The Incidence of Cash for Clunkers
Evidence from the 2009 Car Scrappage Scheme in
GermanyI
Ashok Kaul, Gregor Pfeifer, Stefan Witte
Working Paper
This version: October 2013
First version: April 2012,II
Abstract
Governments all over the world have invested tens of billions of dollars in car scrappage
programs to fuel their economies in 2009. We investigate the German case using a unique
micro transaction dataset covering the years 2007-2010. Our focus is on the incidence
of the premium, i.e., we ask how much of the e 2,500 buyer subsidy is actually captured
by the buyer. A simple heuristic model suggests that the incidence will depend on the
market segment. For cheaper cars, the supply-side is likely to capture some small part
of it while it will offer additional discounts for more expensive cars. Using regression
analysis, we find these hypotheses confirmed. Subsidized buyers of cheap cars paid more
than comparable buyers who did not receive the subsidy, e.g., for cars costing e 12,000, car
dealers reaped about 7% of the scrappage premium, leaving 93% with the buyer. For more
expensive vehicles (cars costing e 32,000), subsidized buyers were granted extra discounts
of about e 1,100 on top of the government premium they received. The results are robust
to extensive sensitivity checks.
I
We are thankful for valuable remarks from Martin Becker, Nadja Dwenger, Marc Escrihuela Villar, Rainer Haselmann, Stefan Kloessner, Dieter Schmidtchen, and Michael
Wolf, as well as the participants of the ACDD conference 2012 in Strasbourg, the Econometric Society Australasian Meeting 2012 in Melbourne, the Warsaw International Economic Meeting 2012, the annual meeting of the German Economic Association 2012 in
Goettingen, and the IIPF conference 2012 in Dresden. Corresponding author is Ashok
Kaul. E-mail address: ashok.kaul@econ.uzh.ch; Tel.: +41 (0)44 634 37 36; Fax: +41
(0)44 634 49 07.
II
University of Zurich, Department of Economics Working Paper No. 68 (http://www.
econ.uzh.ch/static/workingpapers.php?id=745).
1. Introduction
As a reaction to the 2007 financial crisis, governments all over the world
launched car scrappage programs to stimulate the economy in 2009. While
the U.S. spent $3 billion on their “Cash-for-Clunkers” agenda, Germany afforded the most expensive program of all countries with a total volume of
about $7 billion (e 5 billion), a third of the worldwide budget spent on scrappage schemes in this period. Before 2009, similar programs have previously
been implemented, particularly in the 1990s. Since such interventions are
popular amongst policy makers and consumers, we expect similar programs
to be adopted in the future.
In the present contribution, we ask the question how much of the e 5
billion was actually captured by which market side, i.e., we analyze the incidence of the German scrappage program.1 To the best of our knowledge,
this is the first analysis trying to evaluate the incidence of a scrappage rebate. While the subsidy was meant to benefit the consumer, economic theory
suggests that the economic incidence of a subsidy is independent of the statutory incidence.2 Instead, the division of the beneficial amount between buyers
and sellers depends on the relative elasticities of demand and supply. The
1
The paper is closely related to the empirical literature on tax incidence (for the fundamentals and an extensive literature review, see Kotlikoff and Summers (1987) and Fullerton
and Metcalf (2002), since a subsidy is essentially a negative tax.
2
This so-called tax equivalence theorem is a basic fundamental within the incidence context. Ruffle (2005) for instance, shows that this theorem empirically holds. However,
other research (e.g., Busse et al. (2006), Chetty et al. (2009), and Sallee (2011)) implies
that, contrary to standard theories of incidence, the statutory incidence of a policy does
affect the economic incidence.
2
German scrappage program, called Abwrackprämie (scrappage premium) or
Umweltprämie (environmental premium), started in late January 2009. To
receive the lump-sum subsidy of e 2,500 (about $3,500), buyers had to prove
scrappage of an old car and registration of a new one. By September 2009,
the budget was exhausted, having subsidized the purchase of 2 million new
cars. Car dealers in general managed the scrapping of the old car and dealt
with the responsible federal agency and, hence, could identify two different
groups of customers, buyers receiving the subsidy or not. That is why, in our
model framework, we argue that we expect our incidence results to be in line
with an optimal long-run pricing strategy of the supply side reflecting different price elasticities of demand and market conditions in different car price
segments. We therefore expect the effect to be heterogenous over car prices.
To be more precise, we assume that for cheaper cars, the bulk or even all
of the subsidy amount remains with the buyer, implying incidence amounts
of slightly below or at around 100%. For subsidized buyers of large cars we
assume extra discounts on top of the scrappage subsidy amount, implying
incidence amounts of more than 100%.
In the empirical analysis, we use a unique sample of transaction data for
Germany from the years 2007 to 2010. Our focus is on the discount received
by subsidized buyers in comparison to non-subsidized buyers controlling for
covariates. We apply linear regression methods to model the percentage discount from the manufacturer’s suggested retail price (MSRP) as a function
of the scrappage dummy. In a first step, we find that the average effect of
3
the premium on discount was slightly positive, implying that customers captured more than the total amount of the subsidy. Augmenting that model
and allowing for heterogeneity across price segments when comparing subsidized to non-subsidized purchases, we find that these differ significantly.
Subsidized buyers of the first quartile (cheap cars) received less discount
than non-subsidized buyers, implicating a demand-associated incidence of
less than 100%. Somewhere in the second quartile, the difference was zero,
implying just no pass-through of the subsidy to the dealers at all or, put
differently, an incidence of exactly 100%. Above the median MSRP, the discount for subsidized buyers was higher than the discount for non-subsidized
ones, translating into incidence amounts of more than 100%. Consequently,
the empirical results confirm our model assumptions.
Previous work on incidence focused mostly, but not only, on taxes, e.g.,
in Evans et al. (1999), Chetty et al. (2009), Friedman (2009), Hastings and
Washington (2010), Rothstein (2010), and Marion and Muehlegger (2011).
Within the scrappage context however, most papers analyze either sales
(quantities) or environmental aspects—and ignore the incidence of the subsidy, i.e., the price dimension.3 To the best of our knowledge, there exists
only one piece that—amongst others—tries to combine scrappage scheme and
3
For instance, see Adda and Cooper (2000), Licandro and Sampayo (2006), Li et al.
(2013), and Mian and Sufi (2012) for sales effects, and Hahn (1995), Deysher and Pickrell
(1997), Kavalec and Setiawan (1997), Szwarcfiter et al. (2005), and Knittel (2009) for
environmental impacts. This literature mostly finds that the increases in sales during the
program are offset, sometimes completely, by a decrease in later sales as well as the fact
that from an environmental perspective, these programs did not pay off.
4
pass-through questions. Using the car price as the dependent variable, Busse
et al. (2012) estimate whether the U.S. programs rebates did pass through
fully to buyers, without going into a thorough incidence or price discrimination analysis. Instead, they further evaluate whether the rebate crowded
out or stimulated manufacturer incentives, and whether the scrapping of a
large number of vehicles affected prices in the used-vehicle market.4 There
is also some important research regarding incidence within the automobile
market, albeit irrespective of the scrappage context. Busse et al. (2006) analyze cash incentives directed at either the dealer or the customer. They
show that customer rebates are passed to the buyer to an extent of 70% to
90%. Dealer rebates—which are mostly unknown to customers—are passed
through only at about 30% to 40%. Sallee (2011) investigates the case of the
Toyota Prius, a car that was tax-subsidized for its fuel efficiency. Despite a
binding production constraint on the supply side, Sallee finds that the incentives are fully captured by the customers. He suggests that this is due
to a long-term pricing policy of the manufacturer. Verboven (2002) shows
that our approach of combining the two concepts of price discrimination and
incidence, and analyzing how the one translates into the other, indeed is obvious and feasible. He uses existing tax policies toward gasoline and diesel
cars in European countries to analyze quality-based price discrimination and
4
They find that consumers received the full amount of the rebate, that the program stimulated manufacturer rebates, and that the scrapping of old vehicles did not raise prices
in the used-vehicle market.
5
the implied tax incidence.
Our paper contributes to the literature in several ways. First, it fills the
existing gap of evaluating and quantifying the incidence of car scrappage
subsidies, programs that have played an important role in many countries
during the recent financial crisis. Due to exactly that popularity, it is very
likely that such interventions will be put in place again in the future. We analyze the most expensive such program ever launched and therefore focus on
a program with an extremely high potential to analyze this question. Second,
we present a simple heuristic model which helps explaining the mechanisms
at work. Since we develop a very simple and robust estimation strategy that
explicitly takes heterogeneity over different prices into account, we augment
the “standard” model as it is used in related research so far. This kind of
evaluation can easily be applied to similar programs in other countries now
and in the future.
The rest of the paper proceeds as follows. Section 2 gives a short overview
of the German scrappage program and the dataset we use. Section 3 presents
the estimated model. Section 3.1 provides model assumptions, Section 3.2
descriptive evidence, and Sections 3.3 and 3.4 outline the empirical approach
and show the results of the regression. Section 3.5 shows that the data
cover only a limited price range and Section 3.6 presents numerical values
for the price discrimination and the incidence over this price range. Section
3.7 summarizes the main results of the analysis. Section 4 presents a large
variety of sensitivity checks. Section 5 concludes.
6
2. Program and Dataset Description
2.1. Program
Incentives for car replacement designed as consumption subsidies are supposed to have three major benefits: (1) They are potentially environmentalfriendly by replacing old fuel-consuming cars with new ones with better emission standards. (2) They help the automotive manufacturing industry which
plays a particularly important role in Germany. Problems in this sector
would not only come along with the risk of layoffs and the corresponding
negative spill-overs, but also harm consumer confidence severely. (3) They
induce consumers to spend a multiple of the voucher’s value, and thereby
create a multiplier effect in the economy.
The idea for a scrappage program in Germany was introduced by the German vice-chancellor Steinmeier in an interview on December 27, 2008. Only
two weeks later, the government passed an economic stimulus package including a scrappage program. The program officially started on January 14,
2009 and first key points were published on January 16, 2009 by the responsible agency BAFA5 . The subsidy of e 2,500 could be requested by private
individuals who scrapped an old car which was at the time of scrappage at
least nine years old, and which had been licensed to the applicant for at least
12 months prior to the application. The new car had to be a passenger car
5
Bundesamt für Wirtschaft und Ausfuhrkontrolle (Federal Office of Economics and Export
Control).
7
fulfilling at least the emission standard Euro 46 and be licensed to the applicant. While the money was transferred only after the purchase, applicants
could be sure to receive the subsidy if the (simple) requirements were met
and provided that the budget was not yet exhausted. While the money was
granted to car buyers, car dealers in general organized the scrappage and
dealt with the federal agency. Many reported that they even treated the
amount of the subsidy as a down-payment.
The program turned out to be very popular, and the original budget risked
to be used up in April. The government raised the budget to e 5 billion,7
just a few days after switching from a paper-based to an online application
scheme. By September 2, 2009, the budget was depleted, having subsidized
the purchase of 2 million new cars. By the end of 2009, the bulk of requests
had been processed by the agency. National new car registration counts
show that registrations for lower-priced segments (Mini, Small, Medium, and
MPV) roughly doubled in 2009.
2.2. Data
We analyze a unique set of micro transaction data with 8, 156 observations. The data cover information from six randomly chosen car dealers in the
6
European emission standards define the acceptable limits for exhaust emissions of new
vehicles sold in EU member states. Actually, for the German case, this prerequisite was
redundant since all new cars bought in 2009 were Euro 4 equipped anyway.
7
To the best of our knowledge, this is the biggest budget provided for scrapping schemes
in this period. For an overview see http://www.acea.be/images/uploads/files/
20100212_Fleet_Renewal_Schemes_2009.pdf, last accessed on May 30, 2012.
8
center of (West) Germany over six different brands providing information on
the purchase of new cars over a time frame of four years (2007–2010). One of
those dealers covers two distinct brands, and one brand is represented by two
different dealers.8 As we will show in more detail in Section 3.2, this data is
very representative for Germany as a whole. Table A1 in the appendix gives
a summary of the distribution.
The data represent detailed information on the car (brand, vehicle class,
model) and on the transaction, i.e., the MSRP, the actual selling price9 , and
hence the granted discount. They also include dealer specifics, like the corresponding seller as well as buyer specifics, like age, sex and whether the
respective customer was a company employee or purchased a demonstration
car (see below for further explanations). Most importantly, we have information on whether a car was purchased with (CC ) or without (non-CC ) a
Cash-for-Clunkers subsidy within the year 2009.
Note that the MSRP is not a short-term pricing tool for manufacturers.
In general, catalogs and price lists are published once a year without an a
posteriori adjustment of the MSRP. Manufacturers also have much better
means of varying selling prices at their disposal, i.e., dealer and consumer
cash incentives such as those discussed by Busse et al. (2006). In contrast
to the MSRP, these incentives can be changed at very low cost, and are
8
9
For data privacy reasons, we never report the name of a respective dealership or brand.
Note that trade-in values do not affect the data. Trade-ins are treated as fixed-value assets
which are shifted to the used car department of a dealership. Actual trade-in values were
therefore treated as cash-substitutes and consequenlty did not affect the reported prices.
9
unpredictable for the buyers, as well as the dealers who normally do not know
which programs will be issued by the manufacturer next month. In contrast
to the MSRP which is the same all over Germany, incentive programs can also
vary geographically. Manufacturers therefore have good reasons to keep the
MSRP stable and vary incentives in order to meet changing local conditions
without jeopardizing their long-term pricing strategy.10
Table 1 shows how the number of purchases is distributed over the years
2007-2010. Year 2009 is split into non-subsidized (Non-CC ) and subsidized
(CC ) purchases. On average, we observe about 1, 600 sales a year, with twice
that amount in 2009 (1, 649 non-subsidized plus 1, 541 subsidized ones).11
Table 2 provides summary statistics of essential variables. The average car
cost about e 25, 600, and was discounted approximately 17%. Roughly 30%
of all buyers were female. About 16% of all purchases were of demonstration
cars (so-called “floor models”) and 12% refer to sales to employees of auto
manufacturers (called “company employees” henceforth). The average buyer
age was 47 years, but we only observe 1, 425 (out of 8, 156) data points
featuring customer age information.12
10
This is why we consider this variable strictly exogenous, meaning that the MSRP did
not react due to the implementation or the process of the scrappage program.
11
CC purchases are concentrated in the months February to October, and then decline
(see Table A2 in the appendix). This is in line with the distribution of applications for
the subsidy as reported by the BAFA.
12
The remarkably high percentage discount over 50% (max) was due to the fact that
demonstration cars as well as company employees benefit from huge (and) additional
discounts. The high discount of more than e 50, 000 was observed for a demonstration
car of the most expensive category (luxury car segment).
10
Table 1: Number of Purchases over Time by Car Dealers and CC
Dealership
Dealer
Dealer
Dealer
Dealer
Dealer
Dealer
1
2
3
4
5
6
Total
Year of Purchase and Clunker’s Premium
2007
2008
2009
2010
Non-CC
CC
315
250
263
633
81
12
443
235
314
484
67
158
1554
1701
587
268
277
346
60
111
1649
3190
317
330
359
135
43
357
1541
504
381
286
270
43
227
1711
Note: Non-CC are non-subsidized purchases, CC subsidized ones.
Table 2: Summary Statistics: All Data
Variables
Mean
SD
Med
Min
Max
N
Discount in Percent
Discount in 1000 EUR
MSRP in 1000 EUR
Clunker’s Premium (CC)
Demonstration Car (DC)
Company Employee (CE)
Female
Age at Purchase
16.91
4.18
25.62
0.19
0.16
0.12
0.29
47.23
8.68
3.23
14.37
0.39
0.37
0.32
0.45
14.93
16.40
3.44
21.50
0
0
0
0
48
0.00 53.37
0.00 51.81
8.19 198.66
0
1
0
1
0
1
0
1
18
89
8,156
8,156
8,156
8,156
8,156
8,156
8,156
1,425
Note: MSRP is the manufacturer suggested retail price. CC is a dummy variable indicating whether the
buyer of a car received the scrappage subsidy. DC is a dummy variable indicating whether a buyer bought
a demonstration car. CE is a dummy variable indicating whether the buyer was an employee of a car
manufacturing company. Female is a dummy of female buyers, the summary statistics therefore report
the share of women, age at purchase is the age of the buyer at the time of purchase.
11
3. Analysis
In a first step, we present a heuristic model regarding the German scrappage program and its anticipated effects on the subsidys incidence. We then
provide descriptive evidence and, thereafter, our regression analysis. We
start with a standard specification to estimate the average impact of receiving the subsidy on the percentage discount. In this model, we include all relevant control variables as discussed above plus fixed effects for time, brand,
dealership, and seller. Afterwards, we augment this basic specification by
additionally interacting the scrappage dummy with the MSRP, allowing for
heterogeneity across the car price range. This preferred specification reflects
our model assumptions. Taking into account the distribution of purchases
and the share of subsidized purchases over the price range, we show for which
interval of MSRP our results are reliable. To illustrate the estimated differences, we show the magnitude of price discrimination in percentage points
and Euros as well as the incidence over what we consider the relevant price
range. We close this section by summing up and discussing our main findings.
3.1. Model Assumptions
There exists a sizable public finance literature on (tax) incidence in markets with imperfect competition and one could justify just about any difference in terms of incidence across subsidy participants and non-participants
12
as seeming credible.13 We therefore develop sound assumptions regarding our
anticipated evaluation outcomes. Those assumptions are made on grounds
of knowledge of the institutional design of the program, the car market in
general and of different market conditions and price elasticities of demand
across car segments. In essence, we expect our results regarding the incidence
of the scrappage subsidy to be in line with an optimal long-run pricing strategy of manufacturers and dealers in the car market. This pricing strategy is
supposed to reflect different price elasticities of demand and market conditions in different car or price segments. Thus, we expect our results to be
heterogenous over car prices. For the following argumentation, it is pivotal to
remember that dealers were able to reliably identify two different groups of
customers within the year of 2009. Since they managed the scrapping of the
old car and dealt with the responsible federal agency BAFA, dealers always
were able to distinguish between subsidized and non-subsidized buyers.
With regard to the lower price segment, two facts are crucial. On the one
hand, the scrappage program shifted demand heavily toward smaller cars.
This gave dealers market power, and thus allowed for price making. Sudhir
(2001) states that the supply side has a motivation to be aggressive in the
small-car market (the entry level segments) to increase profits and market
13
To name just very few examples, Stern (1987) provides theoretical work on tax incidence
showing that there is the possibility of either over- or undershifting of (different) taxes (to
consumers); Delipalla and O’Donnell (2001) deliver a related application to the cigarette
industry; Anderson et al. (2001), again, show that incidence amounts of more than 100%
are theoretically possible.
13
share. In contrast to non-subsidized customers, buyers receiving the subsidy
and aiming to join this market segment were presumably relatively more
price-inelastic since the subsidy was available only for a very short period
of time and because people were keen on seizing the opportunity of receiving a e 2,500 check. These people wanted to buy now not only because the
subsidy was not available for a long time but also since nobody could exactly know when it would expire. In contrast, the non-subsidized customers
could be more patient. In addition, close substitutes for small cars were not
available since downgrading was hardly possible and the demand shock essentially affected all brands alike. Altogether, this suggests that there was
room for price discrimination based on observables (the scrappage premium
information) in the lower price segment. On the other hand, it is well-known
that competition in the market for small cars is quite high (Sudhir (2001))
and competition presumably increased in 2009 due to the scrappage subsidy.
However, competition limits the scope for price discrimination. In particular, in a competitive market where brand loyalty is not (yet) established,
price elasticity is presumably not spectacularly different across buyer groups.
Moreover, within a certain class of cars, the potential buyer certainly could
choose from different brands and dealerships leaving her with a supposably
higher intra-segment price elasticity of demand.14 Dealers and manufacturers
14
Berry et al. (1995) state that the most elastically demanded cars are that in small market
segments. Cross-price elasticities (large for cars with similar characteristics tend to be
bigger for cheap cars as compared to expensive ones.
14
contemplating higher markups for subsidy receivers therefore had to trade off
higher margins against lower sales in the short run. Also, long-run pricing
considerations may have played a role in the pricing policy (also compare
Sallee (2011)). Together, these suggest why there could be some price discrimination against scrappage premium receivers within the small-car market
but also why this group of buyers should still receive the bulk or even all of
the subsidy.15
Things were very different in the upper price segment. Here, the market was slack and unsold cars were piling up. From an upper-segment car
dealer’s perspective, an interesting and unique potentially profit-maximizing
strategy was possible due to the subsidy. Customers buying in the large car
segment, could be divided into two observable groups (as in the lower price
segment just with different “characteristics”) with distinct price elasticities
15
Busse et al. (2006), who analyze car market cash incentives, find that between 70%
and 90% of the customer promotion amount remains with the buyer, i.e., the seller
reaps only a small fraction of the promotion. Since a customer promotion is quite
comparable to a buyer subsidy granted by the government, the two instruments are
in fact comparable. This supports our assumption that the subsidy should remain—in
large part or even entirely—with the buyer in the small car market. Sallee (2011) finds
that a customer-directed tax subsidy for the Toyota Prius, a car that would fall into
the small to medium-size car market, is fully captured by customers, although sellers
face a binding production constraint. He suggests that this is due to a long-term pricing
policy of the manufacturer. In the case of the German scrappage premium, a production
constraint was also binding in the small car segment since the subsidy caused a run on
these cars. We therefore would argue, again, that in this kind of car price segment,
the subsidy amount should (almost) be fully captured by the consumer. While Sallees
explanation builds mainly on long-run pricing policy of manufacturers, we conjecture
that in the German case, increased competition due to the demand shock induced by
government intervention additionally could lead to the fact that the supply-side would
only capture a small or even negligible fraction of the subsidy in the small car segment.
15
of demand, namely regular large car buyers who did not (and did not tend
to) receive the subsidy, and subsidized buyers who would typically not buy a
large car. Non-subsidized customers in the upper price segment should not
receive exceptional rebates since that would interfere with the well-known
cooperative pricing strategy of car manufacturers toward brand-loyal longterm customers (Sudhir (2001)).16 This would unnecessarily erode margins
without increasing long-term demand in that customer segment. In fact, interviews with car dealers suggest that a selection effect could have worked
in their favor. Subsidized buyers were typically not customers buying pricey
cars, and would usually not upgrade from a clunker to a new expensive car.17
Therefore, their price elasticity of demand for large cars was quite high. To
this buyer group, in contrast to the subsidized buyers within the small car
segment, “substitutes” indeed have been available since downgrading to a
medium priced car was easily possible. All this should lead the supply side
to offer exceptionally high discounts (on top of the subsidy amount) to this
customer group. Moreover, offering high rebates to subsidized customers
would not interfere with long-run pricing considerations of manufacturers
since the new customer group was a one-time target without any significant
downside risk with regard to their long-run car demand for manufacturers of
large cars.
16
17
Also compare Goldberg (1995) regarding brand loyalty in the car market.
Since facing a flat subsidy, buyers should not be willing to trade in a “clunker” worth
more than e 2,500 and have to accept diminishing benefits from purchasing more expensive cars.
16
In summary, we expect our results to be heterogenous over car prices.
Firms in the small car segments tend to be aggressive, and room for price
discrimination based on observable characteristics—such as the information
of receiving the scrappage premium—is limited due to strong competition.
We therefore assume that for cheaper cars, the bulk or even all of the subsidy
amount remains with the buyer, implying incidence amounts of slightly below
or at just 100%. With regard to the larger car segments, aggressive pricing is
usually avoided since such pricing behavior reduces margins without increasing demand of regular customers who tend to be very brand-loyal in that
market segment. However, granting huge discounts to a new group of customers, who could be distinguished from the old and loyal ones based on the
scrappage premium information, offered a one-time opportunity to increase
profits for large car manufacturers and dealers by increasing sales. Hence,
we assume that subsidized buyers of large cars eventually received extra discounts on top of the scrappage subsidy amount, implying incidence amounts
of more than 100%.
3.2. Descriptive Evidence
Figure 1 shows the number of observed purchases for the different vehicle
classes over the observation period.18 Mainly cheap vehicle classes like A
(Mini), B (Small), C (Medium), and M (MPV) benefited from the program.
18
The classification A, B, C, D, E, F, J, M, S is in accordance to the EU classification.
For an overview see http://ec.europa.eu/competition/mergers/cases/decisions/
m1406_en.pdf, last accessed on January 26, 2012.
17
S - Sports Coupés
M - MPV
J - SUV
F - Luxury Cars
E - Executive Cars
D - Large Cars
C - Medium Cars
B - Small Cars
A - Mini Cars
1200
1000
800
600
400
Number of Purchases
200
0
Year of Purchase
2007
2008
2009 CC
2009 other
2010
Note: A, B, C, D, E, F, J, M, S are auto segments according to the EU car classification. 2009 CC are car
purchases in 2009 involving the scrappage subsidy, 2009 other are non-subsidized purchases. SUV stands
for Sport Utility Vehicle, MPV for Multi Purpose Vehicle
Figure 1: Number of Purchases over Time by EU Vehicle Class
18
We do not find many additional purchases in vehicle classes D (Large), E
(Executive), F (Luxury), and S (Sports Coupés).19 Overall, it seems that
subsidized purchases were made over and above the regular purchases, and
were not pulled forward from the following purchase period.20 Note that the
pattern of this sample depiction is almost identical to what new vehicle registration counts for the whole of Germany looked like.21 This indicates that
we are dealing with very representative transaction data, and have sufficient
external validity to transpose our results from the research sample to the
target population (from which the sample was drawn), i.e., car dealerships
in Germany.22
Figure 2 shows the development of the discount over time per vehicle
class. As mentioned previously, inexpensive vehicle classes experienced an
increase in car purchases, while the more pricey segments faced a staggering
or declining demand. We can see that some of the segments which experienced a positive demand shock (Mini and Small) are the ones which receive a
smaller discount throughout 2009 when purchased as a CC car compared to
non-CC cars. For the other, more expensive segments, the opposite happens:
19
This is not surprising since expensive cars are predominantly purchased by corporate
customers, so they obviously played a minor role within the scrapping context.
20
Böckers et al. (2012) analyze the pull-forward effects of smaller vehicle classes in Germany. Heimeshoff and Müller (2011) provide estimates of how many additional cars
were sold due to scrappage programs in 23 OECD countries.
21
Figure A1 in the appendix shows the new car registrations for non-commercial cars in
Germany for the years 2008-2010.
22
We could not have conducted the same analysis by just using the registration count
data, since most of the relevant information is missing therein, for instance the amount
of discount and the indicator for whether a subsidy was received.
19
A - Mini Cars
B - Small Cars
25
20
Discount in Percent of MSRP
C - Medium Cars
24
22
20
18
16
22
20
18
16
14
15
10
D - Large Cars
E - Executive Cars
20
15
10
F - Luxury Cars
25
20
20
15
15
10
10
5
J - SUV
M - MPV
S - Sports Coupés
25
20
18
16
14
12
30
20
20
15
10
10
2007q1
2008q1
2009q1
2010q1
2011q1
2007q1
2008q1
2009q1
2010q1
2011q1
2007q1
2008q1
2009q1
2010q1
2011q1
Quarter of Purchase
Non-CC transactions
CC transactions
Note: Average discount in percent of MSRP over quarters of years across EU vehicle classes. SUV stands
for Sport Utility Vehicle, MPV for Multi Purpose Vehicle.
Figure 2: Percentage Discount over Time by EU Vehicle Class
20
CC customers received a comparatively higher discount.23
The same pattern arises within vehicle classes (see Figure A2 in the appendix), namely that subsidized cars are cheaper than non-subsidized ones.
We therefore control for MSRP in our regression model rather than interactions of “make, model, and turn” as in Busse et al. (2006). More importantly,
using MSRP allows to control for differences in optional equipment since any
additional feature is included in the catalog price.
In a next step, we deepen this discussion a little further by moving from
a graphical to a numerical focus, and present essential figures. First, we take
a closer look at 2009 (Table A4 in the appendix gives summary statistics for
that year only). The average MSRP in 2009 was about e 2, 500 lower compared to the 2007-2010 mean due to a difference in composition: more small
and smallest cars were bought in that period. The average discount in 2009
(17.7%) is relatively stable when compared to the discount in the 2007-2010
sample (16.9%). About 14% and 13% of the 2009 purchases were of demonstration cars and made by company employees respectively. Table 3 shows
the difference for relevant variables between subsidized and non-subsidized
purchases within the year 2009. Non-CC cars received a discount of 17.67%,
23
Summary statistics for the MSRP over vehicle classes are given in Table A3 in the
appendix. It shows that prices rise monotonically over the vehicle classes A through
to F. The mean price of MPVs is similar to Medium Cars; SUVs cost on average as
much as Large Cars; Sports Coupés are comparable to Executive Cars. The standard
deviation of the prices of the last three categories are about twice as big as those of their
respective reference category. The last three vehicle classes are therefore consistent with
the described pattern.
21
whereas CC cars received 16.51%.24 The corresponding absolute values are
e 4, 686 and e 3, 235 respectively. These differences are significant at the
1% level. Yet, we have to take the MSRP into consideration: Non-CC cars
on average cost e 26, 720, whereas CC cars amounted to about e 19, 062.25
This means that customers who called upon the subsidy on average asked for
smaller (cheaper) cars than customers who purchased without the subsidy
denoting differences in the group compositions of CC and non-CC customers.
We therefore have to control for MSRP in our regression analysis rather than
for vehicle class. Furthermore, about 25% of the non-CC group, and about
39% of the CC group was comprised of women. The shares of demonstration
cars and company employees are 19% vs. 10% and 17% vs. 8% (non-CC vs.
CC ) respectively. The last information is important because the unequal
share of the two high-discount categories might be driving the difference in
percentage discount. Both categories make up for a smaller share in the CC
group compared to the reference group, which implies that the average discount of CC purchases would rather be biased downward.26 In the following
analysis, we therefore control for both groups.
Both the descriptive and graphical evidence suggest that price discrimina24
Table A5 in the appendix gives an overview of the development of the percentage discount
over the years including a CC/non-CC distinction.
25
The distribution of the MSRP of subsidized cars is concentrated among lower prices. Its
median is e 17, 000, and the 75th percentile is at about e 22, 000.
26
Table A6 in the appendix shows the percentage discount by different types of purchases.
Standard purchases earned lower discounts (14%) than company employees (26%) or
demonstration cars (23%).
22
Table 3: Summary Statistics: Comparison within 2009 by CC
Variables
Discount in Percent
Discount in 1000 EUR
MSRP in 1000 EUR
Demonstration Car (DC)
Company Employee (CE)
Female
Non-CC
Mean
SD
17.67
4.69
26.72
0.19
0.17
0.25
8.73
3.80
15.27
0.39
0.38
0.43
CC
Mean
16.51
3.24
19.06
0.10
0.08
0.39
SD
Diff
Mean
6.67
2.15
7.56
0.29
0.28
0.49
-1.16
-1.45
-7.66
-0.09
-0.09
0.14
Note: Non-CC are non-subsidized purchases, CC subsidized ones. The last column gives the difference
in means between CC and non-CC purchases. MSRP is the manufacturer suggested retail price. DC is
a dummy variable indicating whether the buyer bought a demonstration car. CE is a dummy variable
indicating whether the buyer was an employee of a car manufacturing company. Female is a dummy of
female buyers, the summary statistics therefore report the share of women.
tion across consumers of different market segments as well as price discrimination between subsidized and non-subsidized buyers may have been present.
Subsidized customers who bought (very) small up to medium cars received
a smaller discount compared to non-subsidized customers; when purchasing
bigger cars the opposite seems to be true, namely that subsidized buyers
received a higher discount than non-subsidized ones. Before drawing further
conclusions however, we need to control for various aspects such as the exact MSRP, the year of purchase, the kind of dealer and brand, as well as
high-discount groups.
3.3. Basic Specification
In our most basic specification, we follow the “standard model” of, e.g.,
Busse et al. (2006) and estimate the incidence effect as a weighted average.
Hence, in this first step, we neglect potential heterogenous impacts of re23
ceiving the subsidy on the percentage discount of car prices. After we get
an idea of the average influence of the government intervention, we then—
in the next section—explicitly consider our heuristic model framework and
allow for heterogeneity across car price segments by augmenting this basic
specification.
We start by estimating the following regression model:
discount = α + βCC + γM SRP + θ0 X + (1)
The dependent variable (discount) is the discount in percent of the MSRP
granted for a single car purchase in percent.27 The key explanatory variable of
interest is CC, the Cash-for-Clunkers dummy, i.e., an indicator as to whether
a car was purchased with the scrappage subsidy (CC = 1) or without it
(CC = 0). M SRP denotes the manufacturer’s suggested retail price or
catalog price (in e 1,000). The vector X contains a set of other controls.
Brands and dealers are modeled as seven brand-dealer dummies, i.e., there
is a dummy for each combination of brand and dealer. Dummies for buyers
who are employees of car manufacturing companies (“company employees”,
27
So it is
discount = 100 ∗
M SRP − Selling P rice
MSRP
with the selling price including the subsidy amount.
24
(2)
CE) and demonstration cars (DC ) are included. Also a dummy for each
individual seller is included, as well as a sex dummy for buyers and year and
month dummies to capture seasonalities and macroeconomic effects. The
error term is represented by .
The estimated coefficients are α, β, γ and the vector θ. The key coefficient
of interest in this specification is β. It measures the percentage difference
in discount a subsidized buyer received in comparison to an non-subsidized
buyer. A positive (negative) estimate of β indicates that subsidized buyers
received a higher (lower) discount than non-subsidized buyers, controlling
for the covariates mentioned above. The coefficient γ measures how dealers’
discount policies differs across price segments. To be precise, γ measures
how the discount changes as the MSRP increases by e 1, 000, holding other
things constant.
Column (1) of Table 4 reports the results of estimating the specification
in Equation (1). The estimated coefficient β measuring the effect of receiving the scrappage subsidy on the discount granted for a car purchase is 0.4.
It is positive and statistically different from zero at the 10%-level.28 This
suggests that the overall pass-through of the subsidy was negative, i.e., dealers grant a 0.4 percentage points bigger discount for CC purchases than for
28
Similar to Busse et al. (2006) who identify the very car based on make, model, and its
very specification, we also ran the regressions with make-model interactions rather than
the MSRP on the right-hand side. In this case, the coefficient of CC gets bigger (0.59
if we control for brands and dealerships, 0.63 if we do not). However, none of these
coefficients is statistically different from the 0.40 of the reported value.
25
non-subsidized ones, controlling for the discussed covariates. Although the
coefficient is quantitatively small (compared to a mean value of about 17%,
see Section 2.2), the result is surprising since a capturing of a subsidy of more
than 100% is not consistent with the related empirical literature.29 The value
of 0.05 for γ suggests that the percentage discount grows at a rate of about
0.05 percentage points with every e 1,000 of MSRP. This means that a difference of e 20,000 implies a higher discount of one percentage point. Before
discussing the controls in vector θ, consider the full model which takes into
account that the effect is heterogeneous over the price range.
3.4. Full Specification
Specification (1) has a shortcoming, namely that it restricts the effect of
receiving the subsidy on the discount to be uniform across price segments.
As discussed in Section 3.1 however, we expect our results to be heterogenous
across car prices. In Section 3.2 we already got an idea how market conditions
and the discounts themselves were different over different vehicle classes and
price segments.
To account for this heterogeneity, we interact the dummy CC with the
MSRP (CC ∗ M SRP ) and estimate the extended regression model in Equa29
Busse et al. (2006) find that 70%-90% remain with the customers, Sallee (2011) finds
that customers capture 100% of the subsidy.
26
tion (3).30
discount = α + βCC + γM SRP + δCC ∗ M SRP + θ0 X + (3)
Results are presented in column (2) of Table 4. Estimating this specification, all the essential coefficients—β, γ, and δ—are statistically significantly
different from zero at the 1% level. The results confirm our expectations:
controlling for individual- and dealer-specifics as well as time trends and
high-discount groups, we find a strong relationship between the MSRP, the
subsidy and the discount in percent. We see that β, the coefficient for CC,
is negative, with −4.4 being rather large,31 and highly significant. The estimate for δ is 0.24 and hence positive, implying that the more expensive a car
was, the more additional discount was granted if the buyer benefited from
the subsidy. The coefficient of M SRP (γ) is 0.03 and thus a little smaller
than in Specification (1), but qualitatively not different.32
Keeping everything else constant, the results allow to depict two different
functions: one for subsidized and one for non-subsidized buyers, denoting the
latter as “baseline function”. Recall that the estimated coefficient for the CC
dummy is −4.4 which means the y-intersect is 4.4 percentage points lower
30
As discussed previously, we cannot simply interact CC with a set of vehicle class dummies
because within each such class, the two groups (subsidized and non-subsidized purchases)
differ.
31
Note that the dummy itself has no meaningful interpretation as it measures the difference
from the overall constant for a price of zero. Interpreting this value as such would be an
inadmissible extrapolation.
32
Clustered standard errors would not change these results, see Section 4.
27
for the CC-function than for the function of non-subsidized purchases. The
coefficient of the interaction term is 0.24, so this function is steeper than the
baseline function with a slope of 0.034 (coefficient for MSRP); with every
additional e 1,000 of MSRP, the expected discount of subsidized purchases
becomes 0.24 + 0.034 = 0.274 percentage points bigger. For non-subsidized
cars, it grows at the rate 0.034 percentage points per e 1,000 of MSRP. All
the relevant coefficients are statistically significant at the 1% level.
Throughout the different specifications, the controls in vector θ remain
stable. For instance, the coefficients of the controls for company employees
(CE) and demonstration cars (DC ) hardly change.33
Note that we do not report the estimated coefficient for sex (taking the
value one if the buyer was female, zero otherwise). In all specifications,
female turns out to be both economically and statistically insignificant.34
Due to the interaction terms, the interpretation of the results is facilitated
if we do not discuss single coefficients, but rather the expected percentage
discount as a (linear) function of the MSRP. For the group of non-subsidized
buyers (CC = 0), this function has a y-intersect (M SRP = 0) at the constant
of 18.05 and a slope coefficient equal to 0.0335.35 For the group of subsidized
33
These percentage values experienced some downward adjustment compared to the descriptive statistics (see Section 2.2), but are still considerably lower compared to a “normal” consumer who bought a “normal” car, i.e., when the purchase involved neither a
company employee nor a demonstration car.
34
This finding is in line with Goldberg (1996) who shows there is no evidence for discrimination against female car buyers.
35
More precisely, the y-intersect depends on the constant as well as the coefficients of
any (binary) control variable. To focus on the relevant part of the function, and since
28
Table 4: Linear Regression Estimation Results of Different Specifications
Dependent Variable: Discount in Percent of MSRP
VARIABLES
CC
(1)
(2)
0.398*
(0.233)
0.0453***
(0.00818)
11.01***
(0.277)
11.50***
(0.313)
17.69***
(1.670)
-4.401***
(0.503)
0.244***
(0.0228)
0.0335***
(0.00800)
10.88***
(0.276)
11.56***
(0.312)
18.05***
(1.673)
8,156
0.488
8,156
0.496
Yes
Yes
Yes
Yes
Yes
n/a
Yes
Yes
Yes
Yes
Yes
18.06
CC*MSRP
MSRP
DC
CE
Constant
Observations
Adjusted R-squared
Year Dummies
Month Dummies
Sex Dummy
Seller Dummies
Dealer Dummies
Intersect
Note: *** significant at the 1%-level, ** significant at the 5%-level,
* significant at the 10%-level. Robust standard errors (HC3) in
parentheses. CC: dummy for subsidized (Cash-for-Clunkers) transaction, MSRP: manufacturer’s suggested retail price in e 1000, DC:
dummy for demonstration car, CE: dummy for employees of auto
manufacturing companies. Year = 2008 (2009) (2010) are dummy
variables for the given years, 2007 is the base year. Intersect indicates where the estimated value for subsidized purchases is equal to
the one of baseline function.
29
buyers (CC = 1), the function has a y-intercept of 18.05 − 4.401 = 13.65
and a slope coefficient equal to 0.2440 + 0.0335 = 0.2775. The latter line is
therefore steeper than the former but starts lower. Thus, the two functions
intersect at
Ilin = −β/δ
(4)
where β measures the downward shift of the CC curve for MSRP zero,
and δ the difference between the slope of the CC and the non-CC functions.
Equation 4 therefore gives the MSRP where both functions intersect. This
value is reported at the bottom of Table 4 (Intersect), it is about e 18,000
for specification (2).
A general conclusion is that subsidized buyers of the first quartile faced
negative price discrimination, i.e., they paid more (experienced a lower discount) if they received the subsidy. Since the scrappage program shifted
demand heavily to the lower-priced segments, car dealers could impose a
price markup by granting less discount. In contrast, subsidized buyers in
the third (and fourth) quartile faced positive price discrimination, meaning
they had to pay less (received more discount) if buying with the subsidy. In
this much slacker part of the car market, dealers used additional discounts
in order to seize a one-time opportunity of selling to very elastic (subsidized)
customers instead of losing them to competitors or lower car segments. At
consideration of these additional controls does not alter the results, we neglect this point
in the discussion.
30
an aggregate level, the positive price discrimination in the upper part of the
distribution overcompensates the negative effect in the lower part.36 Within
the second quartile finally, the difference between the CC and the non-CC
function is just zero. This implies that within the second quartile of MSRP,
car dealers did not price discriminate at all, and CC customers received the
full amount of the subsidy of e 2, 500.
3.5. The Relevant Price Range
But how relevant is the region we are considering? Moreover, are subsidized and non-subsidized purchases sufficiently balanced, meaning whether
the shares of CC and non-CC purchases are rather equal and therefore comparable? If this was not the case, our results might be misleading. Figure 3
gives an insight into the distribution and adds the share of CC purchases by
MSRP.37 The dash-dotted line shows the CC share as a falling function of
MSRP. This is what we expected, given that the lump-sum subsidy matters
relatively more for cheaper cars. However, in a region below e 12,000, the
share is larger than 60%, reaching up to 80% for cars of an MSRP of about
e 9,000. We claim that this part of the distribution lacks common support
because its composition is too unbalanced. The graph of the distribution
(dotted density plot) is very steep on the left side, which means that there
36
The reported coefficient β on the CC dummy from specification (1) of Table 4 can be
interpreted as a weighted average.
37
To calculate the share of CC in Figure 3, we rounded the MSRP to e 1,000 and calculated
the share of subsidized purchases in 2009 for each e 1,000 price interval.
31
were relatively few purchases at a price range of about e 8,000, but already
quite a few at a price of e 10,000 to e 12,000. Cutting off this fringe, we see
that from an MSRP of e 12,000 on, the data points are comfortably dense
enough, and the distribution between CC and non-CC purchases is rather
balanced with about 60% or less. At the other end of the distribution, the
share of CC purchases drops below one third at a price of about e 32,000.
We choose this point as an upper bound for the following discussion. At this
point, we still observe a sufficiently balanced distribution between CC and
non-CC purchases which then steadily shrinks along with the density. In
the following discussion, we therefore focus on a price range from e 12,000
to e 32,000 which we judge to be the most relevant interval of our data with
a solid balance of CC and non-CC purchases.
3.6. Price Discrimination and Incidence
As a next step, we quantify the exact amount of price discrimination and
the corresponding incidence over the price interval for which our results were
found to be relevant. Table 5 yields an overview regarding that quantification
for the linear model (specification (2)). It provides the percentage (PD (%))
and the respective absolute (PD (e )) discount received for a certain MSRP
(what we refer to as “price discrimination”) as well as the corresponding part
(Inc (%)) of the e 2,500 subsidy which remained with the consumer (what
32
.08
Kernel Density of MSRP
.02
.04
.06
Discount in Percent of MSRP
16
18
20
0
14
22
.8
Share of CC-purchases
.6
.2
.4
0
8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
MSRP in 1000 EUR
non-CC linear
Density Plot
CC linear
Share of CC
Note: non-CC linear is the function of non-subsidized purchases based on specification (2), CC linear
is the function of the subsidized ones. The density plot refers to the MSRP in 2009. To calculate the
CC -share, we rounded the MSRP to e 1,000 and calculated the share of subsidized purchases in 2009 for
each e 1,000 price interval. The bold vertical lines indicate the boundaries of the interval we consider
the relevant price range. The thin vertical line at e 18,000 indicates the price where we observe just no
difference in discount received between both buyer groups (the intersect).
Figure 3: Linear Model with Distribution and CC -Share
33
we refer to as “incidence”).38 Recall that the point where the two groups
do not differ at all was at e 18,000. At that point the incidence is 100%, a
result that reflects the findings of Sallee (2011), for a car that would fall into
our second quartile of MSRP. For a cheap car with a MSRP of e 12,000, the
linear model yields a price discrimination of −1.48% or e −178, i.e., dealers
skimmed off about 7% of the subsidy amount (e 2,500). This translates into
an incidence amount of 93%, which would be an upper bound when compared
to the values Busse et al. (2006) find (70 − 90%).
Results for higher-priced cars are more remarkable: a car which cost
e 28,000 and therefore is at the lower end of the fourth quartile of MSRP,
would benefit from an additional discount of 2.42% or e 678, which means
that buyers received an additional discount (on top of the subsidy they received) of around 27%. A car purchase at the very end of our relevant MSRP
range (e 32,000) caused an extra 3.4% or approximately e 1,100, which is
44% of the scrappage subsidy amount. Speaking of incidence this means
that 127% and 144% of the subsidy amount for a e 28,000 and a e 32,000
automobile “remained” with the buyer respectively. Incidence amounts located above the 100%-threshold, in our case clearly distant from that, are
empirically rarely found.
38
The Euro values were calculated from the corresponding percentage values and the
MSRP, not from a separate estimation with discount in Euro as a dependent variable.
34
Table 5: Price Discrimination and Incidence over different MSRPs
MSRP
12,000
14,000
16,000
18,000
20,000
24,000
28,000
32,000
PD (%)
PD (e )
Inc (%)
-1.48
-0.99
-0.50
-0.01
0.47
1.45
2.42
3.40
-178
-139
-80
-2
94
348
678
1088
93
94
97
100
104
114
127
144
Note: The table presents price discrimination for a given MSRP in percentage points of MSRP (PD (%))
and Euro (PD (e )) based on the linear model from specification (2) as well as the respective Incidence
(Inc (%)) which indicates what percentage part of the subsidy remained with the consumer.
3.7. Results
The main result of this paper is that the incidence of the subsidy strongly
and significantly varies across price segments. We focused most of our discussion on three price segments that roughly correspond to the first, second,
and third price (MSRP) quartile.39
In the first quartile that mainly covers mini cars and to some extent small
cars, subsidized buyers received slightly lower discounts than non-subsidized
ones controlling for covariates. In the second quartile—mainly consisting of
small and medium cars—discounts between the two buyer groups did not
differ much, implying that the full subsidy amount remained with the buyer.
The most striking result was found for sales in the upper half of the price
39
We also consider the lower part of the fourth quartile of MSRP since we argue that our
relevant price range reaches e 32,000.
35
distribution. We focused particularly on the third price quartile (mainly
medium and large cars), where subsidized and non-subsidized sales were quite
balanced. In this segment, scrappage premium receivers were granted much
higher discounts than regular customers. The incidence in this price segment
was such that subsidized buyers received huge extra discounts from sellers
over and above the government premium.
Our result for the lower price segments—loosely speaking for the bottom
half of the distribution—is in line with the results in Busse et al. (2006) and
Sallee (2011). Busse et al. (2006) find that between 70% and 90% of the customer promotion amount remains with the buyer, i.e., the seller reaps only
a small fraction of the promotion. Since a customer promotion is quite comparable to a buyer subsidy granted by the government, the two instruments
are in fact comparable, and so are our results of roughly 90% of the subsidy
amount remaining with the buyer in the first quartile of MSRP. Sallee (2011)
finds that a customer-directed tax subsidy for the Toyota Prius, a small car
that would fall into our second price quartile, is fully captured by customers,
although sellers face a binding production constraint. In the case of the German scrappage premium, a production constraint was also binding in the
small-car segment since the subsidy caused a run on these cars. Our results
in the second price quartile are therefore fully in line with Sallee’s results.
While his explanation builds mainly on long-run pricing policy of manufacturers, we conjecture that in the German case, increased competition due to
the demand shock induced by government intervention additionally explains
36
why the supply-side only captured a small or even negligible fraction of the
subsidy in the bottom half of the distribution.
In the upper price segment, non-subsidized (regular) customers did not
receive exceptional rebates since that would interfere with the well-known cooperative pricing strategy of car manufacturers toward brand-loyal long-term
customers. This would have unnecessarily eroded margins without increasing long-term demand in that customer segment. Subsidized buyers, on the
other hand, were typically not customers buying pricey cars, and would usually not upgrade from a clunker to a new expensive car. Therefore, their price
elasticity of demand for large cars was quite high, which lead the supply side
to offer exceptionally high discounts to this customer group, consistent with
our findings of incidence amounts of even more than 140%. Offering such
high rebates to subsidized customers did not interfere with long-run pricing
considerations of manufacturers since the new customer group was a one-time
target without any significant downside risk with regard to their long-run car
demand for manufacturers of large cars.
In summary, our assumptions did hold. We did see that having a clunker
is a signal that (1) you have a car at your disposal, (2) you are not likely to
target the expensive market (because you have a “clunker”) and (3) you want
to buy now, before the program expires. In the upper price segment, factor
(2) dominates and people get an extra discount because they are likely to
be price sensitive customers who differ substantially from the typical luxury
buyer. In that segment, our assumptions in terms of respective incidence
37
amounts have even been exceeded. In the cheap market, subsidized buyers
are less price elastic because they needed to buy now, whereas the nonsubsidized customer could have been more patient. Still, in that segment
dealers are not able to reap off a lot of the subsidy amount due to increased
competition in the year of the policy intervention.
4. Sensitivity Analysis
Our identification strategy employed was a simple difference design where
the discounts received by the subsidized buyers were compared to that of the
non-subsidized buyers. Hence, the assumption required for identification is
knowledge regarding what prices were actually paid by those receiving the
subsidy and a counterfactual estimate of what those people would have paid
if they had purchased the same car without the subsidy. As shown in Section
3.2, the CC and non-CC types differ significantly in almost every observable
way. The estimates in our paper control for all of these factors (company
car etc.), but it is possible that these customers also differ in unobservable
characteristics that are not controlled for and could introduce some bias.40
This is why, in the following sensitivity analysis, we challenge our results in
every reasonable way possible and present a huge variety of robustness checks
to our preferred specification.
40
Certainly, the simple difference design is a limitation of that analysis. A difference-indifferences strategy would be desirable, but the CC versus non-CC group simply cannot
be established outside of the treatment period, i.e., for years other than 2009.
38
So far, the presented results restrict the econometric model to a linear
form which might be too inflexible. We therefore present a quadratic version
of the same underlying economic model as well as a reformulation to a loglinear model. Beyond the functional form, the results might be driven by
(neglected) time effects or by subgroups, or they might be sensitive to the
transformation of the data. Below, we show that a more flexible estimation
would not yield economically different results. A log-lin version of the same
model would actually yield a bigger effect, i.e., a stronger negative price
discrimination for cheaper cars and a stronger positive discrimination for
purchases of expensive cars. We also present sensitivity checks including
additional time controls, interactions, or restricting the data to 2009 and
the relevant price range. After that, we successively drop subsets of the
data which could be driving the results and show results for a variety of
quantile regressions. The results are qualitatively insensitive to any of these
robustness checks.
In a first step, we run a quadratic specification of the model as given in
Equation (5).
discount = α + βCC + γ1 M SRP + γ2 M SRP 2 +
(5)
δ1 CC ∗ M SRP + δ2 (CC ∗ M SRP )2 + θ0 X + Rather than comparing the coefficients of the model, we present the plot
of the linear and the quadratic model in Figure 4 based on the estimated
39
30
Discount in Percent of MSRP
20
25
15
10
20
30
40
MSRP in 1000 EUR
non-CC linear
non-CC quadratic
50
60
CC linear
CC quadratic
Note: Expected discount in percent as a function of MSRP. Functions are given over two models (linear
and quadratic) and two groups, subsidized (CC) and non-subsidized (non-CC) transactions. Parameters
are taken from the regression results above, specifications (2) (linear) and (3) (quadratic). The dashed
vertical lines show the upper borders of the first, the second, and the third quartiles of MSRP in 2009.
The solid vertical lines show the intersects between the CC and non-CC functions, i.e., the prices where
we observe just no difference in discount received between both buyer groups.
Figure 4: Linear and Quadratic Model for Year 2009
40
coefficients.41 We first plot the reference line, i.e., the discount in percent as
a function of MSRP for the two models. These are the lower dashed functions
in the graph, and one can see that they hardly differ. The upper functions
are the respective subsidized purchases where there is a little more difference
between the two models. The dashed vertical lines show the borders of the
first, the second, and the third quartile of MSRP in 2009. The two functions
diverge only from a price of roughly e 40,000 on, and are very close to each
other even for rather low prices of about e 10,000. The divergence in the
upper part is of little importance as the scrappage premium did not play
an important role in these price ranges.42 The vertical solid lines show the
intersects between the CC and non-CC functions, i.e., the point where there
is no difference in discount between a purchase with and without the subsidy.
One can see that in both cases the intersect is located within the second
quartile at a price of about e 18,000 and both functions differ little within
what we defined as the relevant price range in Section 3.5.
As the functional form appears adequate for the observed data, we run a
battery of tests on the linear model as given in Equation (3). Table 6 sums
up the tests we ran on our data sample. It reports the most relevant fig41
Regression results for the quadratic or even cubic formulation of the model are available
upon request.
42
There are two major reasons for this: first, the relative importance of the lump-sum
subsidy decreases as the car price increases. Second, as the subsidy could only be
requested when an old car was scrapped, the old car needed to be of a very low resale
value. In general, buyers of expensive cars benefited more from trading in their old car
than scrapping it for e 2,500.
41
ures: the number of observations (Obs.), the intersect, i.e., the point where
there is no difference in discount between a subsidized and a non-subsidized
car (Inters.), the coefficient for the scrappage premium (CC ), the coefficient
of the interaction term (CC*MSRP), and the percentage levels of price discrimination for a e 12,000, an e 18,000, and a e 32,000 car purchase (PD12,
PD18 and PD32 ).43
The first row represents the linear model based on the full data, i.e., the
reference results from Specification (2) in Table 4. In the second row, we
present an alternative specification of the model. Instead of regressing the
discount in percent, we regress the logarithm of the discount in Euro on the
model as discussed above.44 This specification is more flexible and less liable
to extreme values. However, as we observe the discount to be zero in some
two hundred cases, we lose these observations. We can see that the intersect
changes little (it is actually rather big at 19.63). The coefficients of CC
and the interaction term cannot be compared to the others due to the logtransformation, but they do show the expected sign and order of magnitude.
The observed price discrimination is even more pronounced than in the linear
model. We find −3.65% for an MSRP of e 12,000, −.69% for e 18,000 and
5.72% for e 32,000. This indicates that our preferred estimated results are a
lower bound, and hence a rather conservative measure, of the actual effect.
43
44
Full regression tables are available upon request.
For some variables which imply a level-effect in the discount in percent, such as the
control for demonstration cars or company employees, we add an interaction between
the MSRP and the respective dummy.
42
Table 6: Sensitivity Checks – Overview
Row
Case
Obs. Interc. CC
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
All Data
Log-Lin
Time 1 and 2
2009 Only
Y-M-Int.
Y-BD-Int.
Y-M-BD-Int.
Rel. Price Range
No CE
No DC
No CE, no DC
No VC J, M, S
No BD1
No BD2
No BD3
No BD4
No BD5
No BD6
No BD7
Qreg Q10
Qreg Q50
Qreg Q90
8156
7941
8156
3190
8156
8156
8156
5775
7188
6836
5868
6059
6692
5990
6657
7862
8010
6288
7437
8156
8156
8156
18.06
19.63
18.47
17.28
18.40
19.48
20.06
15.70
18.85
16.76
17.56
16.29
17.58
18.72
15.76
17.89
18.37
16.61
20.95
16.92
16.63
18.62
-4.40
-0.58
-4.73
-3.01
-4.71
-4.04
-4.16
-5.23
-4.77
-4.49
-4.96
-3.90
-3.92
-4.43
-3.29
-4.40
-4.52
-4.49
-5.74
-3.65
-4.10
-3.09
CC* PD12 PD18 PD32
MSRP
0.24
0.03
0.26
0.17
0.26
0.21
0.21
0.33
0.25
0.27
0.28
0.24
0.22
0.24
0.21
0.25
0.25
0.27
0.27
0.22
0.25
0.17
-1.48
-3.65
-1.66
-0.92
-1.64
-1.55
-1.67
-1.23
-1.73
-1.28
-1.57
-1.03
-1.24
-1.59
-0.78
-1.45
-1.57
-1.24
-2.45
-1.06
-1.14
-1.10
-0.01
-0.69
-0.12
0.13
-0.10
-0.31
-0.43
0.77
-0.21
0.33
0.12
0.41
0.09
-0.17
0.47
0.03
-0.09
0.38
-0.81
0.23
0.34
-0.10
3.40
5.72
3.47
2.56
3.48
2.59
2.48
5.44
3.33
4.09
4.08
3.77
3.21
3.15
3.38
3.47
3.35
4.16
3.02
3.25
3.79
2.22
Columns: Case indicates the particularity of each model, Obs. the number of observations, Inters. the
intersect where both models (for subsidized and non-subsidized buyers) intersect. CC is the coefficient for
Cash-for-Clunkers, CC*MSRP the interaction between CC and the manufacturer’s suggested retail price.
PD12, PD18, and PD32 give the price discrimination in percentage points for an MSRP of e 12,000,
e 18,000, and e 32,000 respectively.
Rows: All Data is the linear regression presented above as a benchmark for the sensitivity tests. Log-Lin is
a transformation of our preferred model in which the dependent variable now is the logarithm of the Euro
discount received. Time 1 and 2 includes two time controls for program changes (electronic application
procedure and budget increase). 2009 Only is a restriction to year 2009 only. Y-M-Int. presents results
including year-month interactions in order to control in a more flexible way for time-effects. Y-BD-Int.
presents a regression where all years and all brand-dealer dummies are interacted, Y-M-BD-Int. one where
years, months, and brand-dealer-dummies are interacted. Rel. Price Range restricts the sample to the
interval of e 12,000 to e 32,000. Number 9 to 19 present the exclusion of different groups: CE = company
employees and DC = demonstration cars are high discount receivers, VC J, M, S are the three vehicle
classes which are not part of a natural order, i.e., MPVs (M ), SUVs (J ), and Sports Coupés (S). BD1
to BD7 are different brand-dealer combinations. Number 20 to 22 show the results for different quantile
regressions over the percentiles 10 (Q10 ), 50 (Q50 ), and 90 (Q90 ).
43
For the ease of interpretation, we stick to the preferred model where we model
the dependent variable as the discount in percentage of the MSRP.
In row three, we add two time controls to the model: a dummy for the
switch from the paper-based to the electronic subsidy application procedure
(Time 1 ) and a time dummy for the expansion of the program from e 1.5
to e 5 Billion (Time 2 ). One can see that the inclusion of these further
controls does not alter the results significantly. In row four, we restrict our
sample to year 2009 only. While the price discrimination measured for an
MSRP of e 12,000 is slightly lower, the overall pattern remains the same.
This indicates that we do not simply observe a difference because 2009 was
very special, but that the hypothesis of price discrimination actually holds.
Rows five to seven present more flexible versions of the model. In row five,
we allow for different seasonal effects by interacting year and time dummies.
In row six, we interact brand-dealership combinations with year dummies in
order to allow for differences between the dealerships and over time. In row
seven, we interact all three sets of dummies, i.e., years, months, and branddealership combinations. This means we allow for different time effects over
the brand-dealership combinations and we allow these effects to be different
over the years. The coefficients and hence the measured price discrimination
do not change significantly.
As discussed in Section 3.5, the price range for which our results are
most reliable is the interval between e 12,000 and e 32,000 (MSRP). In row
eight, we therefore restrict the sample to this price range in order to make
44
sure our results are not driven by values far away from what we called the
relevant price range. The restriction makes the CC -function steeper, but
the intersect even smaller. So even though the intersect shifts leftward to
15.7, the observed price discrimination shows the same pattern and same
magnitude as our reference estimates in row one.
As already discussed, our results might be affected by composition effects stemming from unequal shares of high-discount groups. In addition to
correcting for different levels of discount (note that the coefficients are very
stable over all considered specifications), we exclude both groups from the
regression: first, company employees (No CE, row nine), then demonstration
cars (No DC, row ten) and, in row eleven, both. None of these exclusions
changes the results significantly.
In row twelve, we exclude the three vehicle classes that do not enter the
“natural” order. These are SUVs (J ), MPVs (M ) and Sports Coupés (S ).45
Again, the results stay unaffected.
Rows 13-19 show the results when leaving out single brand-dealer combinations. They illustrate that the results of our analysis are not driven by
a single group of those categories. The intersect is rather stable over the
different data restrictions: it may fall down to about e 16,000 (row 15), but
can also move up to roughly e 21,000 (row 19). On average however, it is
45
The other vehicle classes are cardinally ordered by size and, most important, price. The
three excluded vehicle classes are not part of this order and might be special for various
reasons.
45
located around e 17,000 to e 18,000, and therefore meets the dimension of
our full model.
Even though the dependent variable (discount in percent of MSRP) is
almost normally distributed, it is not perfectly so, and it is possible that
the observed effect is not only heterogeneous over MSRP, but also over the
dependent variable. Rows 20-22 therefore present the results from quantile
regressions over the 10th, 50th, and 90th percentile. We can see that the
results do not change either.46
Overall, Table 6 demonstrates that our findings are robust to an extensive variety of sensitivity checks. The log-linear specification shows that
our preferred estimates are rather conservative. Also, it is unlikely that
the presented estimation results are caused by the functional form, ignored
program influences, differences over time, high discount groups or special
vehicle classes, nor single dealers or brands. Quantile regressions confirm
the reported results. In addition to robust standard errors, we also ran
our preferred regression with standard errors clustered either by brand, by
brand-dealer combination, or by year.47 All relevant coefficients remain significant.48
46
Table 6 does not report significance levels, but all relevant coefficients, i.e., the coefficients for CC, M SRP and the interaction are significant at the 1% level.
47
We do this since treating each car sold as a completely independent observation could
tend to be too permissive.
48
In all cases, the coefficient for CC is still significant at the 5%-level and the interaction
term remains significant at the 1%-level. Only the coefficient for MSRP, which is not
crucial for our results, turns insignificant. .
46
We conclude from the robustness analysis that our results are not driven
by unobserved heterogeneity or misspecification. Overall, the preferred model
appears to be rather conservative. The model controls for most relevant influences, with the price of the car being the most important feature.
5. Conclusion
We evaluate the incidence of the German scrappage program from 2009,
the most expensive program of all countries in that time period with a total volume of e 5 billion ($7 billion). Overall, we find that the benefits—as
intended—are captured by the customers, with this effect being heterogeneous over price segments.
Applying linear regression models to a unique sample of micro transaction data, we model the percentage discount as a function of the MSRP, a
dummy for the scrapping premium, and various controls. In a first step, we
find that, on average, the subsidy is captured by the customers by slightly
more than 100%. We therefore allow for heterogeneity across price segments
when comparing subsidized to non-subsidized purchases and find that these
differ significantly. Subsidized buyers of the first quartile (cheap cars) faced
negative price discrimination, i.e., received less discount than non-subsidized
buyers, leaving them with incidence amounts smaller than 100%. Above the
median MSRP (up to expensive cars), the discount for subsidized buyers was
higher than the discount for non-subsidized ones, i.e., Cash-for-Clunkers customers faced positive price discrimination or incidence amounts of more than
47
100%. The absolute amount of price discrimination was substantially bigger
in the upper range of MSRP. Our results can be explained by an optimal
pricing strategy of the supply side which depends on market segments and
different price elasticities of demand.
With regard to the incidence, we showed that the supply-side captured a
small fraction of the subsidy, but only for cars from smaller vehicle classes.
For bigger, more expensive cars, the customers received additional discounts
in order to sign a contract of purchase. In summary, the car industry benefited by selling additional cars, and consumers benefited by being subsidized
by the government and the industry. However, not all consumers benefited
by the same amount and, as the program was tax-financed, most of those
who bore the costs did not benefit from it at all.
While our findings are very robust to changes in the model as well as
different data restrictions, they can only shed light on what happened in
the market of new cars. The demand shock for new cars obviously had
repercussions on the used car market and, as the maintenance of new cars is
usually done by manufacturer-related subcontractors, we also expect negative
impacts on repair shops which are not linked to a particular manufacturer.
Since we do not have data and sufficient information in these respects, we
cannot go into a thorough welfare analysis and therefore strictly separate
efficiency questions from our incidence focus.
48
Appendix
49
Table A1: Number of Purchases over Car Brands and Car Dealers
Car Dealer
Dealer
Dealer
Dealer
Dealer
Dealer
Dealer
Total
1
2
3
4
5
6
1
2
0
1464
0
0
0
0
1464
2166
0
1499
0
0
0
3665
Car Brand
3
4
0
0
0
0
294
0
294
50
0
0
0
0
0
146
146
5
6
Total
0
0
0
1868
0
0
1868
0
0
0
0
0
719
719
2166
1464
1499
1868
294
865
8156
Table A2: Number of Purchases over Month of all Years, 2009 by CC
Month of Purchase
1
2
3
4
5
6
7
8
9
10
11
12
Total
Year of Purchase and Clunker’s Premium
2007
2008
2009
2009
2010
NonCC
CC
103
85
162
154
147
142
118
116
140
109
134
144
1554
109
132
175
225
159
172
149
85
115
145
120
115
1701
Note: Non-CC are non-subsidized purchases, CC subsidized ones.
51
115
139
183
157
115
156
129
107
136
145
137
130
1649
25
93
206
200
221
198
192
150
102
100
41
13
1541
101
120
230
159
183
166
133
133
129
143
114
100
1711
Table A3: Summary Statistics over Vehicle Class
EU Vehicle Class
Catalog price in 1.000 EUR
Mean
SD
A - Mini Cars
B - Small Cars
C - Medium Cars
D - Large Cars
E - Executive Cars
F - Luxury Cars
J - SUV
M - MPV
S - Sports Coupés
Total
11.59
16.09
23.76
38.94
58.32
106.91
40.18
27.25
61.09
25.62
1.29
2.59
4.22
7.87
11.54
15.54
14.68
7.28
25.51
14.37
Note: SUV stands for Sport Utility Vehicle, MPV for Multi Purpose Vehicle.
52
N
649
2611
1649
899
232
19
481
1429
187
8156
Table A4: Summary Statistics: 2009 Only
Variables
Mean
SD
Med
Min
Max
N
Discount in Percent
Discount in 1000 EUR
MSRP in 1000 EUR
Clunker’s Premium (CC)
Demonstration Car (DC)
Company Employee (CE)
Female
Age at Purchase
17.11
3.99
23.02
0.48
0.14
0.13
0.32
46.79
7.82
3.20
12.76
0.50
0.35
0.33
0.46
15.09
16.75
3.25
19.25
0
0
0
0
48
0.00 45.61
0.00 51.81
8.69 130.35
0
1
0
1
0
1
0
1
18
89
3,190
3,190
3,190
3,190
3,190
3,190
3,190
639
Note: MSRP is the manufacturer suggested retail price. CC is a dummy variable indicating whether the
buyer of a car received the scrappage subsidy. DC is a dummy variable indicating whether a buyer bought
a demonstration car. CE is a dummy variable indicating whether the buyer was an employee of a car
manufacturing company. Female is a dummy of female buyers, the summary statistics therefore report
the share of women, age at purchase is the age of the buyer at the time of purchase.
53
Table A5: Percentage Discount over Time by CC
Year of Purchase
Discount in Percent
Non-CC
CC
Total
MeanSD
N MeanSD
N MeanSD
2007
2008
2009
2010
Total
15.69
16.54
17.67
18.03
17.01
9.01
9.64
8.73
8.74
9.09
1554
0 15.69 9.01
1701
0 16.54 9.64
1649 16.51 6.67 1541 17.11 7.82
1711
0 18.03 8.74
6615 16.51 6.67 1541 16.91 8.68
N
1554
1701
3190
1711
8156
Note: Non-CC are non-subsidized purchases, CC subsidized ones. The last three columns of the table
give summary statistics for the whole group.
54
Table A6: Percentage Discount by Type of Purchase
Type of Purchase
Standard
Company Employee
Demonstration Car
Total
Discount in Percent
Mean
SD
14.02
25.84
23.25
16.91
7.44
4.03
8.47
8.68
N
5868
968
1320
8156
Note: Standard are the benchmark purchases, Company Employees are employees of a car manufacturing
company, Demonstration Car denotes cars that are not new but have been used for exhibition and have
been licensed for a maximum of 14 months before the purchase.
55
S - Sports Coupés
M - MPV
J - SUV
F - Luxury Cars
E - Executive Cars
D - Large Cars
C - Medium Cars
B - Small Cars
1000000
A - Mini Cars
500000
0
Number of New Registrations
Year of Registration
2008
2009
2010
Note: A, B, C, D, E, F, J, M, S are auto segments according to the EU car classification. SUV stands
for Sport Utility Vehicle, MPV for Multi Purpose Vehicle.
Figure A1: All New Car Registrations in Germany over Time by Vehicle Class
56
A - Mini Cars
MSRP in 1000 EUR
12.5
12
11.5
11
10.5
B - Small Cars
18
17
16
15
D - Large Cars
C - Medium Cars
26
25
24
23
22
E - Executive Cars
F - Luxury Cars
45
80
130
40
70
120
35
60
110
30
100
50
J - SUV
M - MPV
60
30
50
25
40
20
30
2007q1 2008q1 2009q1 2010q1 2011q1
S - Sports Coupés
90
80
70
60
50
2007q1 2008q1 2009q1 2010q1 2011q1
2007q1 2008q1 2009q1 2010q1 2011q1
Quarter of Purchase
Non-CC transactions
CC transactions
Note: MSRP over quarters of years across EU vehicle classes. SUV stands for Sport Utility Vehicle, MPV
for Multi Purpose Vehicle. The graphs show mean values rounded by quarter.
Figure A2: MSRP over Time by EU Vehicle Class
57
References
Adda, J. and R. Cooper, “Balladurette and Juppette: A Discrete Analysis
of Scrapping Subsidies,” Journal of Political Economy, August 2000, 108
(4), 778–806.
Anderson, Simon P., Andre de Palma, and Brent Kreider, “Tax Incidence in Differentiated Product Oligopoly.,” Journal of Public Economics,
2001, 81 (2), 173 – 192.
Berry, Steven, James Levinsohn, and Ariel Pakes, “Automobile Prices
in Market Equilibrium,” Econometrica, 1995, 63 (4), 841–890.
Böckers, Veit, Ulrich Heimeshoff, and Andrea Müller, “Pull-forward
effects in the German car scrappage scheme: A time series approach,”
DICE Discussion Papers 56, Heinrich-Heine-Universität Düsseldorf, Düsseldorf Institute for Competition Economics (DICE) 2012.
Busse, Meghan, Jorge Silva-Risso, and Florian Zettelmeyer, “$1,000
Cash Back: The Pass-Through of Auto Manufacture Promotions,” The
American Economic Review, September 2006, 96 (4), 1253–1253.
Busse, Meghan R., Christopher R. Knittel, Jorge Silva-Risso, and
Florian Zettelmeyer, “Did ’Cash for Clunkers’ Deliver? The Consumer
Effects of the Car Allowance Rebate System,” November 2012. Working
Paper.
58
Chetty, Raj, Adam Looney, and Kory Kroft, “Salience and Taxation:
Theory and Evidence,” The American Economic Review, 2009, 99 (4),
1145–1177.
Delipalla, Sophia and Owen O’Donnell, “Estimating Tax Incidence,
Market Power and Market Conduct: The European Cigarette Industry.,”
International Journal of Industrial Organization, 2001, 19 (6), 885 – 908.
Deysher, Beth and Don Pickrell, “Emissions Reductions From Vehicle
Retirement Programs,” Transportation Research Record: Journal of the
Transportation Research Board, 1997, 1587, 121–127.
Evans, William N., Jeanne S. Ringel, and Diana Stech, “Tobacco
Taxes and Public Policy to Discourage Smoking,” in “Tax Policy and the
Economy,” Vol. 13 of NBER Chapters, National Bureau of Economic Research, Inc, 1999, pp. 1–56.
Friedman, John, “The Incidence of the Medicare Prescription Drug Benefit:
Using Asset Prices to Assess its Impact on Drug Makers,” 2009. Working
Paper, Harvard Kennedy School, Harvard University.
Fullerton, Don and Gilbert E. Metcalf, “Tax Incidence,” in Alan J.
Auerbach and Martin Feldstein, eds., Handbook of Public Economics, Vol. 4
of Handbook of Public Economics, Elsevier, 2002, pp. 1787–1872.
59
Goldberg, Pinelopi Koujianou, “Product Differentiation and Oligopoly
in International Markets: The Case of the U.S. Automobile Industry,”
Econometrica, July 1995, 63 (4), 891–951.
Goldberg, Pinelopi Koujianou, “Dealer Price Discrimination in New Car
Purchases: Evidence from the Consumer Expenditure Survey,” Journal of
Political Economy, June 1996, 104 (3), 622–654.
Hahn, Robert W., “An Economic Analysis of Scrappage,” The RAND
Journal of Economics, 1995, 26 (2), 222–242.
Hastings, Justine and Ebonya Washington, “The First of the Month
Effect: Consumer Behavior and Store Responses,” American Economic
Journal: Economic Policy, May 2010, 2 (2), 142–162.
Heimeshoff, Ulrich and Andrea Müller, “Evaluating the Causal Effects
of Cash-for-Clunkers Programs in Selected Countries: Success or Failure?,”
2011. Working Paper.
Kavalec, Chris and Winardi Setiawan, “An analysis of accelerated vehicle retirement programs using a discrete choice personal vehicle model,”
Transport Policy, 1997, 4 (2), 95–107.
Knittel, Christopher R., “The Implied Cost of Carbon Dioxide Under
the Cash for Clunkers Program,” 2009. Working Paper, Sloan School of
Management, Massachusetts Institute of Technology.
60
Kotlikoff, Laurence J. and Lawrence H. Summers, “Tax Incidence,”
in Alan J. Auerbach and Martin Feldstein, eds., Handbook of Public Economics, Vol. 2 of Handbook of Public Economics, Elsevier, 1987, pp. 1043–
1092.
Li, Shanjun, Joshua Linn, and Elisheba Spiller, “Evaluating ’Cash-forClunkers’: Program Effects on Auto Sales and the Environment.,” Journal
of Environmental Economics and Management, 2013, 65 (2), 175 – 193.
Licandro, Omar and Antonio R. Sampayo, “The Effects of Replacement
Schemes on Car Sales: The Spanish Case,” Investigaciones Economicas,
May 2006, 30 (2), 239–282.
Marion, Justin and Erich Muehlegger, “Fuel tax incidence and supply
conditions,” Journal of Public Economics, 2011, 95, 1202–1212.
Mian, Atif and Amir Sufi, “The Effects of Fiscal Stimulus: Evidence from
the 2009 Cash for Clunkers Program.,” Quarterly Journal of Economics,
2012, 127 (3), 1107 – 1142.
Rothstein, Jesse, “Is the EITC as Good as an NIT? Conditional Cash
Transfers and Tax Incidence,” American Economic Journal: Economic
Policy, 2010, 2 (1), 177–208.
Ruffle, Bradley J., “Tax and subsidy incidence equivalence theories: experimental evidence from competitive markets,” Journal of Public Economics,
2005, 89, 1519–1542.
61
Sallee, James M., “The Surprising Incidence of Tax Credits for the Toyota
Prius,” American Economic Journal: Economic Policy, May 2011, 3 (2),
189–219.
Stern, Nicholas H., “The Effects of Taxation, Price Control and Government Contracts in Oligopoly and Monopolistic Competition.,” Journal of
Public Economics, 1987, 32 (2), 133 – 158.
Sudhir, K., “Competitive Pricing Behavior in the Auto Market: A Structural Analysis,” Marketing Science, 2001, 20 (1), 42–60.
Szwarcfiter, Lila, Francisco Eduardo Mendes, and Emilio Lebre La
Rovere, “Enhancing the Effects of the Brazilian Program to Reduce Atmospheric Pollutant Emissions from Vehicles,” Transportation Research:
Part D: Transport and Environment, 2005, 10 (2), 153–160.
Verboven, Frank, “Quality-Based Price Discrimination and Tax Incidence:
Evidence from Gasoline and Diesel Cars,” The RAND Journal of Economics, 2002, 33 (2), 275–297.
62