Skip to main content

The incidence of Cash for Clunkers: Evidence from the 2009 car scrappage scheme in Germany

Abstract

This paper investigates the German car scrappage program, focusing on the incidence of the premium. We ask how much of the €2500 ($3500) buyer subsidy is actually captured by the demand side. More precisely, we analyze the program’s impact on different car segments, allowing for heterogeneity in incidence at different points in the vehicle price distribution. Using a unique microtransaction data set, we find that the incidence of the subsidy strongly and significantly varies across price segments. Subsidized buyers of cheap cars paid a little more than comparable buyers who did not receive the subsidy, indicating incidence amounts slightly below 100 %. For more expensive vehicles, subsidized buyers were granted large extra discounts on top of the government premium, translating into incidence amounts considerably greater than 100 %. Taken together, this results in an aggregate incidence amount of up to plus €350 million, suggesting that the positive effect for expensive cars overcompensates the negative effect for small cars.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3

Notes

  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. In contrast, the USA designed its subsidy as two-staged. Depending on the type of car purchased and the difference in fuel economy between the scrapped and the new car, the amount of the subsidy was either $3500 or $4500. The lump-sum subsidy in Germany, therefore, offers a unique and “clean” opportunity to evaluate the incidence of one of the most important scrappage programs worldwide.

  3. This so-called tax equivalence theorem is a basic fundamental within the incidence context. Ruffle (2005) for instance, shows that this theorem holds empirically. However, other research (e.g., Busse et al. 2006; Chetty et al. 2009; Sallee 2011) implies that, contrary to standard theories of incidence, the statutory incidence of a policy does affect the economic incidence.

  4. This intervention subsidized the replacement of household appliances. Unlike the US Cash for Clunkers program, it was administered by the individual states. Moreover, Cash for Appliances required the replacement products to be Energy Star certified, thereby generating a natural control group (other appliances).

  5. For instance, see Adda and Cooper (2000), Licandro and Sampayo (2006), Mian and Sufi (2012), Li et al. (2013), and Copeland and Kahn (2013) 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 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.

  6. Bundesamt für Wirtschaft und Ausfuhrkontrolle (Federal Office of Economics and Export Control).

  7. 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.

  8. For data privacy reasons, we cannot report the name of any car dealership or brand. Table 8 in the Appendix gives a summary of the distribution.

  9. 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 consequently did not affect the reported prices. Moreover, the value of the subsidy is not included in the purchase price of the vehicle and, hence, does not affect reported prices either.

  10. Note that empirical evidence for Germany suggests that things were quite different from the US scrappage program of which we know that it did not lead to long-lasting impacts on new vehicle sales. Mian and Sufi (2012), e.g., state that the U.S. program increased the number of vehicles purchased during its 2-months period but that this effect was completely reversed over the following 10 months. This means that, in the USA, 100 % of the program-induced purchases were pulled forward from future periods. Contrarily for Germany, Klößner and Pfeifer (2015) find that only about 30 % were pulled forward leaving almost one million newly registered cars as program-induced on-top sales, which would not have happened at all in the absence of the policy intervention.

  11. Figure 4 in the Appendix shows new car registrations for all cars in Germany over the years 2008–2010.

  12. 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.

  13. Summary statistics for the MSRP over vehicle classes are given in Table 9 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.

  14. Table 11 in the Appendix gives an overview of the development of the percentage discount over the years including a CC/non-CC distinction.

  15. The distribution of the MSRP of subsidized cars is concentrated among lower prices. Its median is €17,000, and the 75th percentile is at about €22,000.

  16. So, it is

    $$\begin{aligned} \hbox {discount} = 100 \times \frac{\hbox {MSRP}-\hbox {Selling}\,\hbox {Price}}{\text {MSRP}} \end{aligned}$$

    with the selling price including the subsidy amount. We prefer percentage discount over the level for several reasons. To begin with, customers rather think about and make their purchase decision based on the percentage discount instead of an absolute discount value. Moreover, the distribution of absolute monetary values is pretty skewed and residuals are non-normally distributed and highly heteroscedastic, which might lead to problems w.r.t. estimation and inference. However, as suggested by a referee, we present corresponding findings using the Euro level of discount as the dependent variable as additional results when providing essential estimation figures in Table 4, with the advantage that coefficients can be directly interpreted.

  17. Manufacturers 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 unpredictable for buyers as well as 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.

  18. Column (3) presents corresponding results using the Euro level of discount as the dependent variable.

  19. Similar to Busse et al. (2006) who identify the respective 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.

  20. To name just a 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. Busse et al. (2006) and Sallee (2011), though, find that 70–90 and 100 % of the subsidy remain with the customer, respectively. Specification (3) in Table 4 confirms a 100 % pass-through to the customer with a \(\beta \)-coefficient of zero when using the Euro level discount as the dependent variable.

  21. As discussed above, we should not simply interact CC with a set of vehicle class dummies because within each such class, the two groups (subsidized and non-subsidized purchases) differ. However, when conducting sensitivity checks in Sect. 5, we look at a regression considering an interaction of CC with vehicle class while controlling for MSRP.

  22. Column (4) presents corresponding results using the level discount as dependent variable. Here, \(\delta \) provides a direct test of whether customers purchasing higher-priced vehicles capture a greater fraction of the rebate than non-subsidized customers.

  23. 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.

  24. Clustered standard errors would not change the significance levels severely. Table 12 in the Appendix shows our preferred model with standard errors clustered either by brand, dealership, the interaction of brand and dealership, vehicle class, or year. None of these clusters makes our two coefficients of interest (\(\beta \) for CC and \(\delta \) for the interaction term) insignificant.

  25. This result is confirmed when looking at Specification (4) which uses the Euro level discount as the dependent variable. Moreover, our finding is not sensitive to the inclusion or exclusion of controls. Table 13 in the Appendix shows the model (allowing for heterogeneity) built up step by step. While the inclusion of additional controls improves the model fit considerably, it does not affect the main coefficients of interest (\(\beta \) for CC and \(\delta \) for the interaction term) very much. Particularly, the coefficient of the interaction term is very robust to these alterations. Note that Specification (5) of Table 13 in the Appendix corresponds to Specification (2) of Table 4.

  26. For a discussion on high inventories within the expensive car market see http://www.wsj.com/articles/SB10001424052748703296604576005533229837672, last accessed on July 6, 2015.

  27. The reported coefficient \(\beta \) on the CC dummy from Specification (1) of Table 4 can be interpreted as a weighted average.

  28. Consistent with Table 4, Columns (1)–(3) are based on the discount in percent as dependent variable whereas Columns (4) and (5) are based on the Euro level discount as dependent variable.

  29. This difference might be driven by the fact that the distribution of the variables is different and the linear model does not fit the data very well in the lower part of the distribution. Given that the overall pattern is the same, this seems to be unproblematic.

  30. Moreover, unfortunately, it is not possible to link the 2 million subsidized cars to their corresponding trade-ins, i.e., to the vehicles scrapped.

  31. http://www.schwackepro.de/produkte-and-services/fahrzeuge-spezifizieren-und-identifizieren/carconfigurator/, last accessed on 2015-03-23. As prices vary substantially within each model, depending on size, engine, turnover etc., all prices per model (e.g., the Volkswagen Golf) were collected and, lacking information on the distribution of the exact versions in the market, the price was calculated as the unweighted average of the given catalog prices within each model. If the BAFA data grouped similar models, the average retail price was simply the mean of the retail prices calculated for each make-model combination. In case no price was available for a certain model, the catalog price was looked up on the website of the producer.

  32. The total number of subsidized vehicles was slightly smaller than 2 million because the total budget of €5 billion included administrative expenses which could not be spent as subsidies.

  33. Interestingly, Gneezy et al. (2012), in one of their randomized experiments, find quite a parallelism. Instead of comparing subsidized and non-subsidized buyers, they compare Caucasian and African American car buyers. For cheap cars, they do not find discernable differences, e.g., in the average final price quote. In contrast, for high-end cars, they do find significant discrimination between those groups.

  34. We also consider the lower part of the fourth quartile of MSRP since we argue that our relevant price range reaches €32,000.

  35. Similarly, Aldy et al. (2013) find that consumers captured the full incidence of the “Cash for Appliances” program.

  36. Recall that we already discussed that this is possible indeed, both in theory (Stern 1987 or Anderson et al. 2001) and in practical applications (Delipalla and O’Donnell 2001).

  37. Since facing a flat subsidy, buyers should not be willing to trade in a “clunker” worth more than €2500 and have to accept diminishing benefits from purchasing more expensive cars.

  38. Also compare Goldberg (1995) regarding brand loyalty in the car market.

  39. Moreover, the distribution of segments is quite similar between cars bought with and without the scrappage subsidy. See, for example, Fig. 1.

  40. Remember, that 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 €2500.

  41. The mean values are those covered by the relevant price range, that is, €12,000 through €32,000.

  42. Note that we observe quite similar patterns if we interact CC and vehicle class while controlling for MSRP where the latter should correct for the fact that there are observable differences between CC and non-CC cars within vehicle classes. Figure 8 in the Appendix shows the corresponding plot of discount over vehicle classes, restricting the sample to vehicle classes which can be sorted by increasing prices.

  43. The confidence intervals confirm that price discrimination is always negative for cars of an MSRP of €12,000 and always positive for cars of an MSRP of €32,000.

  44. 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.

  45. In Sect. 4.2, we explained why we prefer percentage discount over corresponding level values.

  46. Certainly, we had to make sure to include a valid portion of subsidized purchases which is why we focus on an entire quarter instead of, e.g., just the first month of the intervention (where we could think of comparing similar purchases around the program-start-cutoff).

  47. 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.

  48. The same holds if we leave out single brand-dealer combinations.

  49. In all cases, the coefficient of CC is still significant at the 5 % level and the coefficient of the interaction term remains significant at the 1 %-level. Only the coefficient of MSRP, which is not crucial for our results, might turn insignificant. Recall that even clustered standard errors (on either brand, dealership, the interaction of brand and dealership, vehicle class, or year) do not mitigate the significance of our estimates.

References

  • Adda, J., & Cooper, R. (2000). Balladurette and Juppette: A discrete analysis of scrapping subsidies. Journal of Political Economy, 108(4), 778–806.

    Article  Google Scholar 

  • Aldy, J., Houde, S., & Nazar, H. (2013). Did the stimulus benefit the environment? New evidence from cash for appliances. Working paper.

  • Anderson, S. P., de Palma, A., & Kreider, B. (2001). Tax incidence in differentiated product oligopoly. Journal of Public Economics, 81(2), 173–192.

    Article  Google Scholar 

  • BAFA. (2010). Abschlussbericht–Umweltprämie. Accessed on March 24, 2015 from http://www.bafa.de/bafa/de/wirtschaftsfoerderung/umweltpraemie/publikationen/ump_abschlussbericht.pdf.

  • Busse, M., Silva-Risso, J., & Zettelmeyer, F. (2006). \({\$}1,000\) Cash back: The pass-through of auto manufacture promotions. The American Economic Review, 96(4), 1253–1253.

    Article  Google Scholar 

  • Busse, M. R., Knittel, C. R., Silva-Risso, J., & Zettelmeyer, F. (2012). Did ’Cash for Clunkers’ deliver? The consumer effects of the car allowance rebate system. Working paper.

  • Chetty, R., Looney, A., & Kroft, K. (2009). Salience and taxation: Theory and evidence. The American Economic Review, 99(4), 1145–1177.

    Article  Google Scholar 

  • Copeland, A., & Kahn, J. (2013). The production impact of ’Cash-for-Clunkers’: Implications for stabilization policy. Economic Inquiry, 51(1), 288–303.

    Article  Google Scholar 

  • Delipalla, S., & O’Donnell, O. (2001). Estimating tax incidence, market power and market conduct: The European cigarette industry. International Journal of Industrial Organization, 19(6), 885–908.

    Article  Google Scholar 

  • Deysher, B., & Pickrell, D. (1997). Emissions reductions from vehicle retirement programs. Transportation Research Record: Journal of the Transportation Research Board, 1587, 121–127.

    Article  Google Scholar 

  • Evans, W. N., Ringel, J. S., & Stech, D. (1999). Tobacco taxes and public policy to discourage smoking. In Tax policy and the economy, Vol. 13 of NBER chapters (pp. 1–56). National Bureau of Economic Research, Inc.

  • Friedman, J. (2009). The incidence of the medicare prescription drug benefit: Using asset prices to assess its impact on drug makers. Working paper, Harvard Kennedy School, Harvard University.

  • Fullerton, D., & Metcalf, G. E. (2002). Tax incidence. In A. J. Auerbach & M. Feldstein (Eds.), Handbook of public economics (Vol. 4, pp. 1787–1872). Amsterdam: Elsevier.

    Google Scholar 

  • Gneezy, U., List, J., & Price, M. K. (2012). Toward an understanding of why people discriminate: Evidence from a series of natural field experiments. Working paper 17855, National Bureau of Economic Research.

  • Goldberg, P. K. (1995). Product differentiation and oligopoly in international markets: The case of the U.S. automobile industry. Econometrica, 63(4), 891–951.

    Article  Google Scholar 

  • Hahn, R. W. (1995). An economic analysis of scrappage. The RAND Journal of Economics, 26(2), 222–242.

    Article  Google Scholar 

  • Hastings, J., & Washington, E. (2010). The first of the month effect: Consumer behavior and store responses. American Economic Journal: Economic Policy, 2(2), 142–162.

    Google Scholar 

  • Kavalec, C., & Setiawan, W. (1997). An analysis of accelerated vehicle retirement programs using a discrete choice personal vehicle model. Transport Policy, 4(2), 95–107.

    Article  Google Scholar 

  • Klößner, S., & Pfeifer, G. (2015). Synthesizing Cash for Clunkers: Stabilizing the car market, hurting the environment. Working paper.

  • Knittel, C. R. (2009). The implied cost of carbon dioxide under the Cash for Clunkers program. Working Paper, Sloan School of Management, Massachusetts Institute of Technology.

  • Kotlikoff, L. J., & Summers, L. H. (1987). Tax incidence. In A. J. Auerbach & M. Feldstein (Eds.), Handbook of public economics (Vol. 2, pp. 1043–1092). Amsterdam: Elsevier.

    Google Scholar 

  • Li, S., Linn, J., & Spiller, E. (2013). Evaluating ’Cash-for-Clunkers’: Program effects on auto sales and the environment. Journal of Environmental Economics and Management, 65(2), 175–193.

    Article  Google Scholar 

  • Licandro, O., & Sampayo, A. R. (2006). The effects of replacement schemes on car sales: The Spanish case. Investigaciones Economicas, 30(2), 239–282.

    Google Scholar 

  • Marion, J., & Muehlegger, E. (2011). Fuel tax incidence and supply conditions. Journal of Public Economics, 95, 1202–1212.

    Article  Google Scholar 

  • Mian, A., & Sufi, A. (2012). The effects of fiscal stimulus: Evidence from the 2009 Cash for Clunkers program. Quarterly Journal of Economics, 127(3), 1107–1142.

    Article  Google Scholar 

  • Rothstein, J. (2010). Is the EITC as good as an NIT? Conditional cash transfers and tax incidence. American Economic Journal: Economic Policy, 2(1), 177–208.

    Google Scholar 

  • Ruffle, B. J. (2005). Tax and subsidy incidence equivalence theories: Experimental evidence from competitive markets. Journal of Public Economics, 89, 1519–1542.

    Article  Google Scholar 

  • Sallee, J. M. (2011). The surprising incidence of tax credits for the Toyota Prius. American Economic Journal: Economic Policy, 3(2), 189–219.

    Google Scholar 

  • Stern, N. H. (1987). The effects of taxation, price control and government contracts in oligopoly and monopolistic competition. Journal of Public Economics, 32(2), 133–158.

    Article  Google Scholar 

  • Sudhir, K. (2001). Competitive pricing behavior in the auto market: A structural analysis. Marketing Science, 20(1), 42–60.

    Article  Google Scholar 

  • Szwarcfiter, L., Mendes, F. E., & La Rovere, E. L. (2005). Enhancing the effects of the Brazilian program to reduce atmospheric pollutant emissions from vehicles. Transportation Research: Part D: Transport and Environment, 10(2), 153–160.

    Article  Google Scholar 

  • Verboven, F. (2002). Quality-based price discrimination and tax incidence: Evidence from gasoline and diesel cars. The RAND Journal of Economics, 33(2), 275–297.

    Article  Google Scholar 

Download references

Acknowledgments

The authors are grateful for valuable remarks from the Editor, Ron Davies, and two anonymous referees. Moreover, we thank Martin Becker, Nadja Dwenger, Marc Escrihuela Villar, Paul Grieco, Rainer Haselmann, Stefan Klößner, Dieter Schmidtchen, and Michael Wolf. Also, we are thankful for inputs from the participants of the EEA Conference, the IIOC, the Applied Micro Seminar at Penn State, the ACDD Conference, the Econometric Society Australasian Meeting, the Warsaw International Economic Meeting, the Annual Meeting of the German Economic Association, and the IIPF Conference.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Ashok Kaul.

Appendix

Appendix

See Tables 8, 9, 10, 11, 12 and 13 and Figs. 4, 5, 6, 7 and 8.

Table 8 Number of purchases over car brands and car dealers
Table 9 Summary statistics for MSRP by EU vehicle class
Table 10 Summary statistics: 2009 only
Table 11 Percentage discount over time by CC
Table 12 Linear regression estimation results with clustered standard errors
Table 13 Linear regression estimation results step by step
Fig. 4
figure 4

All new car registrations in Germany over time by EU vehicle class. 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

Fig. 5
figure 5

MSRP over time by EU vehicle class. 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

Fig. 6
figure 6

Linear and quadratic model for year 2009. 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, Specification (2) (linear) and an analogous quadratic specification. 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 intersections between the CC and non-CC functions, i.e., the prices where we observe just no difference in discount received between both buyer groups

Fig. 7
figure 7

Flexible model for year 2009. Expected discount in percent as a function of MSRP. Functions, including 90 % confidence bands, are given for a model including CC*MSRP-interactions considering several MSRP intervals and two groups, subsidized (CC) and non-subsidized (non-CC) transactions. Parameters are taken from regression results, analogous to the linear case, for which thick lines are added as a reference (see Panel a) of Fig. 3)

Fig. 8
figure 8

Linear model with EU vehicle class interaction for year 2009. Expected discount in percent as a function of vehicle class. Functions, including 90 % confidence bands, are given for a model including CC*VC-interactions while controlling for MSRP and considering two groups, subsidized (CC) and non-subsidized (non-CC) transactions. Parameters are taken from regression results, analogous to the cases in Sect. 5.2

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Kaul, A., Pfeifer, G. & Witte, S. The incidence of Cash for Clunkers: Evidence from the 2009 car scrappage scheme in Germany. Int Tax Public Finance 23, 1093–1125 (2016). https://doi.org/10.1007/s10797-016-9396-1

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10797-016-9396-1

Keywords

  • Cash for Clunkers
  • Scrappage scheme
  • Incidence
  • Subsidy
  • Pricing

JEL Classification

  • H22
  • D12
  • L62