https://revistas.ucr.ac.cr/index.php/ingenieria/index
www.ucr.ac.cr / ISSN: 2215-2652
ENERO/ JUNIO 2023 - VOLUMEN 33 (1)
Ingeniería 33(1): 22-33, Enero-Junio, 2023. ISSN: 2215-2652. San José, Costa Rica
DOI 10.15517/ri.v33i1.50910
Esta obra está bajo una Licencia de Creative Commons. Reconocimiento - No Comercial - Compartir Igual 4.0 Internacional
Understanding the Determinants of Fuel Demand in Costa Rica,
1965-2019: Price, Income, and Registered Vehicles
Hacia una mejor comprensión de los determinantes de la demanda
por combustibles en Costa Rica, 1965-2019: precio, ingreso y ota
vehicular
Eduardo Pérez Molina
University of Costa Rica, San José, Costa Rica.
email: eduardo.perezmolina@ucr.ac.cr
ORCID: https://orcid.org/0000-0002-7730-9677
Recibido: 30 de abril de 2022 Aceptado: 23 de setiembre 2022
Abstract
Fuel demand during the last 50 years in Costa Rica has increased constantly. Do the price of fuels
and Gross National Income contribute to explain these trends? This paper explores the existence of causal
relations between economic growth, fuel price, and transport demand (represented by fuel consumption
and registered vehicles). Vector autoregression (VAR) models were estimated with a time series of data of
1965-2019. Causal relations were found between fuel demand and income, but not with registered vehicles.
particular, possible taxation of fuel demand as a carbon mitigation strategy).
Keywords:
Fuel demand, fuel price, income, registered vehicles, VAR, Granger causality.
Resumen
La demanda de combustible durante los últimos 50 años en Costa Rica se incrementó constantemente.
¿Cuánto contribuyen el ingreso nacional bruto y el precio de los combustibles a explicar estas tendencias?
Este artículo explora la existencia de relaciones causales entre crecimiento económico, precio de combustibles
documentaron los efectos del precio sobre la demanda e ingreso, lo cual es muy relevante para la formulación
de políticas públicas (en particular, posibles impuestos a los combustibles como estrategia de mitigación de
emisiones de carbono).
Palabras Clave:
-
sivos, causalidad de Granger.
Ingeniería 33(1): 22-33, Enero-Junio, 2023. ISSN: 2215-2652. San José, Costa Rica DOI 10.15517/ri.v33i1.50910
23
1. INTRODUCTION
The microeconomic analysis of fuel demand has long been considered a subject of importance
[1].
Along this line of reasoning, fundamental research questions have been related to the price
been adopted to the impacts of infrastructure investment and transport costs on development [2].
Fuel demand in Costa Rica has been regularly studied over the last three decades. Generally,
[3], [4],
[5], and government analysis to inform energy sector planning [6], [7] —of strategic importance,
research has resulted in elasticity estimates derived from aggregate demand using time series
econometrics [3], [4], [6], [7]. Blackman et al. [5]
data to conclude the tax on gasoline is progressive and the tax on diesel, regressive in Costa
et al. [8]
in Costa Rica for the energy and transport sector, based on stakeholder input, which they
could be reduced by up to 87 % of the baseline by a combination of modal shift, technology,
and demand management.
[3] for gasoline (-0.33 of price and 0.47
of income) and for diesel (-0,20 of price and 0,33 of income) demand during 1972-1992, after
controlling for vehicle ownership. Adamson [4] used a time series of 1957-1996 to estimate,
for gasoline and diesel, elasticities of price (-0.26 and -0.18, respectively) and income (1.41
[3] nor Adamson [4]
elasticities between fuel prices. Leiva [7], in his most recent energy elasticity estimates for Costa
Rica, used data of 1984-2007 to calculate price and income elasticities for diesel demand (-0.14
estimates only included income elasticity (1.05) because the price elasticity was found to be
positive, likely because of substitution of premium gasoline during high price periods.
This paper reports a time series analysis of fuel demand, Gross National Income (GNI) per
capita (as a proxy variable for household income), and gasoline price. The stationarity of all-
time series was explored, followed by cointegration of all series with a common integration
order. Granger causality was determined. A vector autoregression (VAR) model is reported to
The analysis completed in this paper extends previous work on aggregate fuel demand in
two directions. Firstly, it makes use of longer time series of fuel demand (for the 1965-2019
period, encompassing 55 years), this allows for more rigorous modeling of serial correlation
than previous work [3], [4], [7]
variables. In particular, the developed model extends previous work by considering income as
PÉREZ: Understanding the Determinants of Fuel Demand in Costa Rica, 1965-2019...
24
potentially endogenous –unlike models reported in [3], [4], [7], all of which assumed exogeneity
of price, income, and vehicle ownership.
2. METHODOLOGY
2.1. Data on gasoline demand and price, vehicle ownership, and income in Costa Rica,
1965-2019
Following previous work and, more generally, the literature on the relation between transport
demand and income (e.g. [9], [10]), a series of equations were proposed to explain the relationship
between transport activity (fuel demand and vehicle ownership), fuel price (which is also a good
(1)
(2)
(3)
(4)
with QT
t
the yearly fuel demand per capita (gasoline or diesel), Fleet
t
the number of registered
vehicles per capita, P
t
the fuel price per liter, and IND
t
and μ
i,t
the random error term. The model formed by this set of equations was estimated for
annual data of Costa Rica, 1965-2019.
1
Gasoline price and Gross National Income per capita
Fleet
t
consuming most diesel fuel for transport in Costa Rica (data on the number of vehicles per fuel
is not available in national statistics, requiring the use of these proxy variables). All data series
were transformed into natural logarithms.
TABLE I
2
Variable Mean
Standard
deviation
Minimum Maximum
Gasoline demand per capita [m
3
per person]
141,9 68,8 54,0 266,4
Diesel demand per capita [m per person]
193,9 50,7 93,9 285,9
Registered cars per capita [vehicles per 1000 persons]
78,2 54,0 15,1 193,8
Registered trucks and buses per capita [vehicles per 1000 persons]
32,7 31,1 8,1 51,0
Gasoline price [colones/litre]
617,3 200,0 324,0 1111,1
Diesel price [colones/litre]
439,5 206,5 125,6 880,9
Gross National Income per capita
5645 1914 3023,0 9557,0
1
Fuel demand data were compiled by RECOPE; price data come from statistical compilations of the Dirección Sectorial de Energía, MINAE; gross
national income and international oil prices were taken from World Bank data; regisered vehicle gures are published annually by the MOPT.
2
Real CRC₡, 2020=100; real USD$, 2010=100.
Ingeniería 33(1): 22-33, Enero-Junio, 2023. ISSN: 2215-2652. San José, Costa Rica DOI 10.15517/ri.v33i1.50910
25
Fig. 1. Trends of fuel demand per capita, total registered vehicles per capita, fuel price, and Gross
National Income per capita (1965-2019).
fuel demand per capita to be caused (in the sense of Granger) by other variables, (b) price to
be exogenous, i.e., not caused by other variables, (c) price should be a negative determinant of
both fuel demand and Gross National Income per capita, and (d) Gross National Income per
capita to be a positive determinant of gasoline demand.
Descriptive statistics and trends of the time series are reported in TABLE I and Fig. 1.
display a dip in 1980, a year of sharp economic crisis in Central America) and registered trucks
buses, there is in Fig. 1 a relatively smooth increase from 1965 until the early 2000s, after which
there is a decrease followed by an increase. On the contrary, the number of cars per capita shows
a remarkable growth of over ten times over the 1965-2019 period.
variables —and indeed one could argue they do not exhibit the increasing trend of fuel demand,
with, and likely respond to, the mean price of oil in the international markets (Fig. 1). Additionally,
it is also apparent diesel prices have been larger than the mean towards the end of the study
PÉREZ: Understanding the Determinants of Fuel Demand in Costa Rica, 1965-2019...
26
2.2. Econometric strategy
The quantitative model exploring the endogeneity of the time series, proposed in subsection
time series, y
t
, is said to be caused by another time series, x
t
y
t
can be better predicted by
considering the past of both y
t
and x
t
rather than only the past of y
t
[11]
(5)
(6)
The hypothesis test H0: b
1
= b
2
= … = b
r
= 0 vs. H
A
“not H
0
”, tests if xt does not cause y
t
(in
H
0
d
1
= d
2
= … = d
r
= 0 vs. H
A
“not H
0
” tests
if y
t
does not cause x
t
. Rejecting the null hypothesis implies the existence of Granger causality.
When time series have integration orders greater than 0, conventional Granger causality
tests may present problems due to the inapplicability of conventional asymptotic theory [12].
[12], VAR models can be generated for known orders of
when data have small samples and (b) often, as in the present paper, interest is not on forecasting
[12] suggested
Granger causality should be tested on the levels of a VAR, regardless of integration order, but
they proposed to determine the maximum integration order, I(k), of all time series in the model
and to increase the amount of lags in the VAR by k. These results would be used to perform the
k additional lags would not be included in the test itself.
Thus, the Granger causality tests applied to the set of time series described in subsection A
in each time series, from which their order of integration, I(k), was determined (these results are
r
o
with (r
o
+k)
r
o
lags (i.e., with [(v-1)·r
o
] degrees of freedom, with v the number of variables in the model),
cointegration between the data series was explored (TABLE IV), because cointegration implies
causality, although the absence of cointegration does not necessarily reject it.
relations between variables were explored.
Ingeniería 33(1): 22-33, Enero-Junio, 2023. ISSN: 2215-2652. San José, Costa Rica DOI 10.15517/ri.v33i1.50910
27
3. RESULTS AND DISCUSSION
properties of the time series themselves, the most basic of which is their integration order. TABLE
existence of unit roots in the data.
TABLE II
3
Variable
Level First dierence
Trend Drift Trend Drift
Gasoline demand per capita -1.770 -0.778 -4.309* -4.354*
Diesel demand per capita -2.819 -1.914 -5.121* -5.137*
Registered cars per capita -2.357 -1.191 -6.098* -5.987*
Registered trucks and buses per capita -1.976 -2.476 -3.453
+
-3.116
+
Gasoline price -2.292 -2.258 -5.094* -5.069*
Diesel price -2.035 -1.822 -4.562* -4.571*
Gross National Income per capita -2.402 -0.569 -5.138* -5.190*
* p < 0.01,
+
p < 0.05
The results from TABLE II clearly show (a) log-transformed levels of the variables considered
(the weakest ADF test statistic corresponds to the trucks and buses time series, but it is still
[13].
[12] approach to Granger causality
testing, i.e., estimating a VAR model on the levels of time series (despite the levels not being
stationary per the results of TABLE II), and adding as many lags as the maximum order of
integration of all-time series involved (in the present case, one additional lag, as all-time series
were found to be of order 1). To estimate this VAR model on the levels, the number of lags
gasoline and diesel models), (2) verifying the both models correspond to stable VAR processes,
(3) revising if a greater number of lags might mitigate serial correlation in the error terms
lag but increasing the number of lags exacerbates the problem). It is important to note that the
VAR model for diesel demand excludes the number of registered vehicles (trucks and buses)
because, regardless of the number of lags selected, the resulting models were not stable.
As can be seen in TABLE III, both gasoline and diesel demand are Granger caused by other
between income (Gross National Income per capita), fuel price, and fuel demand are supported
by the model. However, it is noteworthy that the variable cars per capita is not caused by other
3
All variables expressed as natural logarithms.
PÉREZ: Understanding the Determinants of Fuel Demand in Costa Rica, 1965-2019...
28
potential determinants, in particular by the income proxy variable, as mode choice has been
linked to household income (e.g. in origin/destination travel surveys, [14]).
TABLE III
4
Dependent variable
(eect)
Determinants (causes) X
2
(d.f.) Prob.
Gasoline model
Gasoline demand per
capita
Registered cars per capita, gasoline price, Gross
National Income per capita
36.155 (3)* <0,001
Registered cars per
capita
Gasoline demand per capita, gasoline price,
Gross National Income per capita
2.072 (3) 0.558
Gasoline price
Gasoline demand per capita, registered cars per
capita, Gross National Income per capita
1.521 (3) 0.677
Gross National
Income per capita
Gasoline demand per capita, registered cars per
capita, gasoline price
10.993 (3)
+
0.012
Diesel model
Diesel demand per
capita
Diesel price per capita, Gross National Income
per capita
7.932 (2)
+
0.019
Diesel price
Diesel demand per capita, Gross National
Income per capita
2.016 (2) 0.365
Gross National
Income per capita
Diesel demand per capita, diesel price 10.994 (2)* 0.004
* p < 0.01,
+
p < 0.05.
A combination of nonstationary series may be said to be stationary in which case they are
called cointegrated and they share a common stochastic trend. Cointegration is evidence of Granger
causality. The stationary linear combination may be interpreted as a long-run equilibrium [9].
Cointegration, in sum, is an interesting property to explore if present in the data. To detect the
[13]
Eigenvalue statistic for the gasoline fuel VAR model suggests the existence of one cointegration
display any evidence of cointegration. It should be noted that estimating the (unreported)
correction term for gasoline demand and Gross National Income per capita as dependent variables,
4
VAR models on the levels were estimated with 4 lags (optimal VAR was determined to have 3 lags; it was augmented by 1 following Toda and
Yamamoto [12]; Wald tests do not include coecients of this additional lag).
Ingeniería 33(1): 22-33, Enero-Junio, 2023. ISSN: 2215-2652. San José, Costa Rica DOI 10.15517/ri.v33i1.50910
29
TABLE IV
Null
hypothesis
Alternative
hypothesis
Trace statistic
(prob.)
Eigenvalue
statistic (prob.)
Gasoline demand per capita, registered cars per capita, gasoline
price, Gross National Income per capita
r > 3 1.157 (0.282) 1.157 (0.282)
r > 2 7.789 (0.828) 3.633 (0.887)
r > 1 16.654 (0.673) 11.863 (0.573)
r = 0 r > 0 44.232 (0.104) 27.579 (0.047)
+
Diesel demand per capita, diesel price, Gross National Income
per capita
r > 2 0.006 (0.936) 0.006 (0.936)
r > 1 8.034 (0.469) 8.028 (0.384)
r = 0 r > 0 20.330 (0.412) 12.296 (0.532)
+
p < 0.05.
TABLE V
Determinant
(cause)
Gasoline models Diesel models
ΔGasoline
demand
ΔPrice ΔGNI ΔFleet
ΔDiesel
demand
ΔPrice ΔGNI
Intercept 0.038* -0.010 0.016* 0.056* -0.046 0.027 0.018*
ΔQT
t-1
-0.166 0.312 -0.009 -0.004 -0.046 0.225 -0.010
ΔP
t-1
-0.401* 0.162 -0.102* -0.037 -0.236* 0.120 -0.100*
ΔGNI
t-1
0.660
+
-0.232 0.349* 0.100 0.456 -0.462 0.253
+
ΔFleet
t-1
-0.320 0.302 -0.020 -0.244
+
— — —
Adjusted R
2
0.324 <0.01 0.219 0.007 0.135 <0.01 0.290
f stat. 7.241 0.718 4.637 1.094 3.703 0.463 8.085
d.f. 4 & 48 4 & 48 4 & 48 4 & 48 3 & 49 3 & 49 3 & 49
Prob. <0.001 0.584 0.003 0.371 0.018 0.709 <0.001
* p < 0.01,
+
p < 0.05
PÉREZ: Understanding the Determinants of Fuel Demand in Costa Rica, 1965-2019...
30
f statistics.
seem in Fig. 1) and high fuel prices reduce available income.
(negative) and Gross National Income per capita (positive). The only unexpected result is
per capita gasoline demand.
Diesel demand in TABLE V is a much weaker model than gasoline demand, which also
transport. Indeed, previous results also coincide with the results of TABLE V in identifying diesel
price and income elasticities as less than the corresponding elasticities for gasoline [3], [4], [7].
[3]
in fuel price, which may be induced by a tax, reduces per capita demand (as evidenced by the
congestion [3]
public transport (the buses that supply it are dependent on diesel and, because of it, diesel price
[5]) and on freight. Furthermore,
demand constraints imposed by greater price should have limited repercussions on household
income and prosperity.
is available [16]
demand [16]
Ingeniería 33(1): 22-33, Enero-Junio, 2023. ISSN: 2215-2652. San José, Costa Rica DOI 10.15517/ri.v33i1.50910
31
metropolitan region is part) is also 1.19 times greater than the country average [17]
congestion has also been found to be fundamentally centered in Costa Rica [18]. In synthesis, a
there.
The analysis presented in this paper corresponds to a long time series of data —55 years,
between 1965 and 2019. During this period, the Costa Rican economy has undergone various
also have drastically changed the trajectory of the Costa Rica economy. These changes may have
introduced structural breaks into the time series, thus requiring controlling for them. The same
the scope of the models reported in this paper, but it may be a fruitful area for further research
(including a greater frequency of three-monthly data in lieu of annual time series).
4. CONCLUSIONS
The analysis executed and reported in this paper shows evidence of causal relations between
fuel demand and Gross National Income per capita. Time series including gasoline demand and
Gross National Income per capita were found to present one cointegration relation (common
stochastic trend) whereas the diesel fuel VAR resulted in no cointegration relations. Fuel price
was found to be weakly exogenous.
The results of the developed models for gasoline and diesel demand are, on the whole,
consistent with previous analysis of aggregate fuel demand in Costa Rica [3], [4], [6], [7].
of these policies. Fuel demand management is an attractive area for carbon mitigation, as most
fuel consumption (of gasoline and diesel) is explained by transport demand [15], but it also may
provides a quantitative methodological framework which may be used to assess the impacts of
PÉREZ: Understanding the Determinants of Fuel Demand in Costa Rica, 1965-2019...
32
should be concentrated.
5. ROLE OF THE AUTHORS
ACKNOWLEDGEMENTS
The author gratefully acknowledges relevant feedback from two anonymous referees and
improvements to this paper.
REFERENCES
environmental degradation for 75 countries,” Transport Res D-Tr E, vol. 53
10.1016/j.trd.2017.04.031.
J
Transp Geogr
Fuel Taxes and Urban Air Pollution in Developing Countries: The Case of
Costa Rica
Energy
Policy
Rica, 1992.
socioeconomic impacts in Costa Rica,” Energy Strategy Rev
esr.2020.100573.
Transport Res D
Appl Econ, vol.
Ingeniería 33(1): 22-33, Enero-Junio, 2023. ISSN: 2215-2652. San José, Costa Rica DOI 10.15517/ri.v33i1.50910
33
Am J
Polit Sc
processes,” J Econometrics
New Introduction to Multiple Time Series Analysis
Rica, 2019.
