Qualitative Effects of Cash-For-Clunkers Programs

Transcription

1 Intro. Data Treatments Logit Substitution Summary Qualitative Effects of Programs Eugenio J. Miravete 1 María J. Moral 2 1 University of Texas at Austin & Centre for Economic Policy Research 2 UNED March 28, 2012

2 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Disclaimer My apologies but this paper is not about the 2009 U.S. Cash-for-Clunkers Program, although perhaps we can all learn something about its potential effects.

3 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Scrappage Programs Cash-for-Clunkers is a common (Keynesian) policy in time of recession. In addition to direct and indirect jobs, it helps reduce the average age of the car fleet and improve road safety. In the absence of a carbon tax or higher fuel taxes, these programs help achieving environmental goals as long as newer cars are more fuel efficient than the replaced ones.

4 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Adoption of TDI technology The paper introduces the massive adoption of the new TDI technology to (hopefully) a wider audience. Demand for automobiles change dramatically in Europe during the 1990s. Evaluating the effects of the adoption of TDIs is complicated because the automobile industry is not perfectly competitive, vehicles are horizontally differentiated, and diffusion takes an extended period of time. For the current version the key question is: Do clunkers programs induce any permanent qualitative effect beyond a temporary sale spike?

5 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Quantity vs. Quality Sales: Effect of scrappage rates and impact on optimal replacement age. Greenspan and Cohen (1999); Adda and Cooper (2000). Emissions: Combined with compulsory maintenance and competition effects. Hahn (1995); Goldberg (1998). Price Discrimination: Pricing of vehicles with different fuel efficiency responds to mileage driven. Verboven (2002).

6 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Heterogeneous Demand for Diesel Vehicles in Europe EU15 Diesel-(%)

7 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Heterogeneous Demand for Diesel Vehicles in Europe SPAIN Spain: % Diesel EU15: % Diesel

8 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Heterogeneous Demand for Diesel Vehicles in Europe FRANCE: Automobile Sales & Scrapping Program Gasoline Diesel

9 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Heterogeneous Demand for Diesel Vehicles in Europe AUSTRIA: Automobile Sales & Scrapping Program Gasoline Diesel

10 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Heterogeneous Demand for Diesel Vehicles in Europe GERMANY: Automobile Sales & Scrapping Program Gasoline Diesel

11 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Heterogeneous Demand for Diesel Vehicles in Europe SPAIN: Automobile Sales & Scrapping Program Gasoline Diesel

12 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview What do we want to achieve? Determine the main drivers of demand for vehicles and how they change over time. Evaluate whether scrappage programs accelerated the adoption of a new and relatively unknown technology. Identify why this policy intervention was successful in shifting demand by comparing it with other experiences. Describe a process of preference change towards more fuel efficient vehicles that transformed the shape of the Spanish automobile industry within a decade.

13 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview Main Results The effectiveness of Cash-for-Clunkers in changing the nature of demand for automobiles is not immediate but rather it has to be evaluated in the long run. Better mileage is not the only reason why drivers favor diesel over gasoline automobiles. Preferences evolve so that diesel and gasoline models become closer substitutes but their demand respond very differently to each other prices.

14 Intro. Data Treatments Logit Substitution Summary Motivation Literature Agenda Preview What about CARS? The bottom line: It will not succeed alone in shifting the demand towards more fuel efficient automobiles because the policy intervention happens too early in the diffusion path of fuel efficient hybrids.

15 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data What is TDI? Turbo Direct Injection Diesel Engine (Audi-WV, 1989): The most efficient internal combustion engine available. It combines a direct injection pump, a turbo charger, and an intercooler to increase the power while reducing fuel consumption and therefore emissions per gallon. Increased performance and higher torque than gasoline engines. High durability and reliability, zero smell, and quite low clattering.

16 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data The Spanish Economy in the 1990s Strong recession in years Unemployment reached 24% in Very strong growth in the second half of 1990s. Immigration (first wave since 1492) exceeded 10% of population of nationals in just five years.

17 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data The Spanish Automobile Industry in the 1990s Fifth largest in the world by output. Largest exporting industry in the country. Second largest contributor to GDP. Important links to other sectors such as automobile components. The component industry is so diversified that it attracted Japanese manufacturers to open manufacturing plants in Spain after the automobile market was liberalized.

18 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data Program Descriptions Policy intervention was unanticipated (following France s example) and did not target diesel vehicles specifically. First Plan Renove : April 94 to June 94. Between 500 to 600 euros (a 4.89% price discount for gasoline and a 4.40% price discount for diesel models, respectively). Need to buy a new car and replace one seven years or older. Second Plan Renove : October 94 to March 95. About 480 euros. Need to buy a new car and replace one ten years or older.

19 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data Data Description Spanish automobile registrations, : We observe registrations by type of engine. Unbalanced panel with 1,869 model-year observations. 340 models (206 gasoline and 134 diesel). Relative to gasoline models diesel vehicles are... 10% heavier. 15% to 20% less powerful. Between 1,000 and 2,000 euros more expensive. Consume 20% less fuel per mile leading to an average 35% savings in the cost of driving.

20 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data Overall, diesel vehicles aremiravete about 10% & Moral heavier than similar gasoline versions. Diesel and gasoline Table 1: Spanish Automobile Market by Segment and Type of Fuel: gasoline: sales models price fcost km/l90 hp length width weight all 646, , ,227 small 282, , ,900 compact 204, , ,305 sedan 117, , ,621 luxury 35, , ,970 minivan 6, , ,071 diesel: sales models price fcost km/l90 hp length width weight all 371, , ,442 small 91, , ,023 compact 144, , ,409 sedan 116, , ,686 luxury 10, , ,282 minivan 9, , ,414 sales indicate the average yearly vehicle registration between 1991 and models reports the average number of models available for consumers to choose from each year. All other variables are sales weighted averages for the same period. price is the price of automobiles measured in the equivalent of 1994 euros; fcost is the fuel cost in 1994 euros of driving 100 kilometers on a highway; km/l90 reports the kilometers traveled with one liter of fuel driving at a speed of 90 kilometers per hour on a highway; hp denotes the horsepower; length and width are measured in inches and weight in pounds. Table 1 summarizes the features of vehicles sold during the 1990s by segment and fuel type.

21 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data Number of Models: ALL Gasoline Diesel Even before the scrappage programs, there were numerous diesel models available, particularly in the compact and sedan segments For the sake of comparison, today there are 15 hybrid models available in the US but mostly sold in California and with the Prius amounting to over 50% of a 3% of the total automobile market (400,000 units).

22 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data Automobile Sales: ALL Gasoline Diesel Diesel sales grew over 600% from 12% market share in 1991 to 54% in 2000 (and exceeding over 70% in the sedan and minivan segments. Sales of gasoline vehicles declined approximately 17% over the decade.

23 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data Automobile Sales & PC-Income Gasoline Diesel PC-Income 8000 Demand for gasoline vehicles does not increase with income. Diesel vehicles are not inferior goods.

24 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data Automobile Sales and Relative Diesel/Gasoline Price Gasoline Diesel Diesel/Gasoline Price Demand for diesel vehicles takes off when diesel fuel price is cheapest relative to gasoline. Once demand for diesel vehicles shifts, it continued growing despite a steady increase in the relative price of diesel fuel.

25 Intro. Data Treatments Logit Substitution Summary TDI Scrappage Programs Data Figure 3: Diesel vs. Gasoline Sales Automobile Sales: Compact Automobile Sales: Sedan Gasoline Diesel Gasoline Diesel Automobile Sales: Luxury Automobile Sales: Minivan Gasoline Diesel Gasoline Diesel an impressive 48% increase. It could be argued that after this big immediate increase in the sales

26 Intro. Data Treatments Logit Substitution Summary DID Average Treatments Were the effects of scrappage programs on shifting demand toward diesel vehicles immediate? Gasoline: 86,706 units more in 1994 than in 1993 (574,896). Diesel: 78,497 units more in 1994 than in 1993 (163,140). Claim: Shift of preferences is not intended by these programs. DID: The average treatment specification is: ln(sales/1000) = α 0 + α ssegment + α mmanufacturer + α eeconomic variables + α d diesel + α td α tttime trend + α dt diesel d α dtt diesel time trend.

27 Intro. Data Treatments Logit Substitution Summary DID Table 4: Scrappage Programs Average and Dynamic Treatment Effects 19 all small compact sedan luxury minivan constant (4.18) (11.65) (2.00) (7.56) (3.30) (1.49) compact (15.77) sedan (1.57) luxury (9.77) minivan (20.85) diesel (10.20) (9.68) (7.51) (7.89) (3.82) (1.71) D (5.34) (2.77) (4.61) (2.86) (2.37) (0.73) diesel D (11.29) (5.37) (9.29) (12.05) (3.59) (4.64) price (10.83) (22.02) (7.03) (3.19) (7.46) (7.92) gdp-pc (8.68) (4.50) (5.24) (8.81) (3.84) (3.25) fuelprice ((3.35) (4.22) (2.42) (3.95) (1.74) (2.85) R all small compact sedan luxury minivan Sales per model of diesel versions is smaller than for gasoline versions constant (1.26) (1.11) (1.36) (1.80) (0.04) (1.30) compact (16.59) sedan (1.69) luxury (α d < ). This (9.15) is particularly true compact and sedan vehicles. minivan (22.47) diesel (6.70) (7.23) (5.60) (8.59) (2.71) (7.35) d94 The effect of (0.83) scrappage (1.74) programs tends (0.38) to compensate (0.04) this(0.21) effect in the (0.70) d (1.33) (0.50) (2.00) (2.10) (0.07) (1.91) d96 long run (α dt (1.23) > 0) (0.66) (1.65) (1.90) (0.16) (1.78) d (1.24) (0.73) (1.53) (1.98) (0.24) (1.50) d (1.36) (0.96) (1.51) (1.83) (0.12) (1.50) d99 The scrappage (1.25) programs (0.79) have no significant (1.47) effect (1.84) on the demand (0.21) for(1.51) d (1.11) (0.61) (1.42) (1.72) (0.27) (1.53) diesel d (4.43) (1.02) (4.48) (2.89) (1.73) (0.89) diesel d95 gasoline models (6.92) (α t = 0) (1.69) but increases (7.67) long term(4.07) demand (2.23) for diesel (2.48) diesel d (9.19) (2.50) (7.60) (6.64) (3.32) (2.77) diesel d97 vehicles by an (8.93) average of (3.95) 28% (7.24) (8.75) (2.90) (4.18) diesel d (12.26) (5.46) (7.14) (10.67) (2.34) (3.06) diesel d (12.02) (5.13) (7.66) (11.40) (2.02) (3.07) diesel d00 The effect of (13.03) other exogenous (6.36) shifts (8.63) in demand (8.81) (α dtt > ) (3.28) amounts to(3.28) a price (8.95) (17.58) (5.96) (2.77) (7.73) (9.16) gdp-pc 21% increase (1.63) in demand (1.25) for diesel vehicles (1.52) in the (2.14) long run. (0.30) (1.12) fuelprice (0.99) (1.90) (0.21) (0.27) (1.21) (0.90) R Observations 1, Models Firm Dummies OLS estimates. Endogenous regressors are Miravete sales per model & and Moral year in thousand units. Absolute, heteroskedastic-consistent t-statistics are reported

28 Intro. Data Treatments Logit Substitution Summary DID Dynamic Treatments The effect of scrappage programs on changing the preferences of consumers needs not be immediate. This is particularly true if consumers are uncertain about the quality of the new technology and use the share of sales or accumulated sales as a signal to update their prior about the value of this new product. Demand for gasoline vehicles does not increase neither in the short or long run. Dynamic effects on diesel are particularly important for luxury and minivan segments while exogenous demand shifts drive the demand for compact and sedan vehicles.

29 compact (15.77) Intro. sedan Data Treatments Logit (1.57) Substitution Summary DID luxury (9.77) minivan (20.85) diesel (10.20) (9.68) (7.51) (7.89) (3.82) (1.71) D (5.34) (2.77) (4.61) (2.86) (2.37) (0.73) diesel D (11.29) (5.37) (9.29) (12.05) (3.59) (4.64) price (10.83) (22.02) (7.03) (3.19) (7.46) (7.92) gdp-pc (8.68) (4.50) (5.24) (8.81) (3.84) (3.25) fuelprice ((3.35) (4.22) (2.42) (3.95) (1.74) (2.85) R all small compact sedan luxury minivan constant (1.26) (1.11) (1.36) (1.80) (0.04) (1.30) compact (16.59) sedan (1.69) luxury (9.15) minivan (22.47) diesel (6.70) (7.23) (5.60) (8.59) (2.71) (7.35) d (0.83) (1.74) (0.38) (0.04) (0.21) (0.70) d (1.33) (0.50) (2.00) (2.10) (0.07) (1.91) d (1.23) (0.66) (1.65) (1.90) (0.16) (1.78) d (1.24) (0.73) (1.53) (1.98) (0.24) (1.50) d (1.36) (0.96) (1.51) (1.83) (0.12) (1.50) d (1.25) (0.79) (1.47) (1.84) (0.21) (1.51) d (1.11) (0.61) (1.42) (1.72) (0.27) (1.53) diesel d (4.43) (1.02) (4.48) (2.89) (1.73) (0.89) diesel d (6.92) (1.69) (7.67) (4.07) (2.23) (2.48) diesel d (9.19) (2.50) (7.60) (6.64) (3.32) (2.77) diesel d (8.93) (3.95) (7.24) (8.75) (2.90) (4.18) diesel d (12.26) (5.46) (7.14) (10.67) (2.34) (3.06) diesel d (12.02) (5.13) (7.66) (11.40) (2.02) (3.07) diesel d (13.03) (6.36) (8.63) (8.81) (3.28) (3.28) price (8.95) (17.58) (5.96) (2.77) (7.73) (9.16) gdp-pc (1.63) (1.25) (1.52) (2.14) (0.30) (1.12) fuelprice (0.99) (1.90) (0.21) (0.27) (1.21) (0.90) R Observations 1, Models Firm Dummies OLS estimates. Endogenous regressors are sales per model and year in thousand units. Absolute, heteroskedastic-consistent t-statistics are reported in parentheses. All regressions also include manufacturer fixed effects although these estimates are not reported. Variables price, fuelprice, and gdp-pc are measured in the equivalent of thousands of 1994 euros. Segment dummies add up to zero in the estimation that includes all segments. Thus, the estimate for the small dummy is for the average treatment specification and for the dynamic treatment model.

30 Intro. Data Treatments Logit Substitution Summary DID Robustness This is not the ideal data to conduct DID analysis. Collected data on market shares of diesel vehicles and scrappage programs across Europe. Evaluated the effect of these programs on the speed of diffusion: «λit ln = β 0 + δ eeconomic variables 1 λ it + δ ttime trend + δ sscrappage.

31 Intro. Data Treatments Logit Substitution Summary DID term effects on the composition Miravete of automobile & Moral fleets we documented for the Spanish case. Table 3: Scrappage Programs and Adoption of Diesel Technology in Europe Mean Std. Dev. OLS Regressions constant [12.19] [12.10] trend (5.48) [5.63] [5.52] treatment (0.45) [1.89] scrappage (0.20) [1.52] after scrappage (0.42) [1.61] gdp-pc (8.63) [4.41] [4.41] fuelprice: gasoline - diesel (0.11) [3.90] [3.87] fueltax: gasoline - diesel (4.23) [ 0.50] [0.49] R Regressions make use of 277 country-year observations between 1991 and The endogenous variable is the logistic tranformation of the market share of diesel vehicles sold each year in each country. The average diesel penetration is 29% (with a standard deviation of 22%). Fuel prices and gdp-pc are measured in thousand U.S. dollars using PPP exchange rates while fuel taxes are measured as percentage over fuel prices. treatment is a dummy variable that takes value 1 after the application of scrappage program in those countries that sponsored them. Alternatively we define the scrappage dummy for periods when scrappage programs were implemented and after scrappage for those that followed the policy intervention. Absolute-value, heteroskedastic-consistent, t-statistics are reported between brackets. effects of scrappage programs. We thus conclude that, although not definitive, the evidence does not contradict our suggested interpretation that scrappage programs may induce qualitative long

32 Intro. Data Treatments Logit Substitution Summary Diesel vs. Gasoline Equilibrium Analysis Nested Logit Demand Estimation Do consumers buy diesel only because of fuel savings? One-nest logit model: Danger of misspecification. We claim that there are unobservable vehicle features that are highly correlated with the type of engine. It allows us to address endogeneity of vehicle pricing. Demand specification: ln(s j ) ln(s 0 ) = x j β αp j + σ ln(s j g ) + ξ j,

33 Intro. Data Treatments Logit Substitution Summary Diesel vs. Gasoline Equilibrium Analysis Table 4: Multinomial and Nested Logit Demand Estimation Model I Model II Model III constant (6.61) (14.26) (20.88) hp/weight (1.11) (1.57) (2.03) km/l (3.03) (2.22) (3.01) length width (0.13) (0.18) (2.12) compact (0.36) (1.10) (0.84) sedan (1.13) (0.93) (1.37) luxury (1.73) (0.23) (0.54) minivan (0.84) (1.55) (1.44) fuelprice (0.79) (12.17) (13.51) gdp-pc (3.40) (15.87) (16.24) diesel (3.57) (54.10) (56.01) d (1.15) (5.49) (5.57) d (2.07) (8.28) (8.09) diesel d (1.00) (22.28) (36.86) diesel d (2.07) (30.06) (52.96) trend (1.37) (0.18) (0.15) price (0.66) (1.24) (2.46) ln(sj g) (46.50) (62.33) ser sargan [df; χ (df)] [8; 20.09] [7; 18.48] [7; 18.48] No. inelastic demands 1, 835 1, (± 2 s.e. s) (1, 133 1, 869) (592 1, 869) (0 292) LV : I vs. II, III LV : II vs. III IV estimates. Absolute, heteroskedastic-consistent t-statistics are reported in parentheses. price is measured in thousands of 1994 euros. length and width are measured in inches/100; hp/weight is measured as horsepower per 10 pounds of weight. The number of observations is 1,869 and the number Miravete of automobile & Moral models is 340. ser indicates the standard error of

34 Intro. Data Treatments Logit Substitution Summary Diesel vs. Gasoline Equilibrium Analysis Models I III I Logit structure is not flexible enough to address demand even after controlling for the endogeneity of vehicle price. II Price cannot be considered exogenous in a model where individuals first choose the engine type and then the vehicle. III Instrumenting for price leads to a much better fit of the model if we consider that individuals first choose the engine type and then the vehicle. Similar evidence: consumers prefer less expensive, larger, more powerful, and more fuel efficient automobiles. Income affects demand positively and although diesel vehicles are less valued, willingness to pay increases after the scrapping programs were implemented.

35 Intro. Data Treatments Logit Substitution Summary Diesel vs. Gasoline Equilibrium Analysis Oligopolistic Supply The effect of scrapping programs may be overestimated if we ignore the effect that a change of preferences will have on the optimal markups charged by automobile manufacturers in an model of oligopolistic competition with differentiated goods. Equilibrium Model. Marginal cost: Profit Maximization Condition: ln(mc j ) = z j γ + ω j. Pricing Equation: p = mc + 1 (p, x, ξ, α, γ, β)s(p, x, ξ, α, γ, β), ln [ p 1 (p, x, ξ, α, γ, β)s(p, x, ξ, α, γ, β) ] = z j γ + ω j.

36 Intro. Data Treatments Logit Substitution Summary Diesel vs. Gasoline Equilibrium Analysis Table 5: BLP: Equilibrium Model Estimates Means Demand-side Parameters (β s): constant (16.72) hp/weight (6.17) km/l (0.86) (4.55) length width (3.35) diesel (2.22) (3.18) d (1.01) d (4.99) diesel d (0.05) diesel d (3.05) trend (6.83) Price Effect (α i): price (8.54) Cost-side Parameters (γ s): constant (1.13) ln(hp) (4.76) ln(weight) (1.19) ln(q EU) (2.92) diesel (2.15) trend (0.79) non-european (2.81) gmm estimates. Absolute, consistent t-statistics are reported in parentheses. price is measured in thousands of 1994 euros. All other variables are defined in Table 1 and Table 4. The number of observations is 1,869 and the number of automobile models is 340. σ s fuel efficiency when buying an automobile (non-significant mean value of the estimate of km/l90),

37 Intro. Data Treatments Logit Substitution Summary Diesel vs. Gasoline Equilibrium Analysis GMM Estimates. Same demand effects. Consumers appear not to be sensitive to fuel efficiency (although there is individual heterogeneity). Consumers are also heterogeneous regarding their perception of the quality of diesel vehicles, although most think of gasoline as being superior. Diesel becomes however the preferred option after scrappage programs were implemented. There are economies of scale in production and non-european manufacturers enjoy lower costs. Producing diesel and more powerful vehicles is more expensive.

38 Intro. Data Treatments Logit Substitution Summary Figure 4: Elasticities of Substitution by Segments and Fuel Type Elasticity of Substitution: Gasoline Models to Diesel Model Prices Elasticity of Substitution: Diesel Models to Gasoline Model Prices SMALL COMPACT SEDAN LUXURY MINIVAN SMALL COMPACT SEDAN LUXURY MINIVAN Diesel vehicles become much more responsive to prices over time. This asymmetric pattern of cross-price substitution is consistent with the idea of consumer learning of the quality of the new technology. The introduction of diesel models does not affect the cross-price This asymmetric pattern of cross-price substitution is indicative of consumer learning. responsiveness of those consumers who still prefer gasoline vehicles. This group may include customers that are uninformed about the new tdi technology or those high income customers or infrequent drives for which potential savings of buying a diesel are nil. On the contrary, those price sensitive, intensive drivers, and informed customers do substitute away from gasoline vehicles if they become more expensive. Moreover, this response is larger as time goesmiravete by and thus & Moral sales of diesel vehicles likely follows replacement of old gasoline

39 Intro. Data Treatments Logit Substitution Summary Summary The decade of the 1990s in Europe (and Spain in particular) is an interesting application to study the diffusion of a new technology in an environment of oligopolistic competition with horizontally differentiated goods: Scrapping programs have a differentiated effect on the new and old durable good. Long term effects on the demand for diesel vehicles exceeds the immediate impact of scrapping programs. Firms adjust their markups to changing preferences of consumers, thus offering a countervailing incentive to the fast adoption of more fuel efficient and less pollutant technologies. Future extension: Dynamic pricing issues. Learning. Dynamic oligopolstic equilibrium estimation.