Robin Stitzing, Aalto University School of Business 1. Abstract

Transcription

1 3 CO 2 Differentiation of Automobile Sales Tax Rates and Emissions Rates of New Cars Robin Stitzing, Aalto University School of Business 1 Abstract What explains the observed decline of CO 2 emissions rates of new cars in Europe? The European Union introduced mandatory CO 2 emissions standards to manufacturers, accompanied by various domestic fiscal policies targeting consumers. I quantify the effect of drastic domestic tax policy, the 2008 Finnish CO 2 differentiation of ad valorem tax rates, on this key statistic to policymakers. My flexible reduced-form approach uses individual-vehicle registration data, and avoids explicit modeling of economic primitives. Results show that domestic tax policy only accounts for a small part of the large observed decline CO 2 emissions rates of new cars in Finland. An accompanying fuel excise tax hike had almost no effect. Exogenous technological change, likely due to the concurrent introduction of mandatory EU emissions standards to manufacturers, implies that consumers would have largely bought cleaner cars regardless of domestic fiscal policy. JEL Classification: H23, L62, Q53 Keywords: Automobile Taxation, Tax Reform, Environmental Taxation, Emissions-Based Taxation 1 I thank Matti Liski, Janne Tukiainen, Matti Sarvimäki, Kristiina Huttunen, Otto Toivanen, Roope Uusitalo, Marko Terviö, Tuomas Kosonen, Kaisa Kotakorpi, Tuomas Laiho, and Timo Goeschl as well as seminar participants at HECER and VATT for helpful comments. All remaining errors are my own. 67

2 3.1 Introduction Finland taxes the value of passenger cars at first registration in the country with the so-called car tax. The ad valorem car tax is a de facto sales tax on new passenger cars, and is levied in addition to the general value-added tax (VAT). The total tax rate of both taxes combined on tax-free price equaled around 77 percent for the average new car sold in the mid-2000s. Policymakers announced the differentiation of car-tax rates by vehicle-specific average carbon dioxide (CO 2 ) emissions in November 2007, and implemented this environmentally motivated fiscal policy in January Post-reform total tax rates range from 39 to 138 percent. The tax amount thus became an increasing function of both vehicle value and pollution potential. The declared goal of the heterogeneous tax treatment was to incentivize consumers to buy cleaner cars with lower CO 2 emissions. Registration-weighted CO 2 emissions rates of new cars declined quickly from g/km in 2007 to g/km in However, the strong negative trend of new-car average CO 2 emissions cannot be directly attributed to the CO 2 differentiation of car tax rates. The car tax reform coincided with the introduction of mandatory EU CO 2 emissions standards to manufacturers. These supply-side standards, first proposed in 2007 and becoming European law in 2009, essentially forced the industry to produce cleaners cars. Individual EU member countries introduced supplementary demand-side fiscal incentives for consumers to buy cleaner cars. The Finnish CO 2 differentiation of ad valorem tax rates was one of the most drastic demand-side policies, with similar reforms of existing ad valorem taxes also in Ireland and the Netherlands. Disentangling the effect of domestic fiscal policy on CO 2 emissions rates from the supply-side standards, fuel price fluctuations, and general market developments is non-trivial but important for a comprehensive evaluation of the policy. In this study, I quantify this effect of the Finnish car tax reform on CO 2 emissions rates of new passenger cars using detailed monthly registration data from 2006 to I Infer this effect by estimating how the CO 2 differentiation of tax rates shifted the equilibrium relationship between vehicle-specific CO 2 emissions rates and new registrations. The differencesin-differences approach to this equilibrium relationship with CO 2 emissions rates as continuous treatment controls for vehicle characteristics, consumer characteristics, and fuel price developments. The estimated shift is sufficient 68

3 to estimate counterfactual registration-weighted CO 2 emissions rates that would have occurred in the absence of the 2008 car tax reform. The reducedform approach of this study has two advantages over a structural approach with explicit modeling of economic primitives. First, I avoid complex yet arbitrary assumptions about consumer preferences and equilibrium price determination. Second, the estimation does not require price data that are potentially mismeasured due to common but hardly observable bargaining. My results show that domestic fiscal policy had at noticeable but small effect on CO 2 emissions rates of new passenger cars sold in Finland between 2008 and Consumers would have predominantly bought cleaner cars regardless of the January 2008 reform of the car tax CO 2 emissions rates would have been 2.5 g/km (standard error 0.9 g/km) higher, had the car tax remained non-differentiated. While noticeable, the economic importance of this effect is limited given the drastic tax reform and the observed decline of CO 2 emissions rates by 27.8 g/km from 2007 to The negative effect of the car tax reform on CO 2 emissions rates was initially slightly stronger but quickly became weaker as marketed car models became cleaner. This trend, likely driven by the EU supply-side standards, is exogenous to Finland due to the unimportance of the Finnish car market. 2 The unimportance of the market to manufacturers also implies that the estimated effect of the car tax reform on CO 2 emissions rates is a long-run effect. I also find evidence of a positive announcement effect of the car tax reform on CO 2 emissions rates during its two-month announcement period. Ciccone (2015) finds a similar announcement effect in her study of the 2007 CO 2 differentiation of the Norwegian unit tax on new-car registrations. A fuel excise tax hike concurrent to the January 2008 car tax reform had a negligible effect on CO 2 rates. Registration-weighted average CO 2 of new cars in Finland are insensitive to even substantial fuel price changes. Klier and Linn (2013) report similar findings for other Western European car markets. Gerlagh et al. (2015) use cross-country regressions to estimate the effect of the increased CO 2 sensitivity of automobile taxes on emissions rates of new cars. Their result of a minimal impact of differentiated registration taxes confirm the finding of this study. Klier and Linn (2015) study the impact of various CO 2 - related automobile tax reforms on CO 2 emissions rates in France, Germany, and Sweden where the authors also find small effects in the 2 Registrations of new cars in Finland account for less than 0.8 percent of all new-cars registrations in Europe according to industry statistics by the ACEA (2010). 69

4 latter two countries. Another study by Reynaert (2015) hints at the limited effect of demand-side fiscal policy on CO 2 emissions rates of new cars in light of the concurrent introduction of EU supply-side standards. He estimates a structural model of major European car markets to attribute up to 100 percent of the observed decline of CO 2 emissions rates to this supply-side policy. In Stitzing (2016c), I estimate a structural demand and supply model of the Finnish car market to evaluate the distributional welfare effects of the 2008 car tax reform. The structural analysis assumes static consumer preferences, aggregates cars into fewer consumer choices, and uses yearly instead of monthly data. Compared to the estimated effect of this study, the structural model leads an even slightly smaller estimate of the car tax reform s effect on CO 2 emissions. One implication of the findings of these studies and the related literature is that neither domestic fiscal policy or a green evolution of consumer preferences in Finland are the driving forces behind the decline of new-car CO 2 emissions rates but exogenous technological change, likely induced by mandatory EU supply-side standards. 3.2 The Finnish Car Tax Reform The Finnish car tax (autovero) is levied on all passenger cars upon first registration in the country. The de-facto sales tax on new cars is levied in addition to, and is itself subject to the general value-added tax (VAT). It is included in published list prices, and usually paid directly by the dealer. Before the car tax reform of January 2008, the car tax was defined as a 28 percentage share of the tax-inclusive retail value, minus a common discount of 650 euros (450 euros for cars with a diesel engine). Car tax and VAT implied a total tax rate of 76.9 percent on the average car sold between 2004 and The standard deviation of this average tax rate (due to the price-independent common discount) was 3.4 percentage points. Policymakers announced the CO 2 differentiation of the car tax in November 2007, and the tax reform was implemented in January The common discount was abolished, and the car tax share became an increasing function for vehicle-specific CO 2 emissions. Total tax rates of the differentiated car tax and the VAT range from 39 percent (given a CO 2 emissions rate of 60 g/km or less) up percent (given a CO 2 emissions rate of 360g/km or more). Figure (2.1) compares total tax rates of the old and new car tax. 70

5 The pivot point of rate equivalence of both tax systems lies above the CO 2 emissions rate of the average new car registered during the four years before the tax reform. Policymakers intended this implied reduction of the average tax rate to speed up the replenishment of Finland s aging car fleet. Finland further levies an annual registration tax (ajoneuvovero) on passenger cars. This annual registration tax consisted of two components throughout the data period of this study. A fixed component was levied on all cars while a weight-dependent component was only levied on cars with a diesel engine. Policymakers also announced a future CO 2 differentiation of the annual registration tax in November However, the annual registration tax was not differentiated by CO 2 emissions before March Fuel excise taxes were raised in January The tax hike increased excise taxes on petrol fuel from euro cents to euro cents per liter while increasing excise taxes on diesel fuel from euro cents to euro cents per liter. The CO 2 differentiation of the Finnish car tax was part of a series of European Union policy initiatives to reduce CO 2 emissions rates of passenger cars. Mandatory EU CO 2 emissions standard to manufacturers became European law in The goal of reducing the fleet average to 130 g/km by 2015 came with the threat of severe financial penalties essentially forcing the industry to produce cleaner cars. Car manufacturers were aware of the coming supply-side standards already in 2007 when the European Commission proposed these standards. In 2007, the European Commission also encouraged member standards to introduce supplementary tax incentives to promote the purchase of cleaner cars. 4 The Finnish 2008 car tax reform aimed to provide such incentive. 3.3 Data I use vehicle-level registration data from January 2006 to December 2010 provided by the Finnish Traffic Authority and Statistics Finland. The data therefore include two pre-reform years, an three post-reform years. I link the registration data to the Finnish Customs database on car tax payments to identify registrations of new passenger cars. The linking also 3 See Regulation (EC) No 443/2009 of the European Parliament and of the Council of 23 April See Regulation (EC) No 715/2007 of the European Parliament and of the Council of 20 June

6 removes new cars not liable to taxation, such as taxis and driving school cars. Registration entries contain the brand (make), model name (nameplate), date of registration, average fuel consumption in liters per 100 kilometer, and average CO 2 emissions in grams per kilometer. Detailed vehicle specification names (e.g. Volkswagen Golf Petrol 1.4 TSI 103 Comfortline) are not consistently identified in the registration data. I therefore define a vehicle model specification by brand (e.g. Volkswagen), model or program name (e.g. Golf), fuel type (petrol or diesel) 5, and CO 2 emissions rate, and aggregate the data into monthly new registrations by vehicle model specification. Statistics Finland provides monthly average fuel prices for petrol and diesel. I use these averages to compute model-specific monthly fuel costs in euros per 100 kilometer, discounted to 2008 Euros using the official cost-ofliving index Summary Statistics Table 3.1 presents relevant summary statistics by year. The number of brands active in the market and the number of car models defined by brand, model or program name, and fuel type are stable. On average, each model is available in 11.6 specifications with distinct CO 2 emissions ratings prior to 2008, and 12.2 specifications in the later years. CO 2 emissions ratings both of marketed car model specifications and new registrations begin a steady decline in The 27.8 g/km in CO 2 emissions rates from 2007 to 2010 represents a 15.8 percent decline of this key statistic to policymakers in just three year. The share of diesel engines among available vehicle specifications and new registrations is growing over time. Diesel engines emit all else equal less CO 2 then petrol engines and consume less fuel. However, they emit more local pollutants such NO X and particulate matter but are nevertheless treated preferentially by a CO 2 - differentiated tax. Figure 3.3 illustrates the inverse trend of the registration-weighted CO 2 emissions rate and the market share of diesel engines. The illustration shows a spike of CO 2 rates of new registrations in the two-month announcement period of the car tax reform not 5 All cars in the data run on either petrol or diesel fuel. Registrations of alternative-fuel vehicles were economically irrelevant. 6 Average fuel prices by Statistics Finland are based on the weighted average of fuel prices on the 15th day of each month in six Finnish municipalities. I scrapped local fuel prices from the Finnish fuel price website tankkaus.com to ensure that the official averages do not hide economically important regional variation. 72

7 visible from the yearly statistics. The other main vehicle attributes engine power and curb weight do not exhibit strong trends but increase slightly from 2006 to These summary statistics have two main implications. First, the decline of observed average CO 2 emissions of new registrations cannot be attributed directly to the car tax reform. Consumers chose to buy cleaner cars but their choice set became cleaner as well. Second, the decline of CO 2 emissions rates is not a result of downsizing. In fact, cars and their engines became slightly bigger and more powerful while CO 2 emissions rates declined. Figure 3.4 illustrates the substantial fluctuation of tax-inclusive fuel prices in real terms but also the lack of a clear trend. The January 2008 fuel excise tax hike is overshadowed by non-tax price variation. Differential tax treatment accounts largely for the consistent price advantage of diesel fuel over petrol fuel. Figure 3.2 illustrates the strong seasonal fluctuation of new registrations across the calendar year. Registrations are lowest at the end of each calendar year, and highest in the beginning of each year when the industry releases new and updated models. The two-month announcement period of the car tax reform falls into the end of 2007 with exceptionally low registrations. Total registrations of new cars also vary substantially between years registrations are comparable to previous years. A negative trend in total registrations begins to form in 2007, and is temporarily interrupted in was the worst year of the last twenty years in term of market volume. Registration-weighted total tax rates on the tax-exclusive price declined following the car-tax differentiation in 2008, and continued to decline in the following years. The continuous decline of average tax rates under the differentiated car tax is directly linked to the decline of CO 2 emissions rates. While the total tax rate is exogenous, both the equilibrium price and the tax amount of a car under an ad valorem tax are endogenous. Table 3.1 shows a real decline of the average tax-inclusive price following the CO 2 differentiation of tax rates. Other market developments and strategic price setting by manufacturers in response to the car tax reform affect the equilibrium price determination. The econometric model presented in the next section avoids the explicit modeling of equilibrium price and tax determination. 73

8 3.5 Econometric Model The starting point for the empirical analysis is a linear approximation to the equilibrium relationship between vehicle characteristics, consumer characteristics, and new vehicle registrations: ln q j,t = αf j,t + βco2 j + X j,t β + ξ j,t + ν j,t, (3.1) where q j,t is new registrations of vehicle specification j at time t, f j,t is expected fuel costs, CO2 j is the CO 2 emissions rating, X j,t is a vector of other observed vehicle characteristics, ξ j,t is a scalar that represents the total effect of unobserved vehicle characteristics on registrations, and ν j,t is a scalar representing the total effect of consumer characteristics on registrations. Expected fuel costs depend on the fuel type of the car (petrol or diesel), the price of this fuel over the vehicle lifetime, expected kilometers driven, and the average fuel consumption of the car. Conditional on fuel type, average fuel consumption and average CO 2 emissions are directly related to each other. Other vehicle characteristics in X j,t such as horse power and weight are observed by the econometrician. Unobserved vehicle characteristics in ξ j,t include vehicle characteristics unobserved by the econometrician (e.g. whether the vehicle has climate control or four-wheel drive) as well as intangible attributes (design or handling). The consumer characteristics scalar, ν j,t, accounts for the effect of heterogeneous consumer preferences on registrations. For example, large families might be more likely to buy a certain vehicle than other consumers. Klier and Linn (2010) use a similar linear approximation in their study of gasoline prices and new-car fuel economy in the United States. They point out that the linear approximation in equation (3.1) could be derived from a structural demand and pricing model in the Berry (1994) framework. The equilibrium relationship between CO 2 emissions rating and registrations in equation (3.1) is stable. I discuss the empirical verification of this stability assumption below. However, the CO 2 differentiation of car-tax rates in January 2008 likely had an impact on this relationship. High-emissions cars are taxed at higher rates relative to low-emissions cars following the implementation of the tax reform. A negative impact of the car tax reform on the equilibrium effect of CO 2 emissions rating on registrations is thus expected. Similarly, the announcement of the tax reform might affect the equilibrium relationship between CO 2 emissions rating and registrations 74

9 during the announcement period. The announcement effect is potentially positive, if the upcoming CO 2 differentiation induces consumers with a preference for low-emissions cars to postpone purchases while activating consumers with a preference for high-emissions cars. The reform and announcement effects of the tax reform can be included in equation (3.1) by interacting the vehicle-specific CO 2 emissions rating, CO2 j, with two dummy variables equal to one, if time t lies in either the announcement or post-reform period. Formally, ln q j,t = αf j,t + X j,t β + X j,t β + ξ j,t + ν j,t + βco2 j + (3.2) β a CO2 j 1 {t = announcement} + β r CO2 j 1 {t = post-reform}, where the coefficient β represents the equilibrium relationship between CO 2 emissions rating and registrations in the pre-reform period before November The coefficient β a represents the announcement effect of the tax reform on this equilibrium relationship in November and December The coefficient β r represents the effect of the tax reform on this equilibrium relationship fin January 2008 and later. Note that equation (3.2) is a differences-in-differences specification with continuous treatment applied to the reduced-form effect of vehicle CO 2 emissions rating on equilibrium registrations. Unobserved vehicle characteristics are a key challenge to the estimation of equation of equation (3.2). I follow the idea of Klier and Linn (2010) to use fixed effects based on the cyclical nature of the industry. Car manufacturers in Europe operate in annual production cycles with January as the typical introduction month for new or updated vehicle programs (Klier and Linn, 2013). To control for the evolution of observed and unobserved vehicle characteristics from one year to the next, I use unique intercepts by program / model name, fuel type, and year. The intercepts capture observed, non-co 2 vehicle characteristics and unobserved vehicle characteristics on the model and fuel type level. The fuel type distinction is important to account for differences in registrations of diesel and petrol version of the same car model due to the disproportional annual registration tax on diesel cars. The unique intercepts further control for mean consumer characteristics over the year. I additionally add a set of time fixed effects, τ t, that absorb time-specific deviations from the mean consumer characteristics of the year affecting all 75

10 vehicles proportionally. The estimating equation thus becomes ln q j,t = αf j,t + βco2 j + β a CO2 j 1 {t = announcement} + (3.3) β r CO2 j 1 {t = post-reform} + φ j,y + τ t + ɛ j,t, where the residual ɛ j,t includes both the effect of vehicle-specific deviations from the mean consumer characteristics at time t and the effect of vehiclespecific non-co 2 deviations from the mean model characteristics not captured by the model - fuel type - year intercepts. Estimation of equation (3.3) requires an adequate measure of expected fuel costs. I assume that fuel prices follow a random walk, and that fuel prices affect kilometers driven by the same amount for each model. 7 These two assumptions imply that the cost of driving car j for one kilometer at purchase time t is proportional to the expected fuel costs f jt. I thus use the cost of driving 100 kilometers given the price of fuel at time t as a measure of expected fuel costs. Formally, f j,t = p jt afc j, where p j,t is the price of one liter of fuel for car j (either petrol or diesel) in euros, and afc j is the average fuel consumption of vehicle j in liters per 100 kilometer. The parameters of interest are the fuel cost coefficient α and the CO 2 - related coefficients β, β a, and β r. The cost of driving 100 kilometers with vehicle j is a function of the vehicle s fuel type, its average fuel consumption, and the cost of its fuel. Within year changes in diesel and petrol prices disproportionally affect expected driving costs across vehicles. Time-series variation in fuel costs and within variation of fuel consumption and fuel type across models identify the fuel cost coefficient α. Within variation of registrations and CO 2 rates across different specifications of the same car model identify the pre-reform CO 2 coefficient β. Persistent differences of this variation between the pre-reform, announcement, and post-reform periods identify the announcement and reform CO 2 coefficients β a and β r. The inclusion of time fixed effects in equation (3.3) alters the interpretation of the fuel cost coefficient and the CO 2 - related coefficients. The average effect of a fuel price change on registrations is captured by these fixed effects. The fuel cost coefficient α therefore measures the equilibrium effect of a fuel price change on vehicle-specific registrations relative to the average effect. Similarly, the pre-reform CO 2 coefficient, β, measures the equilibrium effect 7 Anderson et al. (2011) provide empirical evidence that U.S. consumer fuel price forecasts are not distinguishable from a no-change forecast. 76

11 of vehicle-specific average CO 2 emissions on registrations relative to the pre-reform average. The announcement and reform effects, β a and β r, thus represent the impact the CO 2 differentiation of car tax rates has on this relative equilibrium effect. Counterfactual analysis using the estimated model to infer the equilibrium effects of the car tax reform or fuel price changes on total registrations is not possible without these average effects. Changes in registration-weighted average CO 2 emissions however do not depend on the average effects. The estimated relative effects of the car tax reform and fuel prices are sufficient to infer equilibrium effects on CO 2 emissions rates of new cars in the announcement and post-reform time periods. A key assumption of equation (3.3) is the stability of the the relationship between vehicle CO 2 emissions rating and relative registrations conditional on institutional stability. For empirical verification of this stability assumption, I estimate a modified version of the econometric model allowing for timespecific CO 2 coefficients. Formally, ln q j,t = φ j,y + τ t + αf j,t + β t CO2 j + ɛ t, (3.4) where the coefficient CO 2 coefficient β t is time-dependent. Substantial differences between CO 2 coefficients β t in either the pre-reform, announcement, or post-reform period would imply a violation of the stability assumption of (3.3). 3.6 Results Parameter Estimates I first present the parameter estimates of equation (3.4). Figure 3.5 illustrates the period-specific CO 2 coefficient estimates for the monthly data in the top panel, and for quarterly aggregated data in the bottom panel. Aggregation of the monthly observations into quarters reduces the likelihood of measurement error at the month-vehicle level at the cost of reduced variation in the data. While I assume this measurement error to be minimal due to the use of official registration data, I present estimates from the aggregated quarterly data alongside the disaggregated monthly data. The 95 percent confidence interval around the point estimates in figure 3.5 is based on standard errors clustered by brand, fuel type, and market segment. Black lines represent a linear fit to the point estimates in the pre-reform and 77

12 post-reform periods. The announcement period does not have linear fit of its own due to a short length. Point estimates of the CO 2 coefficients from the monthly data in the top panel are indeed stable except for noticeable changes at the announcement and implementation of the car tax reform. The difference between pre-reform and post-reform estimates will be captured by the reform coefficient β r in the main regression equation (3.3). The announcement coefficient β a will capture the difference between pre-reform and announcement estimates. Announcement of the car tax reform had a positive effect on the equilibrium relationship between vehicle CO 2 emissions rating and registrations during the announcement period. Estimates obtained from the aggregated quarterly data in the bottom panel look similar with a noticeable exception: the announcement effect vanishes. Aggregation into quarterly observations merges the two-month announcement period with the previous month with much higher registrations due to seasonality and likely due to the announcement itself. Parameter estimates of equation (3.4) generally support the notion of a stable equilibrium relationship between CO 2 emissions and relative registrations. Before presenting estimates of the main equation (3.3) building on this stability assumption, I note a peculiarity of coefficient estimates for time periods in the second half of Point estimates of CO 2 coefficients of this time period lie above the horizontal linear fit to the estimates in the post-reform period. The latter lies inside the 95 percent confidence interval around the point estimates. However, a sensitivity analysis of equation (3.3) to the interaction of CO 2 emissions rates with a placebo treatment dummy for the suspicious time periods finds a statistical significant effect of this placebo treatment. 8 The year 2009 was the worst year for the Finnish car market during the last twenty years in terms of market volume, with exceptionally low sales also in other European countries due to the financial crisis. I therefore also discuss estimates of equation (3.3) obtained without observations from this unusual time period. Table 2.5 presents the relevant parameter estimates from equation (3.3). All Standard errors are clustered by brand, fuel type, and market segment. The first column contains estimates obtains estimates from the disaggregated monthly data. The relative equilibrium effect of vehicle CO 2 emissions on registrations is negative but not statistically significant. The coefficient of 8 Other placebo tests using interactions of CO 2 emissions rating with dummies for fake treatments did not find neither statistical nor statistical significance of these fake treatments. 78

13 the interaction of vehicle-specific average CO 2 emissions with the post-reform dummy measures the shift of this relationship due to the CO 2 differentiation of car tax rates. This estimated reform effect is negative and statistically significant. The differential tax treatment of cars by CO 2 emissions rating had a negative impact on relative registrations of high-emissions cars. The 2008 car tax reform increased registrations of a given car by 4.02 percent relative to its effect on registrations of another car with a 10 g/km higher CO 2 emissions rating. It is likely that the Finnish car tax reform had on average a positive effect on registrations as it implied a reduction of the average tax rate. However, the average effect of the car tax reform cannot be recovered from the included fixed effects. The coefficient of the interaction of CO 2 emissions with the announcementperiod dummy measures the announcement effect of the car tax reform on the equilibrium effect of CO 2 emissions on relative registrations. This point estimate is positive as expected, and also statistically significant. Announcement of the upcoming CO 2 differentiation increased relative registrations of high-emissions cars during the announcement period. For a specific car, announcement of the upcoming car tax reform increased registrations in the announcement period by 6.17 percent relative to the effect on registrations of another car with a 10 g/km lower CO 2 emissions rating. The fuel cost coefficient estimate is negative and statistically significant. Higher fuel prices have a negative effect on relative registrations of cars with a high fuel consumption and higher CO 2 emissions. The estimate can be interpreted as following. Consider two cars with petrol engines and average fuel consumptions of six and eight liters per 100 kilometers. The fuel efficiency of the six-liter car is 39.2 miles per gallon and its average CO 2 emissions are g/km. The fuel efficiency of the eight-liter car is 29.4 miles per gallon and its average CO 2 emissions are g/km. A 25 euro cent increase of the liter price of petrol fuel (in 2008 euros) reduces new registrations of the eight-liter car by around 10.8 percent compared to the effect of the fuel price increase on new registration of the more fuel-efficient six-liter car. The second column of table 2.5 presents parameter estimates of (3.3) estimated with the aggregated quarterly data. The announcement effect estimate is close to zero and statistically insignificant as expected from the bottom panel of figure 3.5. The reason for the disappearance of the announcement effect is apparent in figure 3.2. The reform effect and the effect of fuel costs on relative registrations remain statistically significant. Both 79

14 are slightly stronger in magnitude compared to the monthly-data estimates. The adjusted R 2 of the linear regression increases only marginally, and the aggregation of the monthly observations into quarterly ones does not alter the statistical insignificance of the pre-reform relationship between CO 2 emissions rating and relative equilibrium registrations. The fifth and sixth column of table 2.5 present parameter estimates of equation (3.3) obtained without observations from the second half of The sample limitation does not affect the parameter estimates for the CO 2 coefficients in the pre-reform and announcement period, and also does not affect the fuel cost coefficient. Estimates of the reform s effect on the equilibrium relationship between CO 2 emissions rates and relative registrations are slightly stronger compared to complete-sample estimates. The estimate from the monthly-data specification implies that the CO 2 differentiation of car tax rates increased registrations of a given car by 4.79 percent compared to its effect on registrations of another car with 10 g/km lower CO 2 emissions. Estimation of this effect using the complete sample yields a 4.02 percent effect. I also estimated equation (3.3) without observations from 2009, also excluding the unsuspecting ones from the first half of the year. Results (not reported) of this estimation are similar to the ones in the fifth and sixth column of table (2.5) The Effect of the Car-Tax Reform Effects on CO 2 Emissions Rates The parameter estimates from equation (3.3) allow estimation of the car-tax reform s effect on registration-weighted average CO 2 emissions in the postreform period. To do so, I predict counterfactual registration-weighted CO 2 emissions rates in the absence of the tax reform by setting the post-reform dummy in equation (3.3) to zero. I then use these predicted counterfactual registrations to compute counterfactual weighted CO 2 emissions rates for the post-reform time period. The difference between observed and counterfactual CO 2 emissions rates is the effect of the car-tax differentiation on this key statistic to policymakers. I similarly estimate the announcement effect of the tax reform on CO 2 emissions rates in the announcement period by setting the announcement-period dummy to zero. Note that the predicted counterfactual registrations do not account for the average effect of the car tax reform on registrations. This average effect is captured by the time fixed effects included 80

15 in equation (3.3) yet not relevant for the computation of CO 2 emissions rates. Table 3.3 presents the estimated effects of the CO 2 differentiation of car tax rates on yearly CO 2 emissions rates based on the estimates of equation (3.3) in table 2.5. The uncertainty about these parameter estimates carries over to the effect on CO 2 emissions rates. I use Monte Carlo simulation (n = 500) to compute their standard errors. The impact of the car tax reform on CO 2 emissions rates is small regardless of the used specification. Observed average CO 2 of new registered cars in Finland declined from g/km in 2007 to g/km in According to the monthly specification in the first column, CO 2 emissions rates in 2010 would have only been 2.5 g/km higher, if car tax rates hat not been differentiated by CO 2 emissions in January The use of the aggregated quarterly data results in a stronger effect of the car tax reform of 3.9 g/km in 2010 but even this effect is small compared to the 27.8 g/km decline from 2007 to Excluding observations from the second half of 2009 from the estimation leads to stronger reform effect estimates, and therefore also to stronger effects of the car tax reform on CO 2 emissions rates. These effects remain nevertheless small in comparison to the observed decline of CO 2 emissions rates. The third and fourth column in table 2.5 present these estimates. Estimated standard errors of these effects are small for all specifications. The CO 2 differentiation of car tax rates thus had a measurable and noticeable effect on CO 2 emissions rates of new passenger cars in Finland. Consumers would have nonetheless predominantly bought cleaner cars regardless of the car tax reform, and the effect of the CO 2 differentiation of car-tax rates is declining over time. A potential explanation of the declining impact of the reform over time is that cars marketed in Finland became cleaner throughout the post-reform period. The differentiated-tax system might thus provide less of an incentive to choose cars cleaner than the market average. The reduced-form approach in this study is however limited to the estimation of equilibrium effects, and does not allow their disentanglement. Figure (3.6) illustrates the car tax reform s effect on monthly and quarterly CO 2 emissions rates based on specifications (1) and (2) in table 2.5. The monthly figure in the top panel clearly shows the estimated announcement effect of the car tax reform on CO 2 emissions rates in the last two month of Registration-weighted average CO 2 emissions of new cars in November 2007 would have been 9.6 g/km lower (standard error 1.7 g/km) in the absence of the reform announcement. CO 2 emissions rates in December 2007, 81

16 the second month of the the announcement period, would have been 8.6 g/km lower (standard error 1.5 g/km) The Effect of Fuel Prices on CO 2 Emissions Rates Finnish Policymakers also raised fuel excise taxes when differentiating the car tax in January Excise taxes on petrol fuel increased from euro cents to euro cents per liter excise taxes on diesel fuel from euro cents to euro cents per liter. Regression equation (3.3) permits estimation of the effect of this fuel excise tax hike on CO 2 emissions rates of new cars. I use the fuel cost coefficient estimate to predict counterfactual registrations in the absence of this tax hike affecting all cars disproportionally. Table 3.4 presents the estimated effects of this fuel excise tax hike on yearly registration-weighted CO 2 emissions rates. I only report results based on parameter estimates in the first and second column of table 2.5. Exclusion of the suspicious 2009 observations does not affect the relevant fuel cost coefficient estimate. The estimated effects are statistically significant but economically unimportant for all specifications. The fuel excise tax hike had no part in the strong observed decline of CO 2 emissions rates. I also estimate the effect of a hypothetical fuel excise tax hike of 25 euro cents per liter (in 2008 euros). This counterfactual further demonstrates the general insensitivity of CO 2 emissions of new cars to fuel prices. Table 3.5 presents the estimated effects of this substantial, hypothetical tax hike on yearly CO 2 emissions rates from 2008 to Although substantial, such tax hike would not drive consumers in the market towards much cleaner cars with lower average fuel consumption. Registration-weighted CO 2 emissions rates are generally unresponsive to even substantial fuel price changes. The decline of this limited effect over time relates to the nominal nature of the simulated tax hike. 3.7 Concluding Remarks This study has estimated the effect of the 2008 Finnish CO 2 differentiation of car-tax rates on CO 2 emissions rates of new passenger cars. The econometric analysis uses rich individual-vehicle registration data to estimate the reform s effect on relative equilibrium registrations of cars with different CO 2 emissions rates. The estimated relative effect is sufficient to quantify the tax reform s impact on registration-weighted average CO 2 emissions. This key statistic 82

17 to policymakers declined by 27.8 g/km from 2007 to My results attribute only a small share of this decline to the car-tax CO 2 differentiation. Depending on the data used for the estimation, I find the reform s impact on CO 2 emissions rates in to range 2.5 g/km to 4.7 g/km. A moderate fuel excise tax hike accompanying the CO 2 differentiation of the car tax had no economically important effect. CO 2 emissions rates of new cars in Finland are in general insensitive to even substantial fuel tax hikes. In Stitzing (2016c), I estimate a structural demand and pricing model of the Finnish car market to evaluate the distributional effects of the 2008 car tax reform. The static demand model precludes the evolution of consumer preferences unlike the more flexible reduced-form analysis of this study. The effect of the car tax reform on CO 2 emissions rates estimated with the structural model are even slightly smaller in magnitude than the estimates of the more flexible reduced-form approach to this research question. Both studies thus present additional supporting evidence for the relative unimportance of demand-side fiscal policies to reduce CO 2 emissions rates in Europe in light of the concurrent introduction of mandatory supply-side CO 2 standards. Finnish consumers would have predominately bought cleaner cars regardless of domestic fiscal policy. The reduced-form analysis of this study is limited to the estimation of the tax reform s equilibrium effect on CO 2 emissions rates. An advantage of the reduced-form approach to this specific research question is the use of monthly data to detect an announcement effect of the tax reform. The structural model in Stitzing (2016c) is unable to do due to the use of aggregated yearly data. The unusually short announcement period of the 2008 car tax reform with very few new registrations at the end of the year limits the importance of this announcement effect in absolute terms. Its existence nevertheless shows that markets react to fiscal policy even before the implementation of a tax reform. Optimal tax policy needs to consider potential dynamic responses to fiscal policy. References ACEA European Automobile Manufacturer s Association (2010): Consolidated Registration Statistics, tag/category/by-country-registrations, accessed:

18 Anderson, S. T., R. Kellogg, and J. M. Sallee (2011): What Do Consumers Believe About Future Gasoline Prices?. Berry, S. T. (1994): Estimating Discrete-Choice Models of Product Differentiation, RAND Journal of Economics, 25, Ciccone, A. (2015): Environmental Effects of a Vehicle Tax Reform: Empirical Evidence from Norway, Memorandum. Gerlagh, R., I. van den Bijgaart, H. Nijland, and T. Michielsen (2015): Fiscal Policy and CO2 Emissions of New Passenger Cars in the EU, Working paper. Klier, T. and J. Linn (2010): The Price of Gasoline and New Vehicle Fuel Economy: Evidence from Monthly Sales Data, American Economic Journal: Economic Policy, 2, Klier, T. and J. Linn (2013): Fuel prices and new vehicle fuel economy - Comparing the United States and Western Europe, Journal of Environmental Economics and Management, 66, Klier, T. and J. Linn (2015): Using Taxes to Reduce Carbon Dioxide Emissions Rates of New Passenger Vehicles: Evidence from France, Germany, and Sweden, American Economic Journal: Economic Policy, 7, Reynaert, M. (2015): Abatement strategies and the cost of environmental regulation: Emission standards on the European car market, Working Paper. Stitzing, R. (2016c): Distributional and Environmental Effects of an Emissions-Differentiated Car Sales Tax, Working paper. Tables 84

19 Table 3.1 Summary Statistics by Year Year Registrations Brands Models a Model specifications b Price c Tax rate c N N N N 2008 euros percent , , , , , , , , , , Year Power c Weight c Diesel c Diesel b CO2 c CO2 b hp kg percent percent g/km g/km , , , , , a defined by brand, model name / program name, fuel type b model-average c registration-weighted average 85

20 Table 3.2 Parameter Estimates (1) (2) (5) (6) CO (0.0270) (0.0339) (0.0271) (0.0349) CO 2 announcement *** *** (0.0110) (0.0094) (0.0110) (0.0095) CO 2 post-reform *** *** *** *** (0.0145) (0.0159) (0.141) (0.0155) Fuel cost *** *** *** *** (euros / 100km) (0.0476) (0.0616) (0.0474) (0.0631) Level of observation month quarter month quarter Excluded time periods none none 2009 July to December 2009 July to December Observations 55,782 25,391 50,613 22,902 R Standard errors in parentheses. * p<0.1, ** p<0.05, *** p<0.01 Notes: The dependent variable is log registrations. Standard errors are clustered by brand, market segment, and fuel type. All regressions include time fixed effects, and unique intercepts by model - fuel type - year. 86

21 Table 3.3 The Effect of the CO 2 Differentiation of Car Tax Rates on CO 2 Emissions Rates (1) (2) (3) (4) observed CO 2 emissions = g/km (1.2) (1.4) (1.2) (1.5) observed CO 2 emissions = g/km (1.1) (1.3) observed CO 2 emissions = g/km (0.9) (1.0) (0.8) (1.0) Level of observation month quarter month quarter Excluded time periods none none Notes: The table reports the effect of the 2008 CO 2 differentiation of car-tax rates on registration-weighted average CO 2 emissions (in g/km). Effects estimated using the coefficients reported in table (3.2). Standard errors are estimated using Monte Carlo simulation (n = 500). Table 3.4 The Effect of the January 2008 Fuel Excise Tax Hike on CO 2 Emissions Rates (1) (2) (3) (4) observed CO 2 emissions = g/km (0.1) (0.1) (0.1) (0.1) observed CO 2 emissions = g/km (-0.1) (0.1) (0.1) (0.1) observed CO 2 emissions = g/km (0.1) (0.1) (0.1) (0.1) Level of observation month quarter month quarter Excluded time periods none none Notes: The table reports the effect of the January 2008 fuel excise tax hike on registration-weighted average CO 2 emissions (in g/km). Effects estimated using the coefficients reported in table (3.2). Standard errors are estimated using Monte Carlo simulation (n = 500). 87

22 Table 3.5 The Effect of a Hypothetical 25 Euro Cent Fuel Excise Tax Hike on CO 2 Emissions Rates (1) (2) 2008 (observed CO 2 emissions = g/km) (0.4) (0.6) 2009 (observed CO 2 emissions = g/km) (0.4) (0.5) 2010 (observed CO 2 emissions = g/km) (0.3) (0.4) Level of observation month quarter Model - fuel type - year fixed effects yes yes Excluded time periods none none Notes: The table reports the effect of a hypothetical 25 euro cent / liter fuel price increase on registration-weighted average CO 2 emissions (in g/km). Effects estimated using the coefficients reported in table (3.2). Standard errors are estimated using Monte Carlo simulation (n = 500). 88

23 Figures Figure 3.1 The 2008 Reform of the Finnish Car Tax new car average Total Tax Rate (%) Car Tax CO2 differentiated (2008 and later) non differentiated (before 2008) CO2 (g/km) Figure 3.2 Monthly New Registrations of Passenger Cars m1 2004m7 2005m1 2005m7 2006m1 2006m7 2007m1 2007m7 Month 2008m1 2008m7 2009m1 2009m7 2010m1 2010m7 2010m12 Announcement period New registrations (in 1000's) 89

24 Figure 3.3 Average CO 2 Emissions of New Passenger Cars and Market Share of Diesel-Fueled Cars. Market Share (%) m1 2004m7 2005m1 2005m7 2006m1 2006m7 2007m1 2007m7 Month 2008m1 2008m7 2009m1 2009m7 2010m1 2010m CO2 (g/km) Diesel market share (left y-axis) Announcement period CO2 emissions rate (right y-axis) Figure 3.4 Monthly Average Prices for Diesel and Petrol Euro / liter m1 2004m7 2005m1 2005m7 2006m1 2006m7 2007m1 2007m7 2008m1 2008m7 2009m1 2009m7 2010m1 2010m7 2011m1 Month Petrol Announcement period Diesel Note: in 2008 euros 90

25 Figure 3.5 Estimated CO2 Coefficients Monthly Data beta beta 2006m1 2006m7 2007m1 2007m7 2008m1 2008m7 2009m1 2009m7 2010m1 2010m7 2010m q1 2006q3 2007q1 2007q3 2008q1 2008q3 2009q1 2009q3 2010q1 2010q3 2010q4 Month Quarterly Data Quarter CO2 Coefficient estimate 95% Confidence interval Linear fit Announcement Period CO2 Coefficient estimate 95% Confidence interval Linear fit Announcement period Notes: The figure shows month-specific CO2 coefficients estimated from monthly and quarterly data. The dependent variable are log registrations of car models defined by make, model name, fuel type, and CO2 emissions rating. Standard errors are clustered by brand, market segment, and fuel type. All regressions include time fixed effects, and unique intercepts by model - fuel type - year. 91

26 Figure 3.6 The Effect of the CO 2 Differentiation of Car Tax Rates on CO 2 Emissions Rates CO2 (g/km) m1 2006m7 2007m1 2007m7 Monthly Data 2008m1 2008m7 Month 2009m1 2009m7 2010m1 2010m7 2010m12 Observed No tax reform 95% C.I. Announcement period CO2 (g/km) q1 2006q3 2007q1 2007q3 Quarterly Data 2008q1 2008q3 Month 2009q1 2009q3 2010q1 2010q3 2010q4 Observed No tax reform 95% C.I. Announcement period Notes: based on parameter estimates of specification (2) with clustered standard errors. Confidence interval of counterfactual emissions constructed using Monte carlo simulation. 92