021. Understanding Transport Demands and Elasticities: How Prices and Wooldridge, J.M., 2002. Econometric Analysis of Cross–Section and Panel Data. MIT
Other Factors Affect Travel Behavior, vols. 1–76. Victoria Transport Policy Institute. Press.
http://www.vtpi.org/elasticities.pdf.

241