Few studies have estimated gasoline demand in Canada. Most of these studies have either failed to recognise ali the ways in which consumers can react to gasoline priče changes ог have implemented рге-1973 data which did not provide good estimates of priče elasticities (Dewees et al, 1975; Dahl ,1978; Shalaby and Waghmare, 1980). Some models attempt to estimate the components of gasoline demand using aggre-gated data (Gallini 1983; Dahl 1982). The problem with these models is that they do not identify the relationship between the household decisions on vehicle holdings and on usage. These two decisions are likely to be dependent; the use of a vehicle depends on its type and the type of vehicle chosen depends on its expected use. Since these decisions are made within the household, household data should be used for estimating the various components of gasoline demand. This paper provides an estimate of household gasoline demand in Canada by applying a detailed model to pooled time-series (1969-1988) and cross-sectional provincial data. The model recognises three major behavioural changes that households can make in response to gasoline priče changes: drive fewer miles, purchase fewer cars, and buy more fuel-efficient vehicles. In the model, fuel есопоту is treated in considerable detail. The two components of the fuel есопоту of new cars sold—the technical fuel efficiency of various classes of cars and the distribution of new car sales according to their interior volume rather than their weight — are estimated as functions of economic variables. Car manufacturers are assumed to improve the technical fuel есопоту according to their expectation of consumers[1] response to future changes in gasoline prices and general economic condi-tions. The next three sections of this paper discuss the model, the data, and the estimation results respectively. The conclusions are presented in the final section.

2. The Model

The model presented in this study is of the investment- utilisation type. The basic identity for aggregate gasoline demand in the model, AG, is given by
AG = MS.S.E where: MS = miles driven рег car (vehicle miles); 5 = total cars in the fleet; E = average fuel есопоту of the fleet (miles/gallon). This model has its theoretical basis in the household production literature, that is, Becker (1965), Lancaster (1966), and Pollack and Wachter (1975).

The gasoline per car equation
Following the convention of earlier models, the household demand for gasoline is modelled as the outcome of a utility-maximisation, conditional on vehicle choiće. The solution of such a model yields:
GS = g( Pg/e, YH, UN). E
(1)
which gives the gasoline consumed per car (GS), vvhere: GS = MS*E (vehicle miles х miles per gallon); Pg/e = priče of gasoline per mile, which is the outcome of the priče of gasoline per gallon divided by the fuel efficiency given in terms of miles per gallon; YH = household disposable income; UN = unemployment rate. Furthermore, previous studies have utilised a number of demographic variables. The following demographic variables are included in the model: the percentage of population of driving age (16-65) (POP), and the number of cars per household (AH). Then, the log-linear relationship for equation (1) can be written as: LnGS = A + Bl Ln Pg/e + B2 Ln YH + ВЗ Ln UN + B4 Ln AH + B5 Ln E (2) The stock of cars per household equation
In the literature, the lagged values of the stock of cars, along with the cost of gasoline per mile and the average priče of new cars, have been used as explanatory variables for the stock holding decision. Also, an income per household variable, 17/, and the percentage of population of driving age, POP, are ali utilised. The basic equation for estimating the car holdings per household is given as:
(S/H) = g[Pn9 Pg, (S/H)t_l9 YHy POP]
(3)
where:
S/H = stock of cars per household or (s); Pn = priče of new cars; Pg = priče of gasoline per gallon. The log-linear functional form will be: Ln s = C + D1 Ln Pn + Dl Ln Pg + D3 Ln YH + DA Ln st_x + D5 Ln POP (4) The new car sales per household equation
In this sub-section, the household's decision to buy a new car is modelled. The basic relationship is given by:

194

Transport Gasoline Demand in Canada
M. N. Eltony

(NR/H) = Д Pn, Pg/e, YH, UN)
(5)
where NR/H = new car sales per household or (nr). However, since strikes in the car industry may delay the purchase of a new vehicle, a variable which represents the man-days lost in the car industry due to strikes, ST, is included. Assuming a log-linear functional form, equation (5) gives the following relation: Ln nr = E + Fl Ln Pg/e + F2 Ln Pn + F3 Ln YH + F4 Ln UN + F5 Ln ST (6) The new car fuel-efficiency equation
The fuel есопоту of new cars is defined by the sales-weighted average as follows :
EN = JTENj^Nj/NR
(Ђ
j where: ENj = technical fuel есопоту for the ;-th siže class of cars;[2] Nj/NR= ratio of cars j sold to total new sales in Canada. The relationship in the equation above identifies the two determinants of a change in fuel есопоту: first, the change in the technology of the vehicles; and, second, the change in the distribution of new car sales by siže. The first change is determined by the manufacturers and is the subject of the current sub-section. The second is the result of household preference and is discussed in the next sub-section. A comparison of the fuel есопоту of different classes of cars by their siže (interior volume) rather than their weight has been suggested by both the US Environment Protection Agency (EPA) (1984) and the Society of Automotive Engineers (1985). Both argued that the consumers are probably more interested in the siže and utility of a vehicle rather than its vveight. The EPA developed a new vehicle siže classification based on the interior volume of the automobile which can be used to compare the fuel есопоту of various vehicles. Table 1 gives a description of these classes. The recent use of new ultra-light materials in the manufacturing of cars has made it possible to build a vehicle with more, or at least the same, interior space but with less weight and thus a more fuel-efficient engine. These classes capture such technological advances rather well. Because of the relati vely small Canadian market, car manufacturers are assumed to be primarily concerned with economic conditions in the United States, especially with regard to gasoline prices. Car manufacturers are assumed to make decisions on the design of four interior volume classes of new cars several years prior to the commercialisation of the model. The fuel есопоту chosen for the future model is that level which maximises the present value of revenue minus costs. The manufacturers believe that consumers' demand
Table 1

The Interior Volume Classes ofNew Cars

|Туре* |FP |M3 |
|1. Sub-Compact |3.40 |

* Units are in Cubic Feet & Cubic Metres.

Table 2

Fuel Есопоту Standards In The United States

Year Standard

|1978 |18 mpg |
|1979 |19 mpg |
|1980 |20 mpg |
|1981 |22 mpg |
|1982 |24 mpg |
|1983 |26 mpg |
|1984 |27 mpg |
|1985 and after |27.5 mpg |

for ftiel efflciency in уеаг t depends upon the priče of gasoline that уеаг subject to a particular siže car. Therefore, different lag structures were used for the gasoline priče in the US. A polynomial distributed lag was found to be the most successful. Furthermore, in order to capture the costs of designing a more fuel-efficient car in the interior volume classj9 dummy variables for the siže classes Zj j = 1 ,...,4 are imposed. In 1975 fuel efficiency standards were passed in the US vvhich set sales-weighted fuel есопоту levels for 1978-1990. These are shown in Table 2. Failure to meet these standards meant a fmancial penalty.[3] To capture the response by producers to these standards, several dummy variables were tested. The new car fuel efficiency equation in the log-linear functional form is given as : LnENj = KlZj + K2DST + K3LnPgt_l + K4LnFgt_2 + K5LnPgt_3 + K6LnPgt^ (8)
Transport Gasoline Demand in Canada
M. N. Eltony

where: Zj = dummy variable for each ;th interior volume class; DST = dummy variable for fuel efficiency standards 1975-1985; Pgt_i = US gasoline prices lagged 1,2, 3 and 4 time periods. At this point two important identities should be introduced. The fuel есопоту of the fleet, E, is defined as the harmonic mean of the new car fuel есопоту, EN, and the fuel есопоту of the last year's stock, t/C.
E = EN. NR/S + Et_x. UC/S (9) The proportions of new cars, NR/S, and used cars, UC/S, can be determined from the follovving relationship:
5 = NR + UC (10) which simply states that the addition of used cars, UC, and new cars, NR, is equal to the current stock of cars in the fleet. In equation (4) the stock of cars in the fleet was specified, while in equation (6) the volume of new car sales was specified. Therefore, the proportions of new car sales and used cars, over the fleet, can be determined by the following equation: NR/S = NR/(NR+UQ and UC/S = UC/(NR+ UQ=l-(NR/S) (11)

The sales ratio of nevv cars
Following the previous discussion, there are four classes of new cars to choose from: Sub-Compact, Compact, Mid-size, and Large. Because there are four alternatives available, of which one is chosen by the consumer, the decision can best be modelled in the framework of the multinomial quantitive choice model (McFadden, 1973). In this kind of model, the household is assumed to choose among several alternatives and the decision depends upon characteristics of the household and of the alternatives. The objective of the model is to provide a prediction of the probability that a household with particular characteristics will choose one type of car over another. Let the ratio of the probability of choosing alternative z to the probability of choosing alternative х by household i be Pg/Ppc. Let КГ denote a bundle of characteristics of the household and Lz represent characteristics of the alternative car types. Then, the model of new car choice can be described by the following equation:
P-Z/PjX = e^z + ^z ^ + CLz/ gAx + BxKi + CLx (12) Taking the logarithms of both sides of the above equation yields
Ln (PiZ/Pp) = (Az-Ах) + (Bz-Bx) Ki + C(Lz-Lx) for z = 2, 3,4 (13) Data on the basis of household choice of the type of new car are not readily available; however, the probabilities can be substituted by the relati ve frequencies of the households with the bundle of attributes Kt choosing alternative z. Then substituting the relative frequencies for the probabilities and suppressing the household subscript yields the following equation:
Ln(Nz/Nx) = (Az-Ах) + (Bz-Вх) K + C(Lz-Lx) forz = 2,3,4 (14) In order to be able to estimate the equations in a manner similar to logit estimation (Amemiya, 1981), the coefficient on the characteristics of the vehicle variables, L, are constrained to be equal across the equations.[4] In the estimation of the equations the household disposable income, the regional unemployment rate, the number of man-days lost as a result of strikes in the automotive industry and the percentage of driving age population are the characteristics of the household, KL The difference in the car prices, (Pnz - Pnx)> and in the gasoline cost per mile, Pg (1/ enz - \lenx)y are the vehicle type characteristics, vvhere enz and епх are the fuel efficiency of vehicle type z and type х respectively. Equation (14) can be re-stated as follovvs: Ln(Nz/Nx) = A + BYH+CUN+DST+EPOP +

F (Pnz - Рпх) + G Pg(\lenz - 1/епх) for z = 2, 3,4 (15) where: A = (Az-Ax) B = (Bz- Вх) C=(Cz-Cx) D = (Dz-Dx) E=(Ez-Ex)

3. The Data

Pooled time-series and cross-section data on the Canadian provinces from 1969-1988 were used. Ali prices and income are expressed in relation to the consumer priče index (1981 = 100). Reports published by Statistics Canada provided the data on car stocks, regional unemployment rates, net gasoline sales,[5] the priče of gasoline at the pump in several major Canadian cities, new car registration, residential population in urban areas, driving age population, prime interest rate, number of households by province and the consumer priče index in major cities. The household income was obtained from Statistics Canada' s Household Expenditure Survey. The number of cars per household was obtained

Transport Gasoline Demand in Canada
M.N.Eltony

from the Household Facilities and Equipment Survey. The average fuel-efficiency data were obtained from Statistics Canada' s Household Fuel Consumption Survey. The earlier fuel efficiency series datawere gathered from the Canadian Automobile Survey, the Economic and Technical Review Report, published by Environment Canada. The priče of new cars from four categories, classified by the interior volume of the vehicle, are weighted averages of the prices of the four largest sellers in Canada for that уеаг. The source of these prices was the Canadian Golden Book ofUsed Car Prices 1968-1989. Sales figures for ali models of cars in Canada were made available to this study by R. L. Polk & Co. (Toronto). The sales data were used to construct the sales ratios for fuel categories of cars classified by their interior volume. The fuel economies for these four classes are published annually in the Gas Mileage Guide (EPA, US Department of Energy). The fuel economies data and the sales ratios for the different classes were used to create the sales-weighted average fuel есопоту for new cars (EN) as defined in Equation (7). Several issues of the United States Statistical Abstract 1968-1988 were used as the source for the information on gasoline prices and the consumer priče index in the US.

4. The Estimation Results

A variance components model (Pindyck and Rubinfeld, 1976), which allovvs separate provincial intercepts, is assumed using pooled time-series and cross-sectional provincial data. The estimation procedure employed for most of the equations in the model is Ordinary Least Square (OLS) with auto-correlation correction when necessary.

Estimation of gasoline per car
Equation (1) in Table 3 gives the results for this estimation. The signs of the estimated coefficient on income, unemployment rate, priče of gasoline per mile, fuel efficiency of the fleet and the number of vehicles per household are consistent with economic theory. The priče of gasoline per mile, the fuel есопоту of the fleet, the household disposable income, the unemployment rate and the number of vehicles per household are ali significantly different from zero at the 99 per cent level. The provincial intercept terms are also significant: they account for the differences among provinces vvhich cannot be quantified, that is, the availability of public transport systems, degree of urbanisation and so on. The short-run gasoline priče elasticities per car, holding the fuel есопоту constant, is approximately -0.21. The results give a short-run income elasticity of approximately 0.15.

Estimation of stock of cars per household
The result for this estimation is reported in Table 3 as Equation (2). The priče of gasoline per gallon was found empirically to be more successful than the priče per mile for the stock of cars equation. However, when the priče of gasoline per gallon, lagged one time period, was ineluded better results were obtained. The coefficient for the cost of gasoline per gallon has a negative sign as a refleetion of the fact that gasoline and cars are complementary goods. Researchers have reported that the priče of gasoline and the priče of new cars tend to be correlated. This is because other characteristics of the car which are related to fuel efficiency, such as povver and siže, are not accounted for. Because of this problem of collinearity between the priče of gasoline and the priče of new cars, none of the tested specifications produced a statistically significant coefficient for the priče of new cars; nevertheless it has a negative sign. One possible reason why the new car priče did not behave well in the stock equation is that the new car priče is not the flow cost of the services, which depends on the financing of the car, and is approximately the real interest rate times the priče of the car, plus maintenance costs. But if the real interest rate is relatively constant and maintenance costs are a constant proportion of the car priče over the estimation period, then the priče of new cars represents a good ргоху for the desired variable. The prime interest rate in Canada was ineluded in an alternative specification. Its estimated coefficient has the right sign and it was found to be significant but the priče of new cars remained insignificant. Moreover, the gasoline priče elasticities of the stock of cars per household were found to be small, approximately 0.12. It should be noted that this is a short-run elasticity since in the long run, the household could switch to smaller, more fuel-efficient cars.

Estimation of nevv car sales
The results of the new car sales estimation are given in Table 3 as Equation (3). The coefficient of the new car prices variable has a negative sign and is highly significant, as is the coefficient of the gasoline cost per mile. There is no indication of a collinearity problem between the two prices. Income per household has a positive sign, confirming that cars are normal goods. The coefficient for the unemployment rate (UN) has a negative sign so that at lovver levels of unemployment, new car sales are expected to inerease. The unemployment rate variable (UN) has been ineluded as an indicator of cyclical economic conditions. It provides information about the phase of the business cycle and the expected income which would not be provided by the inelusion of only the disposable income variable. The variable for strikes in the car industry (ST) represents a supply constraint on the availability of new cars, and the coefficient for this variable has a negative sign. Some researchers have argued that the coefficient of this variable should be zero since апу problems with the availability of domestic cars could be made upby an inerease in imports. This is certainly plausible. Hovvever, if imported cars are not a perfeet substitute for domestic ones, then an effect is still expected.

Estimation of fuel efficiency of new cars
The results for this equation are given in Table 4. A polynomial distributed lag of degree 1 is imposed on the gasoline priče with zero restriction on the coefficient of the current

200
Transport Gasoline Demand in Canada M. N.. Eltony

Table 3

Results of Estimations for Canadian Provinces (t-statistics in parentheses)

Equations

period priče. On the basis of both R2 and Minimum Standard Error criteria, the optimum lag length is 4. This indicates that design changes are made 1 to 4 years prior to the уеаг of marketing the final produet. There is evidence from the industry to support this finding. According to the International Business Week Magazine,[6] several car manufacturers spend about 3 to 4 years in designing a new model before the commercialisation уеаг. The coefficients of the gasoline priče per gallon are of almost the same siže, indicating that the design of the new model can be altered up to the last уеаг before the marketing. Further, the gasoline priče elasticity of nevv car fuel efficiency is about 0.8 over the four уеаг design period. In Table 4, the results for the dummy variables for siže classes (Zl,..., ZA) illustrate that the larger the interior volume of the car, the lower the fuel есопоту of the new car. Also, the dummy variable for the fuel есопоту standards indicates a positi ve effect on the fuel efficiency of nevv cars of ali classes.

Estimation of sales ratio of new cars
Three equations with cross restrictions on the estimates of the new car priče variable (Pnz - Рпх) and gasoline cost per mile variable Pg (l/enz - 1/епх) are estimated simultaneously. The three equations were estimated for the sales in categories 2,3 and 4 relative to the sales in category 1. The estimation results for the nevv car sales ratios are given in Table 5. The results shovv that the larger the difference betvveen the priče of new cars in the larger categories and in
Transport Gasoline Demand in Canada M. N. Eltonv

Table 5 Sales Ratios for the New Cars

|Categories |2 |3 |4 |
|Pg(l/enz-l/enx) |-11.738 |-11.738 |-11.738 |
| |(-8.105) |(-5.272) |(-5.996) |
|Pnz - РПХ |-0.1073* IO"3 |-0-.1073*10"3 |-0.1073 *10"3 |
| |(-4.684) |(-3.047) |(-3.465) |
|YH |0.56133 *10"4 |0.6720* IO"4 |0.19093*10^ |
| |(5.728) |(4.496) |(1.876) |
|UN |0.0428 |0.0668 |-0.0443 |
| |(3.582) |(3.629) |(-2.676) |
|ST |-0.14345*10"* |-0.99303* IO"7 |-0.41328 *10-7 |
| |(-3.109) |(-2.409) |(-1.656) |
|POP |-0.08312 |-0.20643 |-0.1797 |
| |(-6.671) |(-10.613) |(-10.634) |
|NFLD |3.6621 |11.482 |12.327 |
| |(6.333) |(11.98) |(14.101) |
|PEI |4.3113 |12.873 |13.257 |
| |(6.324) |(11.624) |(13.335) |
|NS |4.3874 |13.040 |13.163 |
| |(6.261) |(11.469) |(12.903) |
|NB |4.2971 |12.837 |13.465 |
| |(6.399) |(11.737) |(13.686) |
|QUE |4.4644 |13.224 |13.553 |
| |(6.292) |(11.482) |(13.086) |
|ONT |4.4766 |13.423 |13.703 |
| |(6.189) |(11.452) |(12.935) |
|MAN |4.4798 |13.568 |13.775 |
| |(6.128) |(11.493) |(12.98) |
|SASK |4.4935 |13.426 |13.899 |
| |(6.175) |(11.444) |(13.184) |
|ALTA |4.1269 |12.710 |13.316 |
| |(5.926) |(11.256) |(12.993) |
|BC |4.2509 |12.585 |12.926 |
| |(6.033) |(10.973) |(12.442) | category 1 (the smallest siže), the smalleris the ratio oflarger to small car sales. Similarly, as the gasoline cost inereases, the sales of smaller cars become more frequent relative to sales oflarger ones. The coefficients on both variables of car prices and gasoline cost are significantly different from zero at the 99 per cent level. The results also suggest that as household disposable income rises, the number of cars sold in categories 2, 3, and 4 relative to sales in category 1 rises, which is evidenced by the positive sign for the household income variable in ali three equations. A rise in the unemployment rate induces households to move down the siže speetrum, decreasing the ratio of car sales in each category relative to the smallest cars. Furthermore, man-days spent on strike in the Canadian car industry appear to have a negative effect on the sales in categories 2, 3, and 4 relative to category 1. This result is intuitively appealing since a large proportion of the cars in the smallest category are imported, and thus would not be subject to domestic strikes. A rise in the driving age population, POP, also has a negative effect on the sales in category 2, 3, and 4 relative to sales in category 1. Finally, ali the household characteristics variables are statistically significant at the 95 per cent level.

Priče elasticity estimates
The objeetive of developing the model was to determine the household response to gasoline priče changes. A widely used economic indicator of consumer response is the priče elasticity of demand for gasoline. Ву simulating the model over the desired time horizon, 1989-2000, the priče elasticities were determined. A base case was specified in which real household income, the unemployment rate, the real priče of new cars, the interest rate, and the real priče of gasoline per gallon in Canada and the United States are assumed to equal the 1988 values and remain constant for the rest of the time horizon. In an alternative solution of the model only the real prices of gasoline in Canada and the US are assumed to inerease by 10 per cent. The two dynamically controlled solutions of the model were obtained and the percentage change in gasoline consumption computed from the two solutions. Dividing the result for each уеаг by 0.1 yields the dynamic priče elasticity. The priče elasticity as described above ineludes ali direct and indirect effects on gasoline consumption resulting from a 10 per cent change in the priče of gasoline per gallon. The inerease in the priče of gasoline has a direct effect on the number of miles driven per car, the average fuel есопоту of the fleet, new car sales, and on the stock of cars. AH these effects are captured in the priče elasticity estimates. Table 6 gives the short-run (one уеаг) priče elasticity and the longer-run priče elasticity estimates for two to ten years. Because the intercept term is the only parameter that is allowed to vary across provinces, there is little variation in the priče elasticities across provinces; therefore only results for Canada as a whole are reported in Table 6. The short-run (one уеаг) elasticities appear larger than expected. We have noted from Table 3's equation (1) that the direct response to an inerease in the priče of gasoline, holding the stock of cars and the fuel есопоту of the fleet constant,

204
Transport Gasoline Demand in Canada

Table 6

Dynamic Priče Elasticities of Gasoline Demand in Canada
M.N. Eltony

was -0.21. When changes in the ftiel есопоту of the vehicles on the road and fleet siže through new car sales and the scrapping of used cars are included, -0.1 is added to the priče elasticity. This is higher than Gallini' s result of-0.06 and indicates that at least 25 per cent of the decrease in gasoline consumption in the first уеаг results from changes in the average fuel есопоту of the fleet, adjustment to car stocks, and new car sales. The adjustment to the priče increase appears to be rapid, which is indicated by the rise in the priče elasticity within the first four years after a priče change. Ten years after the 10 per cent increase in the gasoline priče, the reduction in consumption settles to approximately 10 per cent of the consumption level without a priče increase.

5. Conclusions

The model presented in this study is one of the few econometric studies that has attempted to model gasoline demand for Canada. The attempt was made to improve upon the existing models through careful description of the underlying decision-making process that faces the household, making the household rather than the indi vidual the focus of the model and significantly extending the time-series beyond the scope of existing studies. In the model, the household which already owns a car can react immediately to a priče increase by driving fewer miles. The household which is planning to buy a new car can either postpone their vehicle purchases or choose a more fuel-efficient new car. Finally, the household which owns an aged car can scrap it in response to a higher gasoline priče. In the long run, the siže and composition of the fleet according to the interior volume of the vehicles can continue to change and necessary miles may fali as households move closer to work. Also, in the long run, car manufacturers can modify the technology of the new cars according to their expectations regarding the future levels of gasoline prices and consumers' demand for more fuel-efficient cars. Another contribution of the model to the literature in addition to the use of the household expenditure survey data is the detailed treatment of the fuel efficiency of the new cars, where they were categorised according to their interior volume rather than their natural weight. The results of estimating the model provided revealing information about elasticity of gasoline demand in Canada. The short-run dynamic own-price elasticity of gasoline' demand ranged between 0.311 to 0.313 in absolute value across the provinces. Almost 75 per cent of the household response to priče change in the first уеаг was accounted for by driving fewer miles. While these results are in line with Gallini (1983), they exhibit one different feature: at least 10 per cent of the response in the first уеаг results from an alteration in the composition of the fleet to more fuel-efficient vehicles. This response accounted for only 4 per cent in Gallini's study. The remaining 15 per cent is attributed to the change in the siže of the fleet. In the intermediate term (five years) priče elasticities range from 0.689 to 0.709, and the long-term elasticities (ten years) range from 0.975 to 1.059. Moreover, the dynamic elasticities imply that the adjustment seems to take place very rapidly during the first four years. The results point to the importance of improving fuel efficiency as an effective means of reducing household gasoline consumption. They also show that the detailed treatment of the fuel efficiency of the fleet was justified. As an explanatory variable, the fuel efficiency variable was found to possess the right sign and to be highly significant in ali equations. The results indicate that in many cases the households were able to keep their favourite car siže because the manufacturers instituted significant improvements in fuel efficiency in new cars of ali classes.

References

Amemiya, T. (1981): "Qualitative Response Models: A Survey". Journal ofEconomic Literature, December, pp. 1483-530.
Bacon, R. W. (1990): RocketsAnd Features: Ћге Asynvnetric Speed Of Adjustment OfU.K. Retail Gasoline Prices To Cost Changes. Oxford Institute Of Energy Studies. Balestra, P. and M. Nerlove (1986): "Pooling Cross Sections and Time Series Data in the Estimation of a Dynamic Model: The Demand for Natural Gas". Econometrica, vol. 34, pp. 586-612. Baltagi, B. H. and J. M. Griffen (1983): "Gasoline Demand in the OECD: an Application of Pooling and Testing Procedures". European Economics Review, vol. 22, pp. 137-77. Baltagi,B. H. and J. M. Griffen (1984): "U.S. Gasoline Demand: WhatNext?" TheEnergy Journal, vol. 5, no. l,pp. 129-39.
Becker, G. S. (1965): "A Тпеогу of Allocation of Time". Economic Journal, vol. 75, pp. 493-517. Berkovvitz, M., N. T. Gallini, E. Miller and R. Wolfe (1990): "Disaggregate Analysis of the Demand for Gasoline". Canadian Journal of Economics, vol. 45, pp. 253-75.

206
Transport Gasoline Demand in Canada
M.N.Eltony

Blomqvist, A. and W. Haessel (1978): "SmallCars, Large Cars, and the Priče of Gasoline". Canadian Journal of Economies, pp. 470-89.
Burright, B. and J. Enns (1975): Econometric Models of Demand for Motor Fuel. Prepared for the National Science Foundation and Federal Energy Administration R-1561-NSF/FEA, The Rand Corporation.
Craig, P. (1984): "Fuel Efficient Incentives for Cars Oil Import Vulnerability Reduction". Energy Policy Forum, The Energy Journal, vol. 5, no. 1, pp. 141-48. Dahl, C. A. (1978): "American Energy Consumption, Extravagantor Economical: ACase Study of Gasoline Demand". Resources andEnergy, pp. 359-73. Dahl, C. A. (1979): "Consumer Adjustment to Gasoline Тах". Revievv of Economies and Statistics, pp. 426- 32.
Dahl, C. A. (1982): "Do Gasoline Demand Elasticities Vary?" Land Economies, pp. 373-82.
Dahl, C. A. (1986): "Gasoline Demand Survev". TheEnergy Journal, vol. 7, no. 1, pp. 67-82.
De Jong, G. C. (1990): "An Indirect Utility Model Of Car Ownership And Private Car Use". European Economic Review, vol. 34, pp. 971-85. Dewees, D., R. Hyndman and L. Waverman (1975): "The Demand for Motor Gasoline in Canada". Energy Policy, pp. 116-23.
Drollas, L. P. (1984): "The Demand for Gasoline: Further Evidence". Energy Economies, Јапиагу, pp. 71-82.
Flaig, G. (1990): "Household Production And The Short and Long-run Demand for Electricity". Energy Economies, April, pp. 116-21. Gallini, N. T. (1983): "Demand for Gasoline in Canada". Canadian Journal of Economies, pp. 299-324. Garbacz, C. (1989): "Gasoline, Diesel and Motorfuel Demand in Taiwan". The Energy Journal, vol. 10, no. 2, pp. 153-63.
Houthakkar, H., P. Verleger and D. Sheehan (1974): "Dvnamic Demand Analysis for Gasoline and Residential Electricity". American Journal of Agricultural Economies, vol. 56, pp. 412-18.
Hsing, Y. (1990): "On the Variable Elasticity of the Demand for Gasoline: The Case of the USA". Energy Economies, vol. 12, no. 2, April, pp. 132-36.
Kouris, G. (1983): "Fuel Consumption for Road Transport in U.S. A." Energy Economies, vol. 5, pp. 89-99.
Lancaster, K. J. (1966): "A New Approach to Consumer Theorv". Journal of PoliticalEconomy, vol. 74, no. 2, April, pp. 132-57.
Lin, A., E. N. Botsas and S. A. Monroe (1985): "State Gasoline Consumption in the U.S.A. An Econometric Analysis". Energy Economies, vol. 7, pp. 29-36. Manning, D. N. (1988): "Household Demand For Energy In The UK". Energy Economies, Јапиагу, pp. 59- 78.
Mannering, F. and K. Train (1985): "RecentDirections in Automobile Demand Modeling". Transportation Research B vol. 19, pp. 265-74. McFadden, D. (1973): "Conditional Logit Analysis of Quantitive Choice Behaviour". In Zarembka (ed.): Frontiers in Econometrics, New York: Academic Press. Mehta, J., L. Marasimham and A. V. Swamy (1978): "Estimation of a Dynamic Demand Function for Gasoline with Different Schemes of Parameter Variations". Journal of Econometrics, vol. 7, pp. 263- 79.
Parbhakar, K. J. (1986): "Fuel consumption for road transport in Quebec". Energy Economies, vol. 8, pp. 165-70.
Pindvck, R. S. (1979): The Structure ofWorld Energy Demand. Cambridge, Mass.: MIT Press. Pindyck, R. S. and Rubinfeld (1979): Econometric Models and Economic Forecasts. McGraw-Hill. Pollack, R. A. and M. L. Wachter (1975): "The Relevence of Household Production Function and its Implications for the Allocatiort of Time". Journal ofPolitical Есопоту, vol. 83, pp. 255-77. Ramsey, J., R. Raasche and B. Allen (1975): "An Analysis of Private and Commercial Demand for Gasoline". Review of Economies and Statistics, vol. 57, pp. 502-7. Rice, P. and P. Frater (1989): "The Demand for Petrol with Explicit New Car Fuel Efficiency Effects - A UK Study 1977-86". Energy Economies, April, pp. 95-104.
Shalaby, A., and R. Waghmare (1980): "AModel for Forecasting Passenger Car Gasoline Demand". Energy DemandModelling. (Model made for the Canadian National Energy Board.)
Simpson, B. H. (1984): US Passenger Car Fuel Есопоту. Society of Automotive Engineers, C441/84, pp.13-28.
Sweeney, J. L. (1979): "Effects of Federal Policies on Gasoline Consumption". Resources and Energy 2, North-Holland, pp. 3-26.
Wabe, J. S. (1987): "The Demand forGasoline: ACommenton Cross-Section Analysis". Energy Economics, vol.9,pp. 287-89.
Wasserfallen W. and H. Guntensperger (1988): "Gasoline Consumption and the Stock of Motor Vehicles: An Empirical Analysis for the Swiss Есопоту". Energy Economics, vol. 10, no. 4, pp. 276-82. Westin, R. (1975): "Empirical Implications of Infrequent Purchase Behavior in a Stock Adjustment Model". American Economic Review, vol. 65, pp. 384-95. Wildhorn, S., B. Burright, J. Enns and T. Kirkwood (1976): HowtoSave Gasoline: Public PolicyAlternatives. Ballinger Publishing Co., Cambridge, Mass. Willett, K. D. and S. Naghshpour (1987): "Residential Demand for Energy Commodities: A Household Production Function Approach". Energy Economics, vol. 9, no. 4, pp. 251-56. Wykoff, F. (1973): "A User Cost Approach to New Automobile Purchases". Review ofEconomic Studies, vol. 40, pp. 377-90.
Yang, B. and T. Hu (1984): "Gasoline Demand and Supply Under a Disequilibrium Market". Energy Economics, vol. 6, pp. 276-82.

Date ofreceipt offinal typescript: July 1992

208
-----------------------
[1] Assistant Professor, College of Commerce and Economics, Kuwait University.
[2] The technical fuel есопоту of cars is the fuel есопоту under test driving conditions. A reliable source for this information is provided by the Environment Protection Agency's publications for ali models of new cars.
[3] The penalty for not achieving the standard is $5.00 per vehicle for each tenth of a mile per gallon below the mandated level. Hovvever, because of the failure of the 1985-1987 model уеаг cars to achieve these standards, the EPA has modified the standards for 1987,1988,1989 and 1990 to 26,26,26.5 and 27.5 mpg respectively.
196
[4] To explain this point further, consider the follovving set of equations: (i) Ln (N2/NI) = (A2 - Л1) + (B2 - BI) K + C21 (L2 - LI)
(u) Ln (N3/NI) = (ЛЗ -Л1) + (ВЗ-BI) АГ+ C31 (L3-L1)
(iii) Ln (N4/N1) = (Л4-Д1) + (B4-Bl)K=C4l (IA-LI) From (i) and (ii), Ln (N2/N3) can be found. That is
(iv) Ln (ШЗ) = (Л2 -Л1) + (B2-Bl)K+C2l (L2-L1) - (ЛЗ -Л1) - (ВЗ-Bl) К-С31 (L3 - LI) If C31 = C21 = C then
(v) Ln (N2№) = (Л2-ЛЗ) + (B2 -ВЗ) К+С(L2 -L3)
The same argument holds for Ln (N2/N4) from (i) and (iii) and ali other possible combinations. See Gallini (1983), p. 305, on this point.
[5] For example, the taxable gasoline which is primarily sold to the drivers of cars, buses and trucks at the gas pump.
198
[6] See: 'The Potholes In Ford' s Road to Riches," 27 November 1989; "For Hyundai, There is No Place Like Home," 20 November 1989; "Infinite and Lexus: Characters in German Nightmare," 9 October 1989; "Japanese Carmakers Flash Their Cash," 13 February 1989; "How Ford And Mazda Shared The Driver's Seat," 26March 1990 . In The International Business Week Magazine, McGraw-Hill Publications.
202

-----------------------
|NFLD |7.2006 |(10.61) |0.2073 |(0.2381) |7.8894 |(4.329) |
|PEI |7.2234 |(10.86) |0.3284 |(0.3754) |7.7907 |(4.284) |
|NS |7.1705 |(10.75) |0.2638 |(0.3019) |7.8021 |(4.286) |
|NB |7.2336 |(10.90) |0.2814 |(0.3216) |7.8878 |(4.335) |
|QUE |7.0088 |(10.44) |0.2256 |(0.2570) |7.8721 |(4.305) |
|ONT |7.0179 |(10.26) |0.2508 |(0.2844) |7.6989 |(4.162) |
|MAN |6.9269 |(10.26) |0.2828 |(0.3215) |7.4788 |(4.065) |
|SASK |7.0283 |(10.44) |0.2475 |(0.2812) |7.3571 |(3.993) |
|ALTA |7.1069 |(10.44) |0.3143 |(0.3547) |7.5888 |(4.086) |
|BC |6.9064 |(10.23) |0.2920 |(0.3310) |7.4933 |(4.068) |
|LnF# |0.1466 |(2.045) |0.1402 |(2.1061) |0.3453 |(2.9921) |
|hnPg/e |-0.2095 |(-3.971) | | |-0.2726 |(-3.002) |
|Ln£ |-0.7420 |(-11.319) | | | | |
|Ln UN |-0.0495 |(-2.298) | | |-0.2956 |(-6.975) |
|Ln AH |-0.4714 |(-7.599) | | | | |
|Ln Pgt_\ | | |-0.1206 |(-3.029) |-0.7179 |(-4.985) |
|Ln Pn | | |-0.0297 |(-1.402) | | |
|Ln st_x | | |0.4955 |(7.789) | | |
|Ln POP | | |0.3906 |(1.825) | | |
|Ln ST | | | | |-0.00395 |(-2.451) |
|R2 |0.9074 | |0.8125 | |0.7655 | |
|SER |0.0500 | |0.0555 | |0.1054 | |
|DW |2.0927 | |2.0670 | |2.0342 | |
|F |130.36 | |62.59 | |41.19 | |
|The notation for the |provinces is: NFLD = Newfoundland; PEI |= Prince Edward Island; NS = |Nova Scotia; |
|NB = New Brunswick; QUE = |:Quebec;ONT = |Ontario; MAN: |= Manitoba; SASK= |= Saskatchewan; ALTA |

(l)Ln GS (2)Lns (3)Lnw

= Alberta; BC = Britisb Columbia.

| |Zl |Z2 |Z3 |Z4 | |
|LnEN |4.0061 |3.9380 |3.7232 |3.5491 | |
| |(29.763) |(29.583) |(27.969) |(26.662) | |
| |DST |Pt-l |Pt-2 |Pt-3 |PtA |
|LnEN |0.082372 |0.20138 |0.19829 |0.19519 |0.19209 |
| |(2.0329) |(3.0178) |(5.3147) |(7.9252) |(4.1945) |
| |/?2 = 0.9991 | |SER = 0.0091 |F = |9621.9 |

Table 4

The New Cars Technical Fuel Efficiency Equation
(t-statistics in parentheses)

|1 |0.3120 |
|2 |0.4673 |
|3 |0.5370 |
|4 |0.5981 |
|5 |0.6984 |
|6 |0.8132 |
|7 |0.8935 |
|8 |0.9478 |
|9 |0.9839 |
|10 |1.0073 |
|11 |1.0192 |
|12 |1.0239 |

Year