The Price and Income Elasticity Demand for Gasoline in the United States By: Mark Groza Undergraduate: University of Akron (BSLE) May 5, 2006 Abstract Why do Americans consume so much gasoline? This is a question policy makers all over the country are dealing with. This empirical analysis does not try to explain why Americans consume so much gasoline but rather how to lower that consumption. A classical demand model is presented though the course of this paper. The goal is to find the price and income elasticity for gasoline. Introduction The appetite for gasoline by American consumers appears to be insatiable. Gasoline prices around the county are at record (nominal) highs, yet consumption does not appear to be falling at a proportionate rate. Today the United States imports 58 percent of the oil it consumes (Blum, 2005). Much of this imported oil is from unstable parts of the world, including the Middle East and parts of South America. OPEC (the Organization of Oil Exporting Countries) has increasing power over American economic matters. If that cartel decides to raise prices substantially the American economy will suffer. America s dependence on foreign oil is a major national security issue. Not just
Groza 2 national security in the traditional sense but also more importantly economic national security. For the United States to become less dependant on foreign sources of oil, gasoline consumption must be cut. To achieve this goal it is important to know the price and income elasticity demands of gasoline. Through the course of this empirical analysis a demand model for retail gasoline will be derived. This is done by using consumption, price, and income data from 49 states and District of Columbia 1. This cross-sectional analysis will provide information on how different retail gasoline prices and different percapita incomes around the county affect gasoline demand. The model will test to see if the demand for gasoline in the United States is price, and or income elastic. Literature Review There is an extensive amount of published works attempting to estimate gasoline demand elasticities. Graham and Glaister (2000) provided a recent overview of many of these estimates. They build on the work of Drollas (1984) which provided an earlier review of fuel demand models. Drollas (1984) derived from previous elasticity estimates the consensus long-run price elasticity demand for gasoline to be inelastic at around -0.80. He also found the short-run price elasticities of gasoline to be about one-third the magnitude of the long run price elasticity. All of the studies reviewed by Drollas (1984) were estimates based on data form the US market. Another study reviewed by Graham and Glaister (2000) is the work done by Bohi & Zimmerman (1986). Bohi & Zimmerman (1984) also did a review of the estimates 1 There is no consumption data available from Alaska. Alaska is omitted from this model.
Groza 3 available at that time. But unlike Drollas (1984) they reviewed demand studies from all over the world. They concluded that previous demand elasticity estimates in the long run ranged form 0.0 to -1.59, and the long run income elasticity estimates range from -.34 to 1.35. The rather large variance in estimates is a result of two things. First Bohi & Zimmerman (1984) examined a huge number of studies done before 1984. Secondly, as they noted, there are literally an infinite number of ways to construct a demand model for retail gasoline. An important conclusion that Graham and Glaister (2000) reached is there are a wide range of estimates on the impact of price on the consumption of gasoline. The type of data used as well as the type and number of variables all influence the magnitude of the parameter estimates. This being the case Graham and Glaister (2000) emphasize the importance of clearly understanding the method and the variables used when constructing a gasoline demand model. Keeping this advice in mind the variables and reason for their inclusion will be clearly explained throughout this empirical analysis. Empirical Model In this section the model used to predict the price and income elasticity of demand for gasoline will be presented. The dependant variable in this model will be consumption of gasoline per-capita in log form (lnconsumption). The independent variables of interest for the sake of this analysis will be gasoline prices (lngasprice) and per-capita income (lnincome), both in log form. To guard against possible omitted variable biases a host of other variables that may affect gasoline consumption are also included.
Groza 4 lnconsumption= b 0 - b 1 lngasprice + b 2 lnincome + b 3 DRIVERS - b 5 CARS + b 6 BIKES - b 7 BUSES - b 10 GREENAUTOS + Ei Where B 0 = Intercept lnconsumption = State per-capita gallons consumed of gasoline in 2003 (log form), lngasprice = State average price for all grades of gasoline in the year 2003, including state taxes excluding federal taxes (log form), lnincome = state per-capita income in 2003 (log form), DRIVERS = percent of state population which has a drivers license, CARS = percent of total number of registered vehicles that are cars, BIKES = percent of total number of registered vehicles that are motorcycles, BUSES = percent of total number of registered vehicles that are buses, GREENAUTOS = percent of total number of registered vehicles that are classified as alternative-fuel vehicles, and Ei = a classical random disturbance term. The dependant variable (lnconsumption) in this empirical model represents the state per-capita consumption of gasoline in gallons. This variable is logged so that a true price elasticity model can be derived. LnCONSUMPTION includes both private and commercial highway consumption of gasoline 2. In this analysis gasoline consumption data from each state is divided by the states total population. While admittedly not every man, woman, and child may not consume gasoline directly it is fair to assume the ratio of consumers to non consumers is fairly equal in each state. 2 Gasoline consumption data is fairly accurate do to states collection of gas taxes. Consumption data for the year 2003 is used for this model. The population data collected was from the closes census to 2003, the 2000 census. Alaska is excluded, Washington D.C. is included.
Groza 5 Total gas prices (lngasprices). This is the independent variable of interest in constructing a price elasticity estimate for gasoline. LnGASPRICES is the states average retail gasoline prices for all grades in the year 2003. This variable is put into log form so a true price elasticity estimate will be derived. LnGASPRICES includes all state gas taxes but does not include federal gasoline taxes. 3 This variable s parameter estimate is predicted to have a negative (-) sign considering a classical demand model. Per-capita income (lnincome). This variable is the per-capita income for residents of each state in the year 2003 in log form. As in any classical demand model income must be included. It is logged so an income elasticity estimate can be derived. Gasoline is most likely a normal good therefore the expected sign on the parameter estimate for the variable (lnincome) is positive (+). That is as income goes up demand of gasoline will go up. Percent of population which has a drivers license (DRIVERS). This variable is derived by dividing the number of registered drivers in each state by that state s total population. The number derived is the percentage of residents in each state that have driver s licenses. The expected sign for this variable is positive (+). As the percentage of population that has a driver s license goes up, consumption can be expected to go up. Percent of vehicles in each state classified as a car (CARS). The U.S. Department of Transportation has four classifications for private and commercial vehicles they are: cars, trucks, motorcycles, and buses. The variable (CARS) is the percent of total vehicles that are classified as cars in each state. Typically cars are more fuel efficient then trucks and buses yet, unlike motorcycles, they can carry more then one or 3 Federal gasoline taxes are constant across the 50 states and the District of Columbia, it is not necessary to include this since it would raise each states gasoline prices proportionately.
Groza 6 two passengers. With more cars there can be more car-pooling (multiple passengers getting to the same place in one vehicle). We can expect as the percentage of total vehicles that are cars goes up, fuel consumption will go down. Therefore, the variables (CARS) will have a parameter estimate with a negative sign (-). Percent of vehicles in each state classified as a motorcycle (BIKES). Motorcycles are more fuel efficient then any other classification of vehicle. However, unlike all other classification of vehicle motorcycles can only carry one or two passengers. Also in many cases motorcycles are used as recreational vehicles. Most people ride them not as a substitute to the miles driven in their cars or tucks, but in addition to their normal daily driving. Individuals who ride motorcycles are inclined to use more gasoline for recreational purposes. The variable (BIKES) is expected to have a parameter estimate with a positive sign (+). As the number of registered motorcycles goes up gasoline consumption will go up. Percent of vehicles in each state classified as a bus (BUSES). This variable represents the percent of registered vehicles that are buses. Private and commercial buses are the only form of mass transportation included in this analysis 4. Mass transportation reduces the per-capita consumption of gasoline. As the percentage of registered buses goes up, gasoline consumption can be expected to go down. The variable (BUSES) will have a parameter estimate with a negative sign (-). Alternative-fuel vehicles (GREENAUTOS). This variable represents the percentage of total registered vehicles that are classified by the U.S. Department of Transportation as alternative-fuel vehicles. Alternative-fuel vehicles includes any 4 This analysis deals only with private and commercial consumption of gasoline, not public consumption. Most mass-transportation systems are publicly ran and therefore excluded from this analysis.
Groza 7 vehicle that runs on an energy source other then gasoline. They do not include gasoline/electric hybrids. The more registered vehicles there are that run on energy other then gasoline the less gasoline will be consumed. The expected sign for this variable s parameter estimate is negative (-). As the number of alternative-fuel vehicles goes up gasoline consumption goes down. Results Two different regressions were run using a combination of different variables (see table 1). Of these two regressions, regression number two was superior. This regression included the variables explained in the previous section. The model has an F-value of 19.84. Therefore we can assume some of these variables do in fact affect the independent variable, the consumption of gasoline. The root mean squared error is 0.09. This model has a very low average error. The adjusted R 2 is 0.729; nearly 73% of that error is explained. Considering the values of these three ways to judge the quality of a regression are all fairly good (in terms of goodness of fit ) we can assume this log-linear regression as a whole is statistically significant. This analysis s major goal was to see if gasoline in the United States is price and or income elastic. In ensuring accuracy for those parameter estimates, and to protect against omitted variable biases a number of other variables were included. The parameter estimates for these secondary variables can be very useful for policy makers interested in lowering Americas demand for gasoline. First however, we must address the independent variables of interest lngasprice and lnincome.
Groza 8 This model predicted the price elasticity of gasoline to be -1.265. This is much more elastic then Drollas (1984) concluded it would be but well within the range Bohi & Zimmerman (1984) predicted. According to this model the demand for gasoline is price elastic. Holding everything else constant as the price of gasoline goes 1% the demand for gasoline will go down 1.27%. This parameter estimate has a T-value of -5.65; statistically significant at the 99% level. The other variable of major interest in this model was lnincome. One of the goals of this analysis was to see if gasoline was income elastic. Unfortunately, this model did not give us a statistically significant parameter estimate for income elasticity. The model produced a parameter estimate of 0.069 and had T-value of 0.59; statistically significant at no useful level. Regrettably, this model did not yield an accurate income elasticity estimate. Possible explanations for this will be explained in the limitations section of this paper. Of the control variables included in this analysis the most significant to policy makers are probably the variables (BUSES) and (GREENAUTOS). The variable (BUSES) had a parameter estimate of -27.89 and a T-value of -2.11. It is predicted that as the percentage of buses as registered vehicles goes up 1% the demand for gasoline will go down 2.8%. This parameter estimate is statistically significant at the 95% level. The variable (GREENAUTOS) had a parameter estimate of -19.41 and a T-value of -2.42. This parameter estimate is statistically significant at the 99% level. It predicts as the percentage of alternate-fueled vehicles goes up by 1% demand for gasoline will go down 1.9%.
Groza 9 The variable (CARS) also yielded a statistically significant parameter estimate at the 99% level. Its parameter estimate was -0.61 and its T-value was -2.56. However, for policy purposes this parameter estimate is most likely not useful. This estimate predicts as the percentage of cars as registered vehicles goes up 1% the consumption of gasoline will go down 0.06%. This is far too small of a change to be a major option for policy makers. The final two variables (DRIVERS) and (BIKES) are also most likely not useful to policy makers. DRIVERS had a parameter estimate of 0.38 and a T-value of 1.27; it is not statistically significant. BIKES had a parameter estimate of 2.23 and a T-value of 1.78. This variable is statistically significant at the 90% but again not very useful for policy makers. An increase or decrease in the number of licensed drivers or registered motorcycles in a given State will not seriously impact gasoline consumption. Table 1 Results of two regressions which show the impact of gasoline prices and income on the consumption of gasoline; including a number of control variables. Dependant variable: Gasoline Consumption Best Regression Regressor (1) (2) lngasprice -1.2843*** -1.2648*** (0.25) (0.223) lnincome.0868 0.07 (0.12801) (0.1176) DRIVERS 0.36110 0.3859 (.3164) (0.303) CARS -0.5716*** -0.61233*** (.2665) (0.23944) BIKES 2.1068 2.23143* (1.44587) (1.25485) BUSES -28.0543*** -27.89475*** (13.50328) (13.21232)
Groza 10 GREENAUTOS -20.54495*** -19.41303*** (8.82313) (8.01465) REGISTERD -0.02399 (.13864) MEANTIME -0.00322.00673 Intercept 11.74323*** 11.73041*** (1.47738) (1.41128) Summary Statistics MSE 0.09256.090 Adj. R 2.7172.7291 F Value 14.81 19.84 N 50 50 These regressions were estimated using fuel consumption data from 49 states and the District of Columbia (no data available from Alaska). The individual coefficient is statically significant at the *90% level or **95% level or ***99% level using a two-sided test. CONCLUSION Statistically speaking this model is defiantly relevant. It has a high F-value, a high adjusted R 2, and a low root mean squared error. Five of the eight parameter estimates yielded are significant at the 99% level. One other estimate is significant at the 90% level. This model is also relevant for practical purposes. According to this model three things can be done to reduce the consumption of gasoline in the United States. They are: raise gasoline prices, increase the number of private and commercial buses in operation, and increase the number of alterative-fueled vehicles. Increasing prices 1% will result in a 1.27% reduction in gasoline consumption. Consumption can be curtailed by raising gasoline taxes which will increase the final price to consumers. However, this
Groza 11 option will undoubtedly have serious negative economic and political consequences. There are much better options for reducing gasoline consumption in the United States. This model shows an increase in private bus and alternative-fuel vehicle registrations will cut the per-capita gasoline consumption in the U.S. If the number of alternative-fuel vehicles as a percentage of total vehicles would go up 10% we can expect to see a 19% decrease in per-capita consumption of gasoline. Also if there was a 10% increase in the number of private buses as a percentage of total vehicles we can expect a 28% reduction in gasoline consumption. Encouraging private bus ownership and alternative-fuel vehicle ownership is a policy that would not have the negative consequences raising gas taxes would have. Limitations This analysis is unique to gasoline demand models because it uses cross-sectional data from one given period of time. One of the major limitations of this study is that it does not take into account price changes in the short or long run. This study tells us how price differences in different states affect the demand for gasoline in one given year. Another limitation in this study is there were no supply issues taken into account. By leaving out supply issues this model is vulnerable to simultaneous causality bias. Throughout the year 2003 there most likely were many price spikes do to lowered supply. None of these supply issues were taken into consideration in this model. Simultaneous causality bias means the parameter estimates in this model are not unbiased nor are they consistent. Future work will have to take into account the simultaneous causality bias present in this study.
Groza 12 One of the other limitations of this study is the income elasticity parameter estimate was not statistically significant. Unfortunately, we can not conclude the variable (lnincome) defiantly affects gasoline consumption; the null can not be rejected. The regression used in this analysis did not take into account cost of living differences in different States. Nor did it take into account different driving patterns for different income individuals. Generally individuals who live in big cities have higher incomes but drive less then people who live in rural areas. Also high income individuals may be more inclined to fly then lowered income individuals. The fact this model did not take into account different driving patterns for different income individuals is a major flaw.
Groza 13 Data Variable Name Description Mean (Standard Data Deviation) Source CONSUMPTION state per-capita gallons consumed of gasoline in 2003 468.833(70.42) (1) Tbl. 7-4 GASPRICE state average price for all grades of gasoline in the year 135.35 (8.919) (1) Tbl. 6-11 2003, including state taxes excluding federal taxes 6-12 INCOME state per-capita income in 2003 30704 (4937) (2) DRIVERS percent of state population which has a drivers license.6945 (.0548) (1) Tbl. 4-2 CARS percent of total number of registered vehicles that are cars.55 (.09) (1) Tbl. 5-1 BIKES percent of total number of registered vehicles that are motorcycles.026 (.013) (1) Tbl. 5-1 BUSES percent of total number of registered vehicles that are buses.00156 (.0016) (1) Tbl 5-1 GREENAUTOS percent of total number of registered vehicles that are run.0025 (.0024) (1) Tbl 7-5 on something other that 100% petroleum products TRUCKS percent of total number of registered vehicles that are buses 0.422 (0.087) (1) Tbl 5-1 MEANTIME average travel tome to work per employee 23.75 (3.571) (1) Tbl 4-1 Notes on data sources: 1. US Department of Transportation, State Transportation Statistics 2004, Washington D.C.: Tables http://www.bts.gov/publications/state_transportation_profiles/state_transportation_statistics_2004/ 2. Bureau of Economic Analysis, Regional Economic Accounts, per capita personal incomes 2003. http://www.bea.gov/bea/regional/spi/drill.cfm
Groza 14 References Blum, Justin Bill Wouldn t Wean US Off Oil Imports, Analysts Say Washington Post On-Line July 26, 2005 March 19, 2006. http://www.washingtonpost.com/wpdyn/content/article/2005/07/25/ar2005072501707.html. Bohi, D. and M. Zimmerman (1984): An update on econometric studies of energy demand Annual Review of Energy 9, 105-154. Bureau of Economic Analysis, Regional Economic Accounts, per capita personal incomes 2003. http://www.bea.gov/bea/regional/spi/drill.cfm Drollas, L. (1984) The Demand for Gasoline: Further Evidence Energy Economics, 6, 71-82. Foos, G. (1986) Die Determinanten der Verkehrnachfrage Karlsruher Beitrage zur Wirtschaftspolik und Wirschaftsfrorschung, 12 Loper Verlag: Karlsruhe. Graham, Daniel J. and Glaister Stephen. (2000) The Demand for Automobile Fuel A Survey of Elasticity s. Journal of transport Economics and Policy V. 36 Part 1, January 2000 p 1-26. US Department of Transportation, State Transportation Statistics 2004, Washington D.C.: Tables http://www.bts.gov/publications/state_transportation_profiles/state_transportatios tatistics_2004/.