Transportation Research Part A 63 (2014) 56–66

Contents lists available at ScienceDirect

Transportation Research Part A

journal homepage: www.elsevier .com/locate / t ra

Asymmetric responses of highway travel demand to changesin fuel price: An explanation via fuel price uncertainty

http://dx.doi.org/10.1016/j.tra.2014.02.0200965-8564/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author. Tel.: +82 53 950 7438; fax: +82 53 950 5429.E-mail addresses: yjkwon@kookmin.ac.kr (Y. Kwon), jm064@knu.ac.kr (J. Lee).

1 Tel.: +82 2 910 4530; fax: +82 2 910 5209.

Yongjae Kwon a,1, Jaimin Lee b,⇑a College of Business Administration, Kookmin University, 136-702 Seoul, Republic of Koreab School of Economics and Trade, Kyungpook National University, 702-701 Daegu, Republic of Korea

a r t i c l e i n f o

Article history:Received 6 February 2013Received in revised form 27 December 2013Accepted 16 February 2014Available online 31 March 2014

Keywords:Fuel pricesHighway travel demandUncertaintyAsymmetryRealized volatility

a b s t r a c t

Previous research has examined asymmetric effects of fuel price uncertainty on energydemand. If we consider that energy demand is related to travel demand, the changes in fuelprices may have asymmetric effects on highway travel demand via fuel price uncertainty.In other words, when in general fuel price is steadily rising, the highway traffic volumedecreases by a small percentage. On the other hand, the highway traffic volume increasesby a large percentage when fuel prices are falling. We hypothesize that the uncertainty infuel prices generates this kind of asymmetric effect on highway traffic volume in Korea. Weuse the Korean monthly fuel price and highway traffic volume data from 2001 to 2009, anddefine the intra-month (weekly) fuel price changes as monthly fuel price volatility which isa proxy for monthly fuel price uncertainty. We found that the direction of the change infuel prices had asymmetric effects on highway travel demand and that the fuel priceuncertainty led drivers to respond asymmetrically to the changes in fuel prices.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

For example crude oil prices rose sharply in the first half of 2008 but plummeted in the second half because of the globalfinancial crisis. Korea’s economy has shown some signs of economic recovery, and subsequently there have been indicationsfor the rise in prices of oil products. This paper hypothesizes that fluctuations in fuel prices increase the fuel price uncer-tainty which in turn has significant effects on behaviors of drivers of passenger vehicles and freight transport firms.

Many studies have shown that changes in fuel prices have considerable influence on behaviors of economic agents or oninvestment and consumption decisions. Hamilton (1983) verified that changes in fuel prices had a significant negative causalrelationship with the US real GNP growth from 1948 to 1980. Mork (1989) found that GNP growth had a significant negativerelationship with increases in fuel prices.

Some studies observed that fuel price uncertainty had significant effects on economic variables such as investment andconsumption. Bredin et al. (2008) and Elder and Serletis (2009) measured oil price uncertainty by using a multivariateGARCH-in-mean model and found that it had a significant negative effect on the industrial production of the G-7 countriesand Canada.

Guo and Kliesen (2005) analyzed the relationship between oil price volatility and the macroeconomic activity of the US byusing quarterly realized oil price volatility, which is the sum of the squared changes in daily prices in a quarter year. They

Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66 57

found that oil price volatility had a significant negative effect on future GDP growth. Cong et al. (2008) also employedmonthly oil price volatility constructed with the average of squared first-log differences in daily oil prices and estimatedthe relationship between oil price shocks and the Chinese stock market.

Kuper and van Soest (2006) used the conditional variance of oil prices from GARCH (1,1) for oil price uncertainty andfound evidence that this uncertainty had an asymmetric effect on energy use and rendered energy-saving technologies lessattractive.2 They claimed that there is a non-zero probability for changes in fuel price reversals to take place as long as fuel priceuncertainty exists. They argued that the higher the level of fuel price uncertainty, the greater the magnitude of changes in fuelprice reversals. That is, they argued that it is price uncertainty that drives the asymmetric effects of fuel price changes on energyuse.

Gately and Huntington (2002) estimated the effects of changes in oil prices on energy and oil demand, and showed thatthe increases and decreases in oil prices had the asymmetric effects on oil demand. Similarly, Dargay and Gately (1997)found that consumers did not respond in the same fashion to rising and falling prices and that fuel demand is not necessarilyreversible to price changes.

Many studies have investigated the relationship between fuel price changes and traffic volume in the transport sector.Dargay et al. (2002), Graham and Glaister (2002, 2004), Goodwin et al. (2004), and Hanly et al. (2002) summarized a negativeeffect of the gasoline prices on traffic volume. That is, an increase in gasoline prices leads the demand for gasoline in thetransport sector to decline, and this demand decrease leads the traffic demand to decrease.

In addition to studies focusing on the relationship between fuel prices and traffic volume, there have been some articles toestimate the relationship between fuel prices and fuel demand in transport sector (Bonilla and Foxon, 2009; de Freitas andKaneko, 2011; Hughes et al., 2006).

However there have been few articles to investigate the relationship between energy price uncertainty and traffic volumelevels. In particular, no study has examined the asymmetric effects of energy price changes on traffic volume.

We analyze three issues surrounding the relationship between fuel price changes and traffic volume. First, our study ana-lyze the effects of fuel price uncertainty on traffic volume. More specifically, the question of whether fuel price uncertaintyhas a significant negative effect on traffic volume is addressed. Second, the asymmetric effects of fuel price changes on trafficvolume are investigated. Third, the first and second are combined to determine whether fuel price uncertainty would induceasymmetric effects of fuel price changes on traffic volume.

Considering the asymmetric responses of agents to price changes, the authors expect that fuel price uncertainty wouldinduce asymmetric responses of drivers to fuel price increases or decreases and that changes in transport volume from fuelprice increases are different from those from fuel price decreases.

In our study, we investigate the effects of fuel price uncertainty on highway travel demand. The results show that fuelprice uncertainty in conjunction with asymmetric behaviors of drivers in response to changes in fuel prices produces asym-metric changes in traffic volume from changes in fuel prices. More specifically, because of the effect of fuel price uncertaintyon the behaviors of drivers, increases in highway traffic volume during periods of fuel price decreases exceed decreases dur-ing periods of fuel price increases.

In this paper, fuel price uncertainty is defined as reversals in fuel price changes (Kuper and van Soest, 2006). If there is anincrease in the level of fuel price uncertainty, then reversals are more likely. In this case, highway commuters are less likelyto respond instantaneously to fuel prices increases because they expect fuel price decreases sooner or later. In addition, thistype of fuel price uncertainty is expected to have an asymmetric effect in the case of fuel prices increases or decreases.

Our study is based partly on Kuper and van Soest (2006), who estimated the relationship between oil price uncertainty andenergy use by using the GARCH (1,1) model to determine fuel price uncertainty. However, our study uses realized volatility.3

Korean Monthly data on highway traffic volume and fuel prices from January 2001 to December 2009 are used in ourstudy. Intra-month (weekly) fuel price changes are used to construct monthly fuel price uncertainty. In the present approach,realized volatility is a model free volatility estimator, whereas GARCH type estimators depend upon the type of model.Therefore, the present approach is more general and robust than that of Kuper and van Soest (2006).

The rest of this paper is organized as follows. Section 2 provides the theoretical background and the model specification.Section 3 describes the data. Section 4 presents the estimation results and their implications, and Section 5 concludes with asummary and limitations.

2. Theoretical background and the model specification

2.1. Theoretical background

2.1.1. Fuel prices with uncertaintyMany studies have investigated the socioeconomic factors influencing demand in the context of road traffic volume. Dar-

gay et al. (2002), Graham and Glaister (2002), and Hanly et al. (2002) summarized the relationship between demand in road

2 Bernanke (1983) and Pindyck (1991) claimed that agents would postpone their decision making or engage in conservative economic activity because ofirreversible characteristics of investment or expenditure plans.

3 Section 2.2 explains why conditional volatilities is replaced with realized volatility.

Table 1Variables in ATFM.

Variables

Demand measures Passengers kilometers, vehicle kilometers, vehicle stocksSupply and service measures Road kilometersMonetary cost components Purchasing and maintenance costs, tax and insurance, and fuel pricesOther variables relating to cars Fuel efficiency of carsOther variables Population, household income, GDP, etc.

Sources: Dargay et al. (2002), Development of an Aggregated Transport Forecasting Model, University College London.

58 Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66

traffic volume and various economic variables and suggested various elasticities in road traffic volume with respect to eco-nomic factors. Dargay et al. (2002) discussed the Aggregated Transport Forecasting Model (ATFM) and explained the rela-tionship between aggregated road traffic volume and various economic factors by using passenger kilometers, vehiclekilometers, and the vehicle stock as demand measures for a proxy for road traffic volume and defining road kilometers assupply and service measures. In addition, they classified purchasing and maintenance costs, tax and insurance costs, and fuelprices as components of the monetary cost and automobile fuel efficiency, the population, household income, and GDP asother variables (see Table 1).

Because the aggregated data on Korean highways are used for our study, the definitions of variables in Dargay et al. (2002)are used. The number of vehicle trips and that of vehicle kilometers on Korean highways are used as variables for travel de-mand. The fuel price per liter, GDP, and Korean highway road kilometers are used as the explanatory variables. Motoringcosts such as purchasing and maintenance costs, tax and automobile insurance costs, and fuel efficiency are excluded be-cause of data availability. GDP is used as a proxy for the population4 and household income, and Korean highway road kilo-meters are a proxy for the total vehicle capacity.

In Dargay et al. (2002), the ATFM employed time series data and a partial adjustment model or an error correction model(ECM). Because road traffic volume, fuel prices, and GDP are non-stationary variables, they may be co-integrated. Actually, acointegration test of the variables of interest is conducted using the Engle–Granger type cointegrating test, and the test re-sults show cointegrating relationships for some variables but not for others.5 Given these mixed results, ECM is not employed.In addition, the present paper study focuses on asymmetric effects of fuel prices on road travel volume and examine whetherthe effects of rising fuel prices on road travel volume are different from those of declining fuel prices. Dargay et al. (2002),Graham and Glaister (2002), Hanly et al. (2002), Bonilla and Foxon (2009), and de Freitas and Kaneko (2011) used ECM inthe transport context, but did not consider asymmetric effects of fuel prices on transport demand. Only Granger and Lee(1989) examined the asymmetric effects of some economic variables (e.g. production and the volume of sales) on othereconomic variables such as changes in investment but still did not estimate the asymmetric effects of some level or differencedvariables on dependent variables. Finally, the present paper employs monthly data from January 2001 to December 2009.Hughes et al. (2006) argued that the long-run relationship or long-run elasticities derived from ECM would not be appropriatefor models using short term monthly data, and therefore the present paper difference the level variables and use first-differencedvariables in the analysis.

The basic estimation model, motivated by Guo and Kliesen (2005), is

4 Mo5 The6 All

d ln ht ¼ a0 þ a1d ln pt þ a2zt þ a3d ln GDPt þ a4d ln xt þ et ð1Þ

h is the number of vehicles or that of vehicle-kilometers on the highway, p the fuel prices, z the fuel price uncertainty, GDPthe gross domestic product, x the length of the highway adjusted for the number of lanes, d the difference notation, ln thenatural log, e the error terms and t is the time (year and month).

In this model, Korean fuel prices, not crude oil prices, are used. The authors isolate the effects of fuel prices on highwaytravel demand from those of exchange rates and taxes. Korean fuel prices are related more to driver behaviors than crude oilprices because of different factors such as exchange rates and taxes.

The number of vehicle trips and that of vehicle-kilometers on the highway are used as a proxy for highway travel demand,and GDP is used to control for economic booms or recesses, which have considerable influence on highway travel demand.The expansion length of Korean highways is used as an explanatory variable. In Korea, new highways are constructed whileexisting highways are widened and therefore, the length of the highway (km) adjusted for the numbers of lanes is used in themodel. To construct the variable, the weight 0.5, 1.0, 1.5, and 2.0 are assigned to two, four, six, and eight lane roads, respec-tively. As of 2009, the length and ratio of four lane roads were 2510 km and 71.80% respectively, and these roads formed thelargest group of roads (www.ex.co.kr). Therefore, four lane roads serve as the reference lane road in the analysis.

Log variables are differenced because all variables have unit roots and are non-stationary.6 Eq. (1) is to estimated to exam-ine whether fuel price changes and uncertainty have significant negative effects on highway travel demand. Previous studies

nthly population data have been provided by the Korean government since 2008.results of the cointegration test are available from the authors upon request.

the variables are I(1) variables according to the unit root test. The results of the unit root test are available from the authors upon request.

Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66 59

have found that increases in energy prices or uncertainty reduce energy use and GDP growth. For example, Guo and Kliesen(2005) showed that oil price uncertainty had a significant negative effect on GDP growth.

Section 4.1 investigates whether fuel prices have similar effects on highway travel demand and whether increases in fuelprices and uncertainty reduce highway travel demand.

2.1.2. Fuel price asymmetry with or without uncertaintyEq. (1) fails to find any asymmetric effects of fuel prices on highway travel demand. To find such effects, Eqs. (2) and (3),

which address the effect of fuel price uncertainty on energy use, are employed.7

7 The

d ln ht ¼ b0 þ b1d ln pt þ b2jd ln pt j þ b3d ln GDPt þ b4d ln xt þ ut ð2Þ

d ln ht ¼ c0 þ c1d ln pt þ c2ztjd ln ptj þ c3d ln GDPt þ c4d ln xt þ v t ð3Þ

where u and v are error terms. Eq. (2) considers asymmetric responses to fuel price increases and decreases without control-ling for fuel price uncertainty and postulates asymmetric responses to fuel price increases and decreases from the elasticitydecomposition (Lee et al., 2009). Therefore, fuel price uncertainty is not included to explain asymmetric responses of driversto fuel price increases or decreases. Here b1 and b2 are expected to be negative and positive, respectively, to capture asym-metric responses to fuel price changes with no fuel price uncertainty. When the proportionate change in fuel prices rises byone unit, the proportionate change in the highway traffic volume changes by (b1 + b2) unit. On the other hand, when the pro-portionate change in fuel prices declines by one unit, the proportionate change in the highway traffic volume increases by(|b1| + b2) unit. If asymmetric responses of drivers are added to changes in fuel prices, then the absolute value of (|b1| + b2)may exceed that of (b1 + b2).

In this paper, fuel price uncertainty is expected to have no independent effect on highway traffic volume, but when com-bined with drivers’ asymmetric responses to fuel price changes, it is expected to amplify the asymmetric responses of high-way traffic volume. Eq. (3) reflects fuel price uncertainty in conjunction with asymmetric characteristics of highway trafficvolume with respect to fuel price changes. Because fuel price uncertainty induces drivers to react asymmetrically to fuelprice decreases or increases, the interaction term for fuel price uncertainty, zt, with the absolute value of the difference ofthe natural log of the fuel price is included in Eq. (3).

Here the coefficients, c1 and c2 are expected to be negative and positive, respectively. When the proportionate change infuel prices rises by one unit, the proportionate change in the highway traffic volume changes by (c1 + ztc2) unit. On the otherhand, when the proportionate change in fuel prices declines by one unit, the proportionate change in the highway trafficvolume increases by (|c1| + ztc2) unit. If responses of drivers to fuel price changes are asymmetric, then the absolute valueof (|c1| + ztc2) exceeds that of (c1 + ztc2).

2.2. Measure of uncertainty

Kuper and van Soest (2006) measured oil price volatility by using the conditional standard deviation produced by univar-iate GARCH models. GARCH-type models are constructed through a combination of constant, past conditional variance, andpast squared return residuals under the assumption that the variance of the current innovation is a function of past innova-tions. Using GARCH models seems harmless because they are widely accepted as a standard and known to be useful for cap-turing typical characteristics of financial volatilities, including persistence and clustering (Engle and Patton, 2001).

However, the conditional variance or conditional standard deviation may not be an appropriate measure of volatility inempirical research because conditional volatility does not contain information on what occurs at certain time intervals. Theconditional variance is merely a fitted value from a time series model. Using fitted values in empirical research can raisesome doubt about reported results because it is difficult to determine whether the chosen model is the best one. In short,the conditional variance depends on the type of model used for the estimation.

This potential problem can be avoided by using model-free volatility estimators. For example, squared or absolute returnscan be good alternatives in that they are independent of any particular models. However, return series do not contain enoughinformation to reflect what happens in the market because they consider only opening and closing prices. To overcome thelimitations of return-based measures, previous studies have suggested new estimators.

For example, Parkinson (1980), Garman and Klass (1980), and Rogers and Satchell (1991) proposed volatility estimatorsconstructed using the highest and lowest values within a certain time interval. These estimators, generally referred to as ex-treme value estimators, were also employed by French et al. (1987) and Schwert (1989), who used daily stock returns tomeasure monthly volatility. With the availability of tick-by-tick financial data to the public, researchers have suggested vol-atility estimators based on intraday price quotes (Andersen and Bollerslev, 1998; Andersen et al., 2001, 2003).

To avoid potential problems in using the conditional variance, weekly fuel prices are used to construct monthly fuel pricevolatility. In fact, several studies have used the model-free volatility of energy prices. For example, Guo and Kliesen (2005)and Cong et al. (2008), examined the relationship between oil prices and the macro-economy by adopting the realized

se equations are proposed in Kuper and van Soest (2006).

60 Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66

variance for a better measure of monthly or quarterly oil price volatility. Following the aforementioned study, monthly fuelprice volatility is constructed by aggregating squared weekly changes in fuel prices.

8 We9 Kor

and tru10 Thi

times bsuch as

11 GD12 KRW

zt ¼Xn

i¼1

r2t;i ð4Þ

where r2t;i is the squared weekly return in the i-th week in month t, and zt is fuel price volatility in month t, which is a proxy

for fuel price uncertainty. Because this measure is just the aggregation of multiple changes in prices, it is not based on anytime series model and free from any specification error.

3. Data

Monthly highway traffic volume, fuel prices, and GDP are used in our study. The intra-month sum of squared weekly re-turns from January 2001 to December 2009 is used for fuel price uncertainty.

The number of vehicle trips and that of vehicle-kilometers on the Korean highway are used as a proxy for highway traveldemand.8 In addition, these numbers are used for the type I vehicles (mainly passenger cars).9

Monthly price data are used for three refined petroleum products; gasoline, diesel, and LPG. These monthly prices arefinal consumer prices, including the valued added tax and the Korean energy taxes. Amounts of gasoline, diesel, and LPG con-sumed in the road transport sector are provided by Petronet (www.petronet.co.kr). Product prices for each are multiplied bythe ratio of the amount for each petroleum product used to that of all petroleum products used (Ferderer, 1996).

Finally, nominal fuel prices are converted into real fuel prices by adjusting for inflation based on the consumer price index(CPI) with 2005 = 100 as the reference year.

WOPy;m ¼X3

i¼1

wiy;mOPi

y;m ð5Þ

WOPy,m is the weighted fuel prices in month m of year y, OPiy;m the price of product i in month m of year y, wi

y;m the ratio of theamount of product i used to the amount of total petroleum products used in month m of year y, i the gasoline, diesel, and LPG,y the year (2001,2001, . . . ,2009) and m is the months (1,2, . . . ,12).

GDP is used as an explanatory variable because economic conditions can affect highway travel demand. However, theBank of Korea does not provide a monthly GDP but yearly or quarterly GDP. Quarterly GDP is transformed into monthlyGDP by using the Industrial Production Index, which is announced monthly by the National Statistical Office of Korea.Monthly GDP is calculated by

Monthly GDPy;m ¼ Quarterly GDPy;q �IPIy;mP3

i¼1IPIy;m;i

ð6Þ

IPI is the Industrial Production Index; y, q, and m are the year, quarter, and month.To capture the supply-side effect of constructing new highways and expanding existing ones, a variable is controlled for

such that it represents the present highway length. To construct this variable, weights of 0.5, 1.0, 1.5, and 2.0 are assigned totwo-, four-, six-, and eight-lane roads, respectively, before the variables for the lane length are added together.10 In this way,the effects of extending and widening highways are controlled for. Highway traffic volume has seasonality, and therefore, thisvariable is seasonally adjusted through the census X12 model in E-Views.11

In December 2009, the numbers of total vehicle trips and that of the type I vehicle trips are 109,600,000 and 89,571,000,respectively. Actually, the number of type I vehicle trips accounts for more than 80% of the number of total vehicle trips onthe highway. The number of total vehicle-kilometers and that of type I vehicle-kilometers are 4,203,099,000 and3,240,734,000, respectively. The number of type I vehicle-kilometers accounts for less than 80% of the number of total vehi-cle-kilometers because the average travel mileage of type I vehicles is less than that of other types of vehicles (see Table 2).

The real weighted fuel price in December 2009 is KRW 1217.18,12 and South Korea’s real GDP in December 2009 is KRW85,446 billion. The length of the highways (km) adjusted for the width of Korea is 4012.7 km. Fuel price uncertainty in Decem-ber 2009 is 0.000087.

As shown in Fig. 1, changes in the natural log of monthly fuel prices show a significant level of heteroskedasticity, which isa stylized fact of financial volatilities. A comparison of Figs. 1 and 2 clearly shows that the larger the change in the fuel price,the higher the level of fuel price uncertainty.

follow the notations and explanations in Lee et al. (2009) to construct the dependent and explanatory variables.ea Expressway Corporation classifies all vehicles into five types. Type II vehicles are small buses and light trucks; type III vehicles, medium sized busescks; and type IV and V vehicles, three and four axle trucks, respectively.s study considers no tolls or fares in the Korean highway system, because they do not change frequently. More specifically, tolls have changed only threey Korea Expressway Corporation since 2000. Even with tolls or fares controlled for, their estimates do not reflect their effects but those of the price levelthe CPI.

P and industrial production index are also adjusted seasonally.denotes the Korean Won, the currency unit for Korea. The KRW/USD exchange rate in December 2009 was 1166.45.

Table 2Data summary (December 2009).

Number of total vehicle trips(in the thousands)

Number of the type I vehicle trips(in the thousands)

Number of total vehicle-kilometers(in the thousands)

Number of the type I vehicle-kilometers (in the thousands)

109,600 89,571 4,203,099 3,240,734

Real weighted fuel prices(KRW)

Real GDP (billion KRW) Highway length adjusted for width(km)

Fuel price uncertainty

1217.18 85,446 4012.7 0.000087

Sources: 1. The Bank of Korea (http://www.bok.or.kr). 2. The Korea Expressway Corporation. 3. The National Statistical Office (http://www.kostat.go.kr/).4. The Petronet (www.petronet.co.kr).

-.100

-.075

-.050

-.025

.000

.025

.050

.075

.100

2001 2002 2003 2004 2005 2006 2007 2008 2009

DLNP

Fig. 1. Changes in the natural log of the monthly fuel prices.

.000

.001

.002

.003

.004

.005

.006

.007

.008

2001 2002 2003 2004 2005 2006 2007 2008 2009

Z

Fig. 2. Monthly fuel price uncertainty.

Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66 61

There were large fuel price volatilities in July 2001, 2002, 2003, and 2004 because of increases by South Korea’s centralgovernment in transport tax rates imposed on diesel in those time periods.13 The South Korean government altered the energytax scheme from 2001 to 2004, and transport tax rates increase for diesel. Incidentally, there were increases in transport taxrates for diesel in July 2001, 2002, 2003, and 2004, resulting in increases in fuel price changes and fuel price uncertainty.

During the first half of 2008, there was a sharp increase in crude oil prices because of speculative demand for crude oil andincreased oil demand from developing countries. During the second half of 2008, however, crude oil prices plummeted as aresult of the global financial crisis.

Fuel price uncertainty is compared under fuel price increases and decreases. Fig. 3 shows the regression lines of fuel priceuncertainties for fuel price changes. Fig. 3 illustrates that an increase in fuel prices induces the series more volatile than adecrease.

As shown in Fig. 3, the magnitude of price change reversals during fuel price increases exceeds that during fuel pricedecreases. This suggests that drivers are more likely to speculate a price change reversal when fuel prices are rising than

13 South Korea’s transport tax is a type of excise tax on gasoline and diesel. It is a revenue source for constructing transport infrastructure such as roads,railroads, airports, and harbors. In the case of LPG, the Korean government imposes excise tax.

-0.07 -0.02 0.03 0.08

dlnp

0.000

0.002

0.004

0.006

z

Fig. 3. Monthly fuel price uncertainty and changes.Notes: The graph displays the relationship between the monthly returns and monthly volatility of fuel prices. The plotted line is a spline-smoothing curve,which shows the asymmetry of the fuel price uncertainty with respect to changes in the price.

TabEsti

d

z

d

d

C

NRFAS

***

** S* S

a

b

numc

62 Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66

declining. A rise in fuel price is not likely to reduce travel demand to the same extent as a decline in fuel prices increasestravel demand. This suggests asymmetric responses of highway traffic volume to fuel price changes.

4. Estimation results

4.1. Fuel price uncertainty without asymmetric effects

Table 3 shows the estimation results for highway travel demand with respect to fuel price changes and uncertainty with-out considering asymmetric effects. The percentage change in the real GDP has a positive effect on that in highway trafficvolume. However, this effect is statistically significant only when the number of total vehicle trips is used as a dependentvariable. The percentage change in the length of highways adjusted for the number of lanes has a significant positive effecton that in highway traffic volume. As Korean Expressway Corporation has extended the length or lanes of its highways, it isnatural that highway traffic demand increases in numbers.

The percentage change in fuel prices has a significant negative effect on that in highway travel demand. However, fuelprice uncertainty has a significant positive effect on the percentage change in highway travel demand, which is contrary

le 3mation results without asymmetric effects.

Number of totalvehicle trips

Number of the type Ivehicle trips

Number of totalvehicle-kilometers

Number of the type Ivehicle-kilometers

lnpt �0.1104* �0.0988** �0.2830*** �0.3450**

(0.0560)a (0.0448) (0.1000) (0.1527)t 3.6274*** 3.3044** 5.2505** 5.8355*

(1.1682) (1.3236) (2.0674) (3.4395)lnGDPt 0.2219*** 0.1203 0.1194 0.0255

(0.0826) (0.0811) (0.1752) (0.2621)lnxt 0.4504*** 0.4146*** 0.5081*** 0.4923***

(0.1502) (0.1275) (0.1519) (0.1727)onstant 0.0003 0.0012 �0.0006 �0.0001

(0.0018) (0.0019) (0.0032) (0.0044)

o. of observationsb 108 108 108 1082 0.1911 0.1195 0.0829 0.0543-statistic 6.0829*** 3.4952** 2.3268* 1.4784ICc �4.8715 �4.7011 �3.4826 �2.8309Cc �4.7474 �4.5770 �3.3585 �2.7068

Significance at 1% significance level.ignificance at 5% significance level.

ignificance at 10% significance level.The numbers in the parenthesis are Newey–West standard errors.To avoid the loss of the first observation in return calculation, fuel price data for December 2000 was added to the original dataset. By doing so, theber of observations, 108, was intact.

AIC and SC represent the Akaike Information Criteria and Schwarz Criteria respectively.

Table 4Estimation results without the fuel price uncertainty.

Number of total vehicletrips

Number of the type I vehicletrips

Number of total vehicle-kilometers

Number of the type I vehicle-kilometers

dlnpt �0.0733 �0.0661 �0.2309** �0.2874**

(0.0589)a (0.0470) (0.0952) (0.1368)|dlnpt| 0.0118 0.0211 0.0338 0.0400

(0.0896) (0.1035) (0.1643) (0.2244)dlnGDPt 0.2047** 0.1069 0.0980 0.0023

(0.0879) (0.0875) (0.1823) (0.2687)dlnxt 0.4485*** 0.4125*** 0.5047*** 0.4884***

(0.1509) (0.1286) (0.1551) (0.1765)Constant 0.0024 0.0028 0.0021 0.0028

(0.0025) (0.0028) (0.0044) (0.0058)

No. ofobservationsb

108 108 108 108

R2 0.1580 0.0945 0.0634 0.0414F-statistic 4.8312*** 2.6878** 1.7442 1.1125AICc �4.8314 �4.6731 �3.4617 �2.8174SCc �4.7073 �4.5490 �3.3375 �2.6932

* Significance at 10% significance level.*** Significance at 1% significance level.** Significance at 5% significance level.

a The numbers in the parenthesis are Newey–West standard errors.b To avoid the loss of the first observation in return calculation, fuel price data for December 2000 was added to the original dataset. By doing so, the

number of observations, 108, was intact.c AIC and SC represent the Akaike Information Criteria and Schwarz Criteria respectively.

Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66 63

to the author’s expectation and the findings of Guo and Kliesen (2005), Bredin et al. (2008), and Elder and Serletis (2009). Theestimation results show that highway travel demand increases with an increase in fuel price uncertainty. The close relation-ships of travel demand to energy use and GDP growth suggest a negative effect of fuel price uncertainty on highway traveldemand, and therefore Eq. (1) fails to sufficiently explain the relationship between fuel price uncertainty and highway traveldemand.

4.2. Asymmetric effects without fuel price uncertainty

Table 4 represents the estimation results for highway travel demand with respect to fuel price changes without fuel priceuncertainty. The percentage change in the real GDP has a positive effect on that in highway traffic volume. However, thiseffect is not statistically significant except for the number of total vehicle trips. The percentage change in the length of high-ways adjusted for the number of lanes has a significant positive effect on that in highway traffic volume.

As expected, the percentage change in fuel prices and the absolute value of this percentage change have negative andpositive effects, respectively, on the percentage change in highway travel demand. However, no coefficients for the percent-age change in fuel prices are statistically significant in the four cases. In addition, no coefficients for the absolute value of thepercentage change in fuel prices are statistically significant. This suggests that Eq. (2) is not an appropriate model for rep-resenting the asymmetric characteristic of highway traffic volume with respect to changes in fuel prices.

4.3. Asymmetric effects with fuel price uncertainty

As mentioned earlier, Kuper and van Soest (2006) found that asymmetric responses of energy demand to changes in oilprices are due to oil price volatility. If oil price volatility influences asymmetric responses of energy demand, then it is verylikely that travel demand asymmetrically responds to changes in prices because of the volatility.14 To test this hypothesis, thespecifications proposed in Kuper and van Soest (2006) are applied.

According to the results, the effects of the percentage change in real GDP and the length of highways adjusted for thenumber of lanes on the percentage change in highway travel demand are similar to those in Tables 3 and 4 with respectto their signs, significance, and magnitude (see Table 5).

The percentage change in fuel prices has a significant negative effect on highway traffic volume. As expected, a decrease(increase) in fuel prices increases (decreases) highway traffic volume. The interaction terms of fuel price uncertainty and theabsolute value of the percentage change in fuel prices have a significant positive effect on the percentage change in highwaytravel demand. This suggests that an increase in fuel price uncertainty induces drivers to react more asymmetrically to fuelprice increases or decreases.

14 As long as drivers pay for the gas, this point of view is worth being investigated.

Table 6The proportionate change of the highway travel demand with respect to the proportionate change of the fuel prices.

Number of total vehicle trips Number of the type I vehicles trips

Fuel price increase Fuel price decrease Fuel price increase Fuel price decrease

�0.0909 0.1520 �0.0818 0.1394

Number of total vehicle-kilometers Number of the type I vehicle-kilometers

Fuel price increase Fuel price decrease Fuel price increase Fuel price decrease

�0.2528 0.3344 �0.3114 0.4019

Notes: The average value of 0.000646 for zt was used.

Table 5Estimation results with the fuel price uncertainty.

Number of totalvehicle trips

Number of the type Ivehicles trips

Number of totalvehicle-kilometers

Number of the type Ivehicle-kilometers

dlnpt �0.1215** �0.1106** �0.2936*** �0.3566**

(0.0587) a (0.0467) (0.1045) (0.1591)zt|dlnpt| 47.2521*** 44.6069*** 63.1666** 70.0487*

(15.4920) (16.0706) (25.7250) (40.8295)dlnGDPt 0.2353*** 0.1337 0.1351 0.0429

(0.0841) (0.0833) (0.1811) (0.2716)dlnxt 0.4547*** 0.4187*** 0.5137*** 0.4985***

(0.1504) (0.1275) (0.1530) (0.1739)Constant 0.0011 0.0018 0.0008 0.0013

(0.0016) (0.0017) (0.0029) (0.0038)

No. of observationsb 108 108 108 108R2 0.1892 0.1198 0.0790 0.0517F-statistic 6.0084*** 3.5058** 2.2094* 0.2380AICc �4.8692 �4.7015 �3.4785 �2.8282SCc �4.7450 �4.5773 �3.3543 �2.7040

*** Significance at 1% significance level.** Significance at 5% significance level.* Significance at 10% significance level.

a The numbers in the parenthesis are Newey–West standard errors.b To avoid the loss of the first observation in return calculation, fuel price data for December 2000 was added to the original dataset. By doing so, the

number of observations, 108, was intact.c AIC and SC represent the Akaike Information Criteria and Schwarz Criteria respectively.

64 Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66

When a proportionate change in fuel prices increases by one unit, a proportionate change in total vehicle trips and that intype I vehicle trips decrease by 0.0909- and 0.0818-unit, respectively.15 However, when a proportionate change in fuel pricesdecreases by one unit, a proportionate change in total vehicle trips and that in type I vehicle trips increase by 0.1520- and0.1394-unit, respectively.16 The absolute change in highway travel volume when fuel prices decrease exceeds that when fuelprices increase (see Table 6).

With the number of vehicle-kilometers used as a dependent variable, similar results are observed. When a proportionatechange in fuel prices increase by one unit, a proportionate change in total vehicle-kilometers and that in type I vehicle-kilometers decrease by 0.2528- and 0.3114-unit, respectively. When a proportionate change in fuel prices decrease byone unit, a proportionate change in total vehicle-kilometers and that in type I vehicle-kilometers increase by 0.3344- and0.4019-unit, respectively.

Graham and Glaister (2002) and Hanly et al. (2002) provided the symmetric elasticities of vehicle trips or vehicle trip kilo-meters with respect to fuel prices by using estimates derived from previous reports. Graham and Glaister (2002) showed thatthe short run elasticities of vehicle trips and that of vehicle trip kilometers were �0.16 and �0.15, respectively. Hanly et al.(2002) reported that the short run elasticities of vehicle trip kilometers was �0.10. Our estimates for the elasticities ofvehicle trips are consistent with these findings, but those for the elasticities of vehicle trip kilometers are higher. However,this paper provides elasticity estimates under both fuel price increases and decreases, which is the major contribution for ourstudy.

15 0.0909 unit and 0.0818 unit decreases are calculated as (�0.1215 + 0.000646 � 47.2521) and (�0.1106 + 0.000646 � 44.6069), respectively.16 0.1520 unit and 0.1394 unit increases are calculated as (0.1215 + 0.000646 � 47.2521) and (0.1106 + 0.000646 � 44.6069), respectively.

Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66 65

When fuel prices increase, fuel price uncertainty weakens the negative effect of fuel price changes on highway travel de-mand, reflecting an asymmetric characteristic of highway travel demand with respect to changes in fuel prices. This suggeststhat drivers consider a price change reversal to be more likely when fuel prices increase than when they decrease.

The elasticities of type I vehicle trips with respect to changes in fuel prices are lower than those of total vehicle trips.However, the elasticities of type I vehicle trip mileage are higher than those of total vehicle trip mileage. This implies thatdrivers of passenger vehicles are more likely to respond to changes in fuel prices in terms of trip mileage than in terms ofactual trips.

Finally, even when fuel prices change by the same amount, a decrease in fuel prices has a greater effect on highway trafficvolume than an increase because of drivers’ asymmetric behavior in response to fuel price changes derived from fuel priceuncertainty.

5. Conclusions

We find that fuel price uncertainty has a significant effect on highway travel demand, which induces drivers to respondasymmetrically to fuel price changes. In addition, fuel price uncertainty has a significant effect on asymmetric responses ofhighway travel demand to changes in fuel prices.

These empirical results have two important policy implications. First, fuel taxes may not be effective in reducing fuel usein uncertain times because people respond to fuel price uncertainty asymmetrically. In uncertain times, for example, traveldemand may not decrease by much even when fuel prices rise because people expect fuel prices to decrease sooner or later.However, travel demand is likely to increase sharply when fuel prices drop because people expect fuel prices to increasesoon. In this regard, fuel taxes may not be an effective measure for reducing fuel demand. Second, reducing and managinguncertainty may be effective in promoting the policy agenda (as mentioned in Kuper and van Soest (2006)) Oil-importingcountries such as Korea have to secure stable energy sources for oil, shale gas, coal, and alternative and renewable formsof energy, among others, and develop transport methods for using alternative and renewable forms of energy, to reduceand manage fuel price uncertainty.

This paper makes two important contributions to the literature. First, many studies have focused on the relationships offuel price changes to energy use, investment, and other economic factors, whereas this paper addresses transport demandwith respect to changes in fuel prices and fuel price uncertainty. Second, the paper employs realized volatility for fuel priceuncertainty instead of using the conditional standard deviation for prices. Kuper and van Soest (2006) used the GARCH (1,1)model to obtain the conditional standard deviations for fuel prices as a proxy for fuel price uncertainty. However, GARCHmodels may not be suitable for this kind of study because they produce volatility measures conditional on past information.Simply put, a measure of volatility at a certain time interval should keep all the variation in the time interval. We can use avolatility measure in empirical studies only when this condition is satisfied. By adopting the concept of the realized volatilityproposed in French et al. (1987) and Schwert (1989), we could construct a measure of fuel price volatility suitable for empir-ical studies. With this measure, we could avoid potential problems associated with using model-dependent volatilitymeasures.

However, this paper considers short-term data from 2001 to 2009 and weekly fuel prices instead of daily fuel prices toobtain monthly fuel price uncertainty, which is a limitation. We look forward to seeing future research that utilizes moreadequate and elaborate data and that deals with various transport modes.

References

Andersen, T., Bollerslev, T., 1998. Answering the skeptics: yes, standard volatility models do provide accurate forecasts. Int. Econ. Rev. 39, 885–905.Andersen, T., Bollerslev, T., Diebold, F.X., Ebens, H., 2001. The distribution of stock return volatility. J. Financ. Econ. 61, 43–76.Andersen, T., Bollerslev, T., Diebold, F.X., Labys, P., 2003. Modeling and forecasting realized volatility. Econometrica 71, 579–626.Bernanke, B.S., 1983. Irreversibility, uncertainty, and cyclical investment. Quart. J. Econ. 98, 85–106.Bonilla, D., Foxon, T., 2009. Demand for new Ca fuel economy in the UK, 1970–2005. J. Transp. Econ. Policy 43, 55–83.Bredin, D., Elder, J., Fountan, S., 2008. The Effects of Uncertainty about Oil Prices in G-7, UCD Business Schools. Working Paper No. 08/08, UCD.Cong, R.G., Wei, Y.M., Jiao, J.L., Fan, Y., 2008. Relationships between oil price shocks and stock market: an empirical analysis from China. Energy Policy 36,

3544–3553.Dargay, J., Gately, D., 1997. The demand for transportation fuels: imperfect price-reversibility? Transp. Res. Part B 31, 71–82.Dargay, J., Goodwin, P., Hanly, M., 2002. Development of an Aggregated Transport Forecasting Model (ATFM). University of London, Centre for Transport

Studies, ESRC Transport Studies Unit, London.de Freitas, L.C., Kaneko, S., 2011. Ethanol demand under the flex-fuel technology regime in Brazil. Energy Econ. 33, 1146–1154.Elder, J., Serletis, A., 2009. Oil price uncertainty in Canada. Energy Econ. 31, 852–856.Ferderer, P.J., 1996. Oil price volatility and the macroeconomy. J. Macroecon. 18, 1–26.Engle, R.F., Patton, A.J., 2001. What good is a volatility model? Quant. Finance 1, 237–245.French, K., Schwert, W.G., Stambaugh, R.F., 1987. Expected stock returns and volatility. J. Financ. Econ. 19, 3–29.Garman, M.B., Klass, M.J., 1980. On the estimation of security price volatility from historical data. J. Bus. 53, 67–78.Gately, D., Huntington, H.G., 2002. The asymmetric effects of changes in price and income on energy and oil demand. Energy J. 23, 19–55.Goodwin, P., Dargay, J., Hanly, M., 2004. Elasticities of road traffic and fuel consumption with response to price and income: a review. Transp. Rev. 24, 275–

292.Graham, D.J., Glaister, S., 2002. Review of Income and Price Elasticities of Demand for Road Traffic. Centre for Transport Studies Imperial College of Science,

Technology, and Medicine, London.Graham, D.J., Glaister, S., 2004. Road traffic demand elasticity estimates: a review. Transp. Rev. 24, 261–274.

66 Y. Kwon, J. Lee / Transportation Research Part A 63 (2014) 56–66

Granger, C.W.J., Lee, T.H., 1989. Investigation of production, sales, and inventory relationships using multicointegration and non-symmetric error correctionmodels. J. Appl. Econ. 4, S145–S159.

Guo, H., Kliesen, K.L., 2005. Oil price volatility and U.S. macroeconomic activity. Fed. Reserve Bank St. Louis Rev. 87, 669–683.Hanly, M., Dargy, J., Goodwin, P., 2002. Review of Income and Price Elasticities in the Demand for Road Traffic. University of London, Centre for Transport

Studies, ESRC Transport Studies Unit, London.Hamilton, J.D., 1983. Oil and the macroeconomy since World War II. J. Polit. Econ. 91, 228–248.Hughes, J.E., Knittel, C.R., Sperling, D., 2006. Evidence of a Shift in the Short-Run Price Elasticity of Gasoline Demand. NBER Working Paper, No. 12530. NBER,

Cambridge, MA.Kuper, G.H., van Soest, D.P., 2006. Does oil price uncertainty affect energy use? Energy J. 27, 55–78.Lee, J., Han, S., Lee, C., 2009. Oil Price and Travel Demand. Korea Transport Institute, Korea.Mork, K.A., 1989. Oil and the macroeconomy when prices go up and down: an extension of Hamilton’s results. J. Polit. Econ. 97, 740–744.Pindyck, R.S., 1991. Irreversibility, uncertainty, and investment. J. Econ. Lit. 29, 1110–1148.Parkinson, M., 1980. The extreme value method for estimating the variance of the rate of return. J. Bus. 53, 61–66.Rogers, L.C.G., Satchell, S.E., 1991. Estimating variance from high, low and closing prices. Ann. Appl. Probab. 1, 504–512.Schwert, W.G., 1989. Why does stock market volatility change over time? J. Finance 44, 1115–1153.