ESTIMATING ELASTICITIES OF HOUSEHOLD DEMAND FOR FUELS FROM CHOICE ELASTICITIES BASED ON STATED PREFERENCE

Transcription

1 ESTIMATING ELASTICITIES OF HOUSEHOLD DEMAND FOR FUELS FROM CHOICE ELASTICITIES BASED ON STATED PREFERENCE Zeenat ABDOOLAKHAN Information Management and Transport School of Economics and Commerce, M261 University of Western Australia 6009 Abstract: Potential responsiveness to new vehicle technology and fuels, including ethanol, biodiesel and various hybrid alternatives, has been explored in a stated choice survey. Respondents chose between four hypothetical fuel and vehicle alternatives in each of nine scenarios. The results have been modelled with nested logit (NL) and choice elasticities calculated. The NL model has also been re-estimated with endogenous weights to match actual and expected consumption distributions and the corresponding choice elasticities derived. Only in the case of work trips can choice elasticities generally be interpreted as demand elasticities. Otherwise the choice element is only one part of demand. A particular demand elasticity may be similar to the corresponding choice elasticity while in other cases they will differ substantially, the difference being due to the income effect. The choice elasticities have been modified to restore symmetry and zero share weighted column sums. The respective demand elasticities have then been derived by adding the calculated income term for each fuel. The changes in household demand elasticities depend on two factors: the fuel share and the income effect applied to the elasticity. Generally when the shares are relatively small, the own-choice and own-price elasticities are similar. This is due to the fact that when the share is multiplied by group elasticity, the result is very little difference between choice and demand elasticities. In the case of petrol, because it is the dominant product in the group, its own-price elasticity differs greatly from the choice elasticity. Keywords: Choice, Demand, Elasticity, Income effect

2 1. INTRODUCTION Current changes in vehicle technology and fuels seem to be leading in the short term to a fleet still powered by internal combustion engines with either conventional or hybrid drive. Already available fuels such as LPG, biodiesel and ethanol favour the internal combustion engine. However unfamiliarity with the alternatives seems to create an inertia effect retarding the move away from the traditional petrol car. In Australia, the most common alternative fuel is liquefied petroleum gas (LPG) but the uptake of LPG by private motorists has been limited despite the demonstrable savings. The potential responsiveness to new vehicle technology and fuels was explored in a stated choice survey (Abdoolakhan, 2006) in which respondents chose between four hypothetical fuel and vehicle alternatives in each of nine scenarios. The survey design was D-optimal. A sample scenario is shown in the Appendix; 231 respondents provided usable data, giving 2039 observations. The structure of the nested logit (NL) model used to analyse the responses is shown in Figure 1. Conventional fuels Alternative fuels Petrol Diesel Hybrid Petrol E20 B20 Hybrid E20 Hybrid LPG Dual LPG Dedicated LPG Figure 1: Nested Logit (NL) Tree Structure 2 CHOICE ELASTICITIES When elasticities are derived from the coefficient estimates, the probability weighted averaging tends to reduce the rigid equality of cross choice elasticities while the branching structure of NL results in the average being the sum of a branch elasticity and a choice elasticity within the branch (Hensher et al, 2005). Cross choice elasticities in different branches may differ substantially. Since the focus is on choice of vehicle and fuel type with respect to fuel price, combined elasticity values have been used for the car types which run on the same fuels. For non-nested alternatives, changes in attributes result in the direct and cross-elasticities being identical to those of the MNL model. This is not the case for changes in the attributes of nested alternatives, the differences being attributable to the value of the scale parameter, θ in the elasticity equation (Koppelman et al, 2006). When θ is equal to one, its maximum value, the Ê1-q ˆ N expression, Á in the elasticity formulae becomes zero and the direct and crosselasticity expressions for the nested alternatives collapse to the corresponding MNL Ë qn expressions. This is the case with the Conventional branch since θ was set to 1 for ease of interpretation. The direct and cross-elasticities of Petrol, Diesel and Hybrid Petrol from the NL model are therefore those of a simple MNL model. 2

3 In the combined choice elasticity results of Table 1 the own-elasticities of petrol and diesel indicate that respondents are more sensitive to the price of diesel. An increase in the price of petrol has a greater impact on the probability of choosing conventional petrol cars than hybrid petrol. This is sensible due to the fact that the hybrid petrol is a more efficient technology. The same observation can be made for Hybrid E20 and E20 whereby E20 is more elastic. Table 1: Choice elasticities with respect to five fuels Petrol Diesel E20 B20 LPG Conventional Petrol Diesel Hybrid Petrol E B Hybrid E Dual LPG LPG Hybrid LPG In order to explore the elasticities further, it is worthwhile to assume different plausible distributions of fuel use likely to occur in the future. This is discussed in the next section. 3 FROM CHOICE TO DEMAND ELASTICITIES In general, choice elasticities cannot be interpreted as demand elasticities. Kremers et al s (2002) conclusion that there is little difference between choice and demand elasticities generally holds for work trips but that does not alter the fact that choice is only part of demand. A particular demand elasticity h ij may be similar to the corresponding choice elasticity m ij while in other cases they will differ substantially. In addition to choice, a complete representation of demand must include the income effect. Smith (2006) showed that: h ij = m ij + w j [ fh Fm - w F h Fm ( 1+ fh Fm )] (1) where w F is fuel group share of household expenditure, w j is individual share of group budget, h Fm is fuel group income elasticity and f the Frisch income flexibility. In the fuel case budget shares are treated as approximately equivalent to effective physical shares because the high performance of diesel and low energy density of LPG are approximately balanced by their relatively high and low litre costs in Australia. This means that there is no adjustment for different expenditure and quantity shares. Clements (2008) uses the values, f = -0.5 and h Fm = 1.0. However with these values too many negative cross-elasticities were obtained and it became cumbersome to interpret. Different values for f and h Fm were tested and the final ones adopted were -0.4 and Endogenous Weighting NLogit4 allows the analyst to weight the data so that it conforms to some prior view of the market. Endogenous weighting is used when the true market shares of alternatives are applied to the model (Hensher et al., 2005). In this study, SP data is used and therefore the 3

4 true market shares for all alternatives are not known. Nevertheless one can make reasonable assumptions about the future market for car/fuel types and use this method to obtain elasticity estimates. Two scenarios are presented. The first is that of a fuel type distribution representing a near future scenario and the other representing the longer term. Table 2 shows the distribution of choices from the data set as well as the two new sets of distributions assumed for endogenous weighting. The 2006 passenger vehicle survey (ABS, 2006) in litres of fuel consumed gives: 91.4% petrol, 4.12% diesel, 4.47% LPG (and other). The distributions used in the two scenarios have been developed from these relative shares. Table 2: Data choice shares and projected household choice shares Choice shares from data set (%) Near future distribution (%) Long term distribution (%) Conventional Petrol Diesel Hybrid Petrol E B Hybrid E Dual LPG LPG Hybrid LPG Petrol share is obtained by difference once all the other shares are established. It is assumed that petrol will still be the dominant fuel such that petrol and hybrid petrol represent 63% as shown in the long term distribution. It could well be that Hybrid Petrol will be more than 15% as cost of fuel continues to rise and hybrid cars become more competitive with their conventional counterparts. In other countries diesel is a favoured fuel on energy efficiency grounds whereas Australia has a long established policy of pricing it out of the car user's reach mainly due to cost recovery from trucks for road wear. Although diesel is expensive worldwide, this might not last. For this reason, diesel is believed to become a more important fuel in the longer term. E20 contains 80% petrol so that its share is expected to remain small in the longer term. The other assumed shares may be moderately realistic although there is much uncertainty about the development of alternative fuels. 3.2 Two Scenarios: Near Future and Long Term For the two scenarios, the data is weighted with the respective shares and the choice elasticities are generated by the NLogit4 software. As with the choice elasticities discussed earlier, the columns of the original estimates are combined such that each column represents a fuel. The 9 5 elasticity matrix is then normalised with the help of genetic algorithm (GA) software so that the two properties, symmetry and zero share weighted column sums of the elasticity matrix are restored (Taplin et al, 1999). The symmetry calculations take account of the multiple uses of petrol, E20 and LPG. The zero share weighted column sum simply implies that someone switching from j must adopt one of the other alternatives. The resulting 4

5 choice elasticities for the near future case are shown in Table 3 and the long term projection in Table 4. Table 3: Estimated choice elasticities for the near future Petrol Diesel E20 B20 LPG Share % Conventional Petrol Diesel Hybrid Petrol E B Hybrid E Dual LPG LPG Hybrid LPG Share Wtd Column Sum Table 4: Estimated choice elasticities for the long term Petrol Diesel E20 B20 LPG Share % Conventional Petrol Diesel Hybrid Petrol E B Hybrid E Dual LPG LPG Hybrid LPG Share Wtd Column Sum From these modified choice elasticities the respective demand elasticities are then derived. The Australian household allocation of 3.615% of expenditure to fuel is used (ABS, 2004). Table 5 shows the specific income effects for the two scenarios for each car/fuel type based on the income term from Equation (1): w j [ fh Fm - w F h Fm ( 1+ fh Fm )]. Table 5: Income effect Fuel Near Future scenario Long Term scenario Conventional Petrol+Hybrid Petrol Diesel E20+Hybrid E B LPG+Dual LPG+Hybrid LPG

6 3.3 Near Future Demand Elasticities The ordinary elasticities of demand were derived from the choice elasticities by adding to each element in a column of choice elasticities its respective income effect from Table 5. Table 6 shows the derived ordinary demand elasticities with respect to fuel price under the near future scenario. All the own and cross elasticities have the expected signs. Hybrid petrol is slightly less elastic compared to conventional petrol vehicles. This is consistent with hybrids being more efficient. The underlined cross-elasticities are actually constrained to the minimum value Table 6: Near future car/fuel type demand elasticities with respect to fuel prices Petrol Diesel E20 B20 LPG Conventional Petrol Diesel Hybrid Petrol E B Hybrid E Dual LPG LPG Hybrid LPG The results indicate that the demand for diesel cars is more elastic compared to petrol or hybrid petrol alternatives. Although diesel cars have higher performance than conventional petrol cars, the litre cost of diesel remains higher than petrol and diesel cars tend to be those with big engines. The relatively high diesel elasticity may suggest a shift towards smaller engines since moving from diesel to any other fuel while keeping a big engine vehicle would not be any cheaper to run due to loss of performance. The estimated demands for E20 cars, hybrid E20 and B20 are relatively elastic. This reflects their high substitutability. The sensitivity to the price of these fuels could be due to their newness in the market. In the case of the LPG options, demand for Hybrid LPG is fairly inelastic compared to Dual LPG and LPG. Once again, this is consistent with the fact that hybrids are more efficient technologies compared to the conventional counterparts. 3.4 Long Term Demand Elasticities The estimated demand elasticities under the long term scenario are shown in Table 7. Table 7: Long term car/fuel type demand elasticities with respect to fuel prices Petrol Diesel E20 B20 LPG Conventional Petrol Diesel Hybrid Petrol E B Hybrid E Dual LPG LPG Hybrid LPG

7 The comparison of immediate future and long term demand elasticities does not reveal any dramatic changes. From the immediate future to the long term demand elasticities, conventional and hybrid petrol become slightly more elastic with respect to petrol price. The larger petrol share in the immediate future scenario, i.e. 71% Petrol + 8% Hybrid Petrol, enlarges the cross-elasticities in the petrol column making the absolute own-elasticities smaller. The following example illustrates this point: È h LPG.Pet Pr ice = h Petrol.LPG Pr ice Í Î ( 71% + 8% ) (3% + 4% + 1%) This has made h LPG. Pet Pr ice bigger in this case (0.329) than in the Long Term scenario case (0.235) where the total petrol share is 48%+15% as well as the LPG share being larger. Whereas one might expect a decrease in the demand elasticity for diesel cars, due to an increase in its share from 6% to 10%, demand for diesel cars actually becomes a little more elastic in the long term scenario. This is mainly due to the greater income effect. Hybrid E20, B20 and LPG become slightly less elastic compared to the immediate future elasticities which is partly due to their bigger shares under the long term scenario. 3.5 A Restricted Case: Three Fuels Only A final estimation was carried out for the current three conventional fuels: petrol, diesel and LPG. In this case, the endogenous weighting corresponds to the 2006 shares modified to take account of recent trends. The specific income effects for each fuel are based on the market shares and the income term, w j [ fh Fm - w F h Fm ( 1+ fh Fm )], but here f = -0.5 and h Fm = 1.0, as in Clements (2008). The resulting income effects are: Petrol Diesel LPG The demand elasticities are derived from the choice elasticities by adding to each element in a column on the left hand side of Table 8 its respective income effect to obtain the demand elasticity in the corresponding column on the right hand side of Table 8. Table 8 Estimated choice and demand elasticities Choice Elasticities Demand Elasticities Petrol Diesel LPG Share % Petrol Diesel LPG Petrol Diesel LPG Share Wtd Col Sum This result indicates that demand for petrol, the fuel with the major market share, is much more elastic than the own-choice elasticity for petrol. In contrast, the own-choice and owndemand elasticities differ only a little for each of the two minor fuels. 7

8 4 CONCLUSIONS As the distribution of fuel consumption changes from one scenario to another, the change in demand elasticities will depend on two factors: the change in the consumption share and the income effect applied to the elasticity. While some choice elasticities are similar to demand elasticities (Kremers et al., 2002), others differ radically. Generally when the shares are relatively small, the own-choice and own-price elasticities are similar. This is due to the fact that when the share is multiplied by group elasticity, the result is very little difference between choice and demand elasticities. In the case of petrol, because it is the dominant product in the group, its own-price elasticity differs greatly from the choice elasticity. REFERENCES Abdoolakhan, Z, 2006, Choice factors in the adoption of alternative fuels and new technology vehicles: results from a pilot survey, PATREC Research Forum ABS, 2004, Household Expenditure Survey, Australia: Detailed Expenditure Items, , Cat. No ABS, 2006, Survey of Motor Vehicle Use, Australia 12 months ended 31 October Clements, K, 2008, Price elasticities of demand are minus one-half, Economics Letters 99, Hensher, D, Rose, J and W Greene, 2005, Applied Choice Analysis (Cambridge UP) Koppelman, FS and Bhat C, 2006, A self-instructing course in mode choice modelling: multinomial and nested logit models. U.S Department of Transportation, Federal Transit Administration. Kremers, H, Nijkamp, P and P Rietveld, 2002, A meta-analysis of price elasticities of transport demand in a general equilibrium framework, Economic Modelling, 19, Smith, Brett, 2006, A method to infer theoretically consistent demand elasticities from mode choice models, 11th International Conference on Travel Behaviour Research, Kyoto. Taplin, J., Hensher, D. and Brett Smith, 1999, Preserving the symmetry of estimated commuter travel elasticities, Transportation Research Part B 33,

9 APPENDIX: Example of a choice scenario from the online survey 9