It's the economy, stupid: Increasing fuel price is enough to explain Peak Car in Sweden

Transcription

1 It's the economy, stupid: Increasing fuel price is enough to explain Peak Car in Sweden Anne Bastian and Maria Börjesson Centre for Transport Studies, KTH Royal Institute of Technology CTS Working Paper 2014:15 Abstract It has long been well-known that economic variables such as GDP and fuel price as well as socio-demographic characteristics and spatial distribution are key factors explaining car use trends. However, due to the recently observed plateau of total car travel in many high income countries, it has been argued that other factors, such as changes in preferences, attitudes and life-styles, have become more important drivers of car use. This would imply that traditional variables are no longer enough for explaining car travel trends. However, in this paper we show that economic variables alone can explain the observed car use trends in Sweden We also find that urban populations, in particular those with low incomes, respond stronger to fuel price increases and economic downturn, i.e. are reducing car travel more. Among high income urban populations, however, we find signs of saturation in car ownership and distances driven. This underscores the importance of accounting for differences in accessibility with other travel modes and income distribution when explaining the Peak Car phenomenon. Keywords: Peak Car, fuel price elasticity, transport model, urban car travel Centre for Transport Studies SE Stockholm Sweden

2 1 INTRODUCTION In the last decade the previously inexorable rise in car distances driven in Sweden and in many other high income counties has plateaued or declined (Goodwin, 2013; Millard Ball and Schipper, 2011). Although it is well-known that traditional factors such as economic variables and demographics are important for car use trends, this recent development has led some to argue that other factors, such as changes in preferences, attitudes and life-styles, have become more important drivers of car use. The OECD report on Long-run Trends in Car Use (2013a), using country level data, concludes that economic downturn and rising fuel prices explain some of the trends in car distances driven but not all. The authors, however, make the strong implicit assumption that fuel price elasticity is constant across and within countries, despite substantial differences in absolute fuel prices and accessibility to attractive mode and destination alternatives. Yet, Blow and Crawford (1997), Goodwin et al. (2004) and Wadud et al. (2010) suggest that fuel price elasticity varies between urban and rural populations and by the absolute fuel price. The purpose of this study is to explore to what extent economic variables (GDP, fuel price, income distribution, population growth and location) can explain the trends in car distances driven in Sweden from 2002 to We use time-series data on car vehicle kilometers travelled (VKT) per adult and municipality in Sweden. We control for differences between urban areas and other regions as well as income differences across municipalities. The data is based on distance meter readings of over two thirds of cars in Sweden. It is therefore robust with respect to random data collection errors. The literature divides potential drivers of VKT trend changes into two broad categories. The first category includes economic variables as listed above. The second category comprises more heterogeneous factors such as shifts in preferences, attitudes and lifestyles. The question tackled in this paper, to what extent the first category of factors can explain the recent trends in VKT, is a key issue in forecasting. Transport forecasting models rely on the assumption that the parameters explaining past travel behavior remain constant over time (Fox and Hess, 2010). These parameters typically capture responses to economic and socio-demographic variables. Hence to the extent that trends are driven by such variables, a good transport model would have a chance of predicting future travel demand. To the extent, however, that trends in car use are driven by the second category of non-economic factors, transport models would not be able to predict it. The observed VKT trend break has been seen as a challenge for travel demand forecasting (Millard Ball and Schipper, 2011). The United States Energy Information Administration reduced their reference case projection of annual VKT growth rates until 2035 from 1.5% (projection from 2010) to 0.9% (projection from 2014). The Swedish Transport Administration currently predicts a corresponding growth of 1.2% annually until Accurate transport forecasts are important in appraisal, because many investment and policy decisions depend on long-term VKT forecasts, sometimes with horizons of up to 30 years. Our main result, that economic factors are the dominant drivers of change in VKT per adult in Sweden from 2002 to 2012, is thus reassuring and lends some credibility to transport forecasting. Moreover, our data indicates higher fuel price elasticity in urban 2

3 areas compared to other parts of Sweden. This is plausible given that urban residents have higher accessibility to public transport (Wadud et al., 2010) and a larger variety of destination alternatives. Headicar (2013) also finds a stronger decline in car use in London compared to the rest of the UK. Within urban areas we find that residents of low income suburbs are most sensitive to fuel price increases and economic downturn. They have reduced their driving substantially since 2008 in response to increasing fuel prices. This is consistent with Blow and Crawford (1997), who showed that urban low income households in the UK are most responsive to fuel price changes, while rural high income households are least responsive. In high income urban municipalities we find signs of saturation in car ownership and use, which matches findings by Grimal et al. (2013) in France. These results provide a strong warning for analyzing car use trends without accounting for accessibility with other travel modes and socio-economic differences. Our results suggest that land use and transport options are important determinants of absolute VKT and fuel price elasticity of VKT. Land use changes and public transport improvements in Sweden from 2002 to 2012, however, are not the main cause of VKT trend changes, having changed only marginally during this relatively short period. Section 2 describes the Swedish data used. Section 3 gives an overview of aggregate trends in VKT and economic variables in Sweden. Section 4 highlights the role of foreign immigration to Swedish VKT per adult trends. Section 5 presents results of GDP and fuel price elasticity modeling. Section 6 discusses the role of income distribution. Section 7 examines the impact of urbanization. Section 8 concludes. 2 DATA AND DEFINITIONS To measure VKT (vehicle kilometers traveled) in Sweden we use distance meter readings from mandatory passenger car inspections. The data is aggregated per municipality and calendar year and spans the period from 2002 to Actual distances driven are recorded for about two thirds of all cars registered in Sweden, both privately owned and company cars. This data is used to estimate distances driven for the remaining one third of cars in Sweden, based on car age, model and ownership type. Among the cars without recorded distances 60% are newer than three years old. The estimated yearly distance of these new cars equals the yearly average distance of the three year old cars (within the same class) inspected for the first time 1. This implies a risk that a fraction of the inter-temporal changes in driving distances lag behind in the data. The distance of each car is assigned to the municipality where the car is registered. Comparisons across municipalities within the same year can be somewhat biased, because company cars are sometimes registered in a municipality different from the residence of the user. Comparisons across time are more reliable where the share of company cars per municipality remains rather stable over time. The municipalities of Solna, Malmö and Lund are excluded altogether from the analysis because of their high concentration and fluctuation of company car registrations. In this paper we define urban areas as the 25 municipalities within Stockholm County (excluding Solna municipality) and the municipalities of Gothenburg and Mölndal. The latter two municipalities constitute the Gothenburg urban area. We apply this relatively 1 The estimation method is documented in (Trafikanalys 2013). 3

4 narrow definition of urban areas, because the extensive public transit networks, labor market size and population density make the Stockholm and Gothenburg urban areas distinct from other parts of Sweden. Stockholm County (excluding Solna municipality) has 2.1 million inhabitants. Stockholm has a well-developed public transport network including metro, commuter trains, trams and buses. The share of public transit trips to and from the inner city reaches 75% during peak hours. The inner city of Stockholm is built on several islands, connected by bridges, which used to imply relatively high road congestion for a city of comparable size. The bridges connecting the inner city to the outer city are the major bottlenecks in the road network, and congestion charges are levied on these since January The share of driving is about 30% of all trips. Gothenburg is Sweden s second largest city. Gothenburg and Mölndal municipalities have inhabitants in total. The public transport system comprises trams, commuter trains and buses. Road congestion is fairly limited. The share of driving is about 42% of all trips, while the corresponding share in the rest of Sweden is 63%. In this paper we measure car use as VKT per adult. Adults are defined as registered residents of Sweden aged Adults residing in the Stockholm or Gothenburg urban areas are referred to as urban adults. Population, GDP, income and driving license statistics are taken from Statistics Sweden. We use real fuel pump volume sales prices as provided by the Swedish Petrol and Biofuel Institute (SPBI) as well as the World Bank. We do not consider changes in diesel prices separately. Diesel volume prices in Sweden increased somewhat faster than fuel prices between 2003 and 2008 and have been matching fuel prices since We do not adjust for changes in fuel efficiency, car sizes or rebound effects. The elasticity estimates by Goodwin et al. (2004), which we apply, were also derived without separating out such effects. Elasticity in this paper is defined as the change in VKT per adult over the change in real fuel price (or real GDP per adult). 3 AGGREGATE NATIONAL TRENDS From 2002 to 2007 total car VKT in Sweden grew by about 1.4 % annually and remained roughly stable from 2008 to 2012, see Figure 1. Considering a population growth of about 1% annually, the stability in absolute VKT from 2008 to 2012 corresponds to a 5% decline in VKT per adult. This decline is not a consequence of changes in car ownership per adult, which has been roughly stable at the national level. Hence, on average the distance driven per car slightly declined since Patterns in population growth and VKT per adult vary between different parts of Sweden, as one would expect. In the Stockholm and Gothenburg urban areas VKT per adult declined by 8% from 2007 to 2012, while the adult population grew by a total of 10% during this time. Accordingly the absolute VKT in urban areas increased by 1%. Urban car ownership per adult declined by 3% from 2007 to Hence, the average distance driven per car declined somewhat in urban areas. The period from 2002 to 2012 is marked by strong increases in real fuel prices, 39% in total. Sweden, along with other European countries, has among the highest absolute consumer price of fuel in the world. In 2012 a liter of gasoline in Sweden cost the equivalent of US $ The average price paid in the United States was about half as much (World Bank, 2014). 4

5 index It's the economy, stupid: Increasing fuel price is enough to explain Peak Car in Sweden 140% 135% 130% 125% Total Sweden: index versus 2002 real gas price index real GDP per adult index 120% 115% 110% 105% 100% 95% year Figure 1: based on data from Statistics Sweden, SPBI and Trafikanalys (car inspection data) adult population index total VKT index cars per adult index Real incomes of employed and retired residents of Sweden have continued to increase, even since the 2008 crisis (Statistics Sweden, 2014a). Differences in disposable income between employed versus unemployed, foreign born versus Swedish born and high income versus low income groups have rapidly increased since the middle of the last decade (Statistics Sweden, 2014a). In the following two sections we explore potential explanations for the recent decline in VKT per adult and why this decline is stronger in urban areas than in the rest of Sweden. 4 IMMIGRATION MATTERS FOR SWEDISH VKT PER ADULT Immigration is important for VKT per adult trends in Sweden because of its contribution to urban population growth, income inequality and driving licence gaps: Immigrants lower licence holding levels, lower average income and higher concentration in urban areas contributes to the stronger decline of VKT per adult in urban areas. The majority of foreign immigrants to Sweden settle in the urban areas. About one third of mid-aged adults presently living in Stockholm were not born in Sweden (Statistics Sweden, 2014c). Since 2006, when there was a change in immigration regulations, foreign immigration to Stockholm County jumped from previously around 18,000 to around 30,000 people per year. About half of the immigrants were born outside the European Union or Nordic Countries (Stockholm County, 2011). The Stockholm suburbs Södertälje, Sundbyberg and Upplands-Väsby experienced the strongest decline in VKT per adult among all urban area municipalities since 2007 (about 10% decline), more than even the city municipalities of Stockholm and Gothenburg (about 6% decline). These municipalities have among the highest shares of foreign-born residents in Sweden, but also good public transit access and below average incomes. We cannot isolate the effects of foreign-born populations from income effects in our data, because the two measures are highly negatively correlated. The average economic standard of non-eu/nordic immigrants is 64% relative to Swedish born residents (Statistics Sweden, 2014a). 5

6 Foreign-born young adults are much less likely to acquire a driving licence than their counterparts born in Sweden, see Figure 2. Moreover, licences that immigrants acquired outside of Europe are typically not valid in Sweden and costly to replace. This divide by foreign background is most pronounced among women, in particular low educated women (Statistics Sweden, 2014c). Sweden s remaining gender gap in licence holding is due to the substantial gender gap among foreign-born populations and due to the oldest generation, see Figure 2. The gender gap among seniors will continue to shrink as today s younger retirees replace the oldest generation. However the gender gap among foreign-born residents as well as the gap between Swedish and foreignborn residents may well remain, especially as immigration numbers are expected to remain high over the coming years (Statistics Sweden, 2014b). 100% Share of car driving licence holders among residents of Sweden in 2013 by gender, age and birth country 80% 60% 40% 20% 0% age age age age age age age age 60+ Men born in Sweden Women born in Sweden Men born abroad Women born abroad Figure 2: Based on Swedish driving licence register from Statistics Sweden 5 GDP AND FUEL PRICES EXPLAIN VKT PER ADULT TRENDS Total VKT have grown only slightly from 2002 to 2007 and remained stable from 2007 to 2012 in both the urban areas and the rest of Sweden. However, most of the adult population growth occurred in the urban areas, in particular since This means that VKT per adult have declined much more in urban areas since 2007 than in the rest of Sweden. This section explores the observed trend decline in VKT per adult by applying standard elasticities to GDP per adult and fuel prices. We expect to find smaller absolute fuel price elasticities outside of urban areas, as shown by Blow and Crawford (1997) for the UK. Consequently we analyze VKT per adult separately for urban and other areas of Sweden. The UK literature review by Goodwin et al. (2004) suggests long term elasticities of about -0.3 for fuel prices and 0.5 for GDP on car traffic volumes. The Swedish National Transport model SAMPERS, which is estimated based on National Travel Survey data, produces an average fuel price elasticity of -0.3 for regional trips (Börjesson, 2012), in line with earlier literature. Applying these standard elasticities to the Swedish real fuel price and real national GDP per adult trends produces a rise and fall pattern in VKT per adult from 2002 to This modeled trend closely matches the actual VKT per adult trend in the urban areas, as shown in Figure 3. 6

7 index It's the economy, stupid: Increasing fuel price is enough to explain Peak Car in Sweden 145% 140% 135% 130% 125% index vs real gas price index real GDP per adult index 120% 115% 110% 105% 100% 95% 90% year elasticity model: -0.2 gas price, 0.5 GDP per adult VKT per adult index, rest of Sweden elasticity model: -0.3 gas price, 0.5 GDP per adult VKT per adult index, urban areas Figure 3: Elasticity modeling results and statistics from Statistics Sweden, SPBI and Trafikanalys (car inspection data) In line with our a priori expectation, we had to reduce the fuel price elasticity, to -0.2, to fit the modeled VKT trend per adult to the observed trend in the non-urban parts of Sweden. A tentative but plausible explanation for why the elasticity would be lower outside the urban areas is better urban access to attractive mode and destination alternatives. To explore this further we break VKT per adult trends down to the municipality level. Indeed we find a positive relationship between the VKT per adult in 2007 and the growth in the same measure from 2007 to 2012, see Figure 4. Hence, residents of municipalities that were already very car dependent in 2007 reduced their driving less when fuel prices increased and the economy slowed down. There is more variation in VKT per adult growth rates between different small municipalities than between larger municipalities. This is to be expected, because smaller populations imply more random variation. 7

8 VKT per adult growth index from 2007 to 2012 It's the economy, stupid: Increasing fuel price is enough to explain Peak Car in Sweden 110% Sweden's municipalities VKT per adult 2007 vs. change in in VKT per adult from 2007 to 2012 bubble area = 2007 adult population count 105% 100% 95% 90% 85% 80% VKT per adult in 2007 rest of Sweden municipalities urban area municipalities Figure 4: Based on Trafikanalys (car inspection data) and Statistics Sweden population statistics 6 INCOME DISTRIBUTION MATTERS FOR ELASTICITIES There are signs of saturation among high income municipalities, both in terms of car ownership and VKT per adult. As Swedish income growth becomes increasingly concentrated among high income groups, average income growth becomes less influential for average VKT per adult trends. National average real incomes have risen since Still, car ownership per adult has remained roughly stable, both nationally and in high income urban municipalities. In lower income urban municipalities, however, car ownership per adult has declined slightly. Among the urban municipalities it is those with lower incomes that show the strongest reduction in VKT per adult since Residents of high income urban municipalities on the other hand adjusted their VKT much less. This would be expected by an argument of diminishing utility of money. If we would not control for differences in population size and urbanity across municipalities, our data would suggest that residents of higher income municipalities had reduced their VKT more than residents of low income municipalities since However, controlling for these variables reveals that this is a spatial effect, because larger and more urban municipalities tend to have higher incomes, see Figure 5. This demonstrates the importance of separating the analysis of urban areas from less accessible areas. 8

9 VKT per adult growth index from 2007 to 2012 It's the economy, stupid: Increasing fuel price is enough to explain Peak Car in Sweden 110% 105% 100% 95% 90% 85% Sweden's municipalities median income 2007 vs. change in in VKT per adult from 2007 to 2012 bubble area = 2007 adult population count 80% median individual annual income in 2007 in Swedish krona rest of Sweden municipalities urban area municipalities linear trendline all of Sweden's municipalities Figure 5: Based on Trafikanalys (car inspection data) and Statistics Sweden population and income statistics 7 URBANIZATION IS NOT A MAIN DRIVER OF RECENT VKT TRENDS Urbanization, in the sense that urban area municipalities experienced a faster population growth than other municipalities, is not a major explanation to the national decline in VKT per adult since Had the average VKT per adult of each municipality remained constant from 2007 to 2012, the national VKT per adult would have declined by only 0.5% due to population count changes, compared to the observed 6%. The impact of urbanization on national VKT per adult is so limited, because the somewhat stronger marginal population growth in urban municipalities, as shown in Figure 6, only slowly affects absolute population count ratios between different municipalities. It remains an open question if and how a shift in the socio-demographic composition of the populations within urban areas and the rest of Sweden may have contributed to observed VKT trends, but these changes are also too slow to explain trend breaks in VKT per adult over 5-10 years. 9

10 population growth index from 2007 to 2012 It's the economy, stupid: Increasing fuel price is enough to explain Peak Car in Sweden 120% 115% 110% 105% 100% 95% Sweden's municipalities VKT per adult 2007 vs. adult population growth from 2007 to 2012 bubble area = 2007 adult population count 90% VKT per adult per day in 2007 rest of Sweden municipalities urban area municipalities Figure 6: Based on Trafikanalys (car inspection data) and Statistics Sweden population statistics 8 CONCLUSIONS This paper suggests that the observed rise and fall in Swedish VKT per adult from 2002 to 2012 can be explained by applying standard elasticities to trends in fuel price and GDP per adult. Our results further suggest that absolute fuel price elasticities are highest among urban populations, in particular among those with lower incomes. This is in line with earlier research and likely due to higher access to attractive mode and destination alternatives in urban areas. We therefore conclude that changes in economic factors (GDP, fuel price, population growth, income growth and distribution) and their interaction with land-use and accessibility with different travel modes are key drivers behind recent changes in VKT per adult in Sweden. Hence, the trend decline in car travel during the recent decade is not sufficient to reject the assumption underlying transport forecasting: that travel time and cost parameters, explaining travel behavior, remain constant over time. However, the accuracy of forecast models depends not only on model parameters but also on projections of a number of input variables. Long-term projections of GDP, income growth and distribution, immigration, population location, land-use and transport options are uncertain. For instance, the economic downturn and fast increase in fuel prices after 2008 was almost impossible to predict. Our results are thus consistent with De Jong et al. (2007), finding that transport model forecasting errors driven by input uncertainty (regarding the future value of the input variables) are usually larger than errors driven by model uncertainty (model specification and parameters used). Our data further indicates saturation in driving among high income groups. Hence, accounting for changes in income distribution should improve models of travel demand and car ownership. Moreover, if income acts as a proxy for other variables, e.g. degree of the job specialization, the effect of income on behavior in the cross-section will be different from the effect over time. To capture such dynamics, disaggregate longitudinal 10

11 data is required in model estimation, in addition to the cross-sectional data that is typically applied. Because the absolute price elasticity of VKT is highest among the urban populations and those with lower incomes, an important message to policy makers is that it is a combination of increases in the cost of driving, dense and mixed land-use as well as accessibility by alternative modes that seems to be effective in reducing car travel. For policy makers the inevitable remaining uncertainty about future levels of car travel suggests that transport and land-use initiatives should be evaluated not just by their feasibility and expected social return but also by their flexibility and resilience to changes in demand, technology, financing and priorities. We have shown that there is no need to assume changes in fundamental attitudes towards car driving to explain recent VKT trends in Sweden, which is in line with findings from the Netherlands (Netherlands Institute of Transport Policy Analysis, 2014). This does, however, not necessarily imply that changes in preferences, attitudes, lifestyles and social trends are not relevant or important for behavioral changes. Instead it seems likely that changes in these factors to a large extent are induced by long-term effects of economic incentives and that the effects of the two categories of car use trend drivers are inseparable in the long-run. In fact, many long-term lifestyle trends, such as urbanization and more youth acquiring university education, seem to be driven by economic incentives. This suggests that adaptations in terms of shifts in lifestyles, preferences and attitudes reinforce the effectiveness of economic incentives in the long-run and probably also increase the acceptability of them. There are indications in the transport sector that this is the case. For instance travelers adapt more easily to economic incentives over a longer time (Börjesson et al., 2012; Goodwin et al., 2004) and are often not even aware that they change their attitudes or their travel behavior in response to economic incentives, as found by Eliasson (2014) with regard to the Stockholm congestion charge introduction. Our findings from Sweden are likely transferable to some European countries with similar conditions in fuel prices, incomes, urbanization levels and public transit supply. However, in countries with very different conditions forecasting models likely need different specifications and parameter values. For example in the United States the number of cars exceeds the number of license holders (United States Department of Energy, 2013). Dargay et al. (2007) also emphasize the need to adjust transport models to local conditions, by showing that saturation levels for car ownership are significantly lower with high urbanization and population density. 9 REFERENCES Blow, L., Crawford, I. (1997). The distributional effects of taxes on private motoring. IFS Commentaries C065. Institute for Fiscal Studies: London, UK. Börjesson, M., Eliasson, J., Hugosson, M. B., Brundell-Freij, K. (2012). The Stockholm congestion charges 5 years on. Effects, acceptability and lessons learnt. Transport Policy, 20 (0), doi: /j.tranpol Börjesson, M. (2012). Forecasting demand for high speed rail. CTS Working Paper 2012 (12). Dargay, J., Gately, D., Sommer, M. (2007). Vehicle Ownership and Income Growth, Worldwide: , The Energy Journal, 28 (4),

12 De Jong, G., Daly, A., Pieters, M., Miller, S., Plasmeijer, R., Hofman, F. (2007). Uncertainty in traffic forecasts: literature review and new results for The Netherlands, Transportation, 34 (4), Eliasson, J. (2014). The role of attitude structures, direct experience and framing for successful congestion pricing. Transportation Research A, 67, Fox, J., Hess, S. (2010). Review of Evidence for Temporal Transferability of Mode- Destination Models, Transportation Research Record, 2175(1), 74-83, doi: / Grimal, R., Collet, R., Madre, J. (2013). Is the Stagnation of Individual Car Travel a General Phenomenon in France? A Time-Series Analysis by Zone of Residence and Standard of Living. Transport Reviews, 33 (3), Goodwin, P. (2013). Peak Travel, Peak Car and the Future of Mobility: Evidence, unresolved issues, policy implications and a research agenda. In OECD and ITF, Longrun Trends in Car Use, OECD Publishing/ITF. Goodwin, P., Dargay, J., Hanly, M. (2004). Elasticities of Road Traffic and Fuel Consumption with Respect to Price and Income: A Review. Transport Reviews, 24 (3), doi: / Headicar, P. (2013). The Changing Spatial Distribution of the Population in England: Its Nature and Significance for Peak Car. Transport Reviews 33 (3): doi = / Millard Ball, A., Schipper, L. (2011). Are We Reaching Peak Travel? Trends in Passenger Transport in Eight Industrialized Countries. Transport Reviews, 31 (3), doi: / Netherlands Institute of Transport Policy Analysis (2014). Not car-less, but car-later. (Accessed on ). OECD (2013a). Executive Summary. In OECD/ITF, Long-run Trends in Car Use, OECD Publishing/ITF. OECD (2013b). Report. Long-run Trends in Car Use, ITF Round Tables, No. 152, OECD Publishing/ITF. Santos, G., Catchesides, T. (2005). Distributional Consequences of Fuel Taxation in the United Kingdom. Transportation Research Record, 1924(1), doi: / Statistics Sweden (2014a). Hushållens ekonomiska standard (The economic standard of households). Report BV/EV - pm 2014 (2). Sweden.se. Statistics Sweden (2014b). Sveriges framtida befolkning (Sweden s future population). Report BE18SM Sweden.se. Statistics Sweden (2014c). Sweden.se ( Accessed on ). 12

13 Stockholm County (2011). Flykting i Stockholms län (Refugee in Stockholm County). Report 2011:20. ISBN Trafikanalys (2013). Trafikarbete på svenska vägar - en översyn av skattningsmetoden (Traffic on Swedish roads an overview of the estimation method). PM 2013:8. Trafikverket (2014). Prognos för Personresor Trafikverkets basprognos 2014 report. (Prediction for passenger travel Swedish Transport Administration base prediction. Report). United States Department of Energy (2013). Transportation Energy Data Book, 32nd (Oak Ridge, TN: Oak Ridge National Laboratory). (Accessed on ). United States Energy Information Administration (2014). Annual Energy Outlook. (Accessed on ). Wadud, Z., Graham, D.J., Noland, R. B. (2010). Fuel demand with heterogeneity in household responses. The Energy Journal, 31 (1), World Bank (2014). ( Accessed on ). 13