Climate, Technology and Value: Insights from the First Decade with Mass-Consumption of Electric Vehicles

Abstract

Adoption of low-carbon technology is key to mitigating climate change. A possible unwanted consequence of fast technological progress is that products get outdated before their technical lifetime is over. We investigate whether the market value of electric vehicles, characterized by rapid technological progress, declines faster over their lifetime than gasoline vehicles, which represent a mature technology. We use novel data from Norway, the market with the highest market shares for electric vehicles in the world. The data are from the largest web platform for secondhand vehicles over the period 2011–2021. Prices of electric vehicles decline faster than gasoline vehicles. This seems to be driven by the electric vehicles with below median driving range. We hypothesize that the large price drop is mainly due to the fast technological improvement of electric vehicles.

1 Introduction

Widespread deployment and continued progress of low-carbon technologies are crucial to reduce greenhouse gas emissions (Barrett 2009; Stock 2020; Pathak et al. 2022; Dhakal et al. 2022). The adoption of low-carbon small-scale technologies, such as electric vehicles and solar power, has accelerated the last decade. These technologies have improved fast, both regarding performance and costs (e.g. Pathak et al. 2022). Better and cheaper low-carbon technologies enable more consumers to take part in the energy transition—but also make existing goods obsolete faster than before. We investigate how the market values technology throughout the lifetime when the technology is rapidly improving.

In this paper, we specifically investigate the case of electric vehicles. For electric vehicles, the most important improvement the last decade is the increased driving range, which is due to better and cheaper batteries. This has also allowed production of larger models catering to a larger range of consumers. Gasoline vehicles, on the other hand, is a mature technology. We find the price path of a car over its lifetime by looking at cars of the same model year, make and fuel sold secondhand at different ages. To study the effect of particularly rapid technological progress on market prices, we compare the price path for ten-year-old electric vehicles and gasoline vehicles.

Improved technology performance and cost reduction are often necessary to increase technology adoption for low-carbon technologies. But for durable goods technological progress can be a paradox. As the resale value may be lower if the newest technology available is much better and cheaper than the used technology, this increases the total cost of ownership. How much better the newest technology has to be compared to the used technology to lower the resale value is an empirical question. We investigate this specifically for electric compared to gasoline vehicles, using driving range as a proxy for technological improvement. The valuation of used technology is relevant for other durable, low-carbon, small-scale technologies as well, such as solar power. The potentially reduced resale value for low-carbon technologies with large technological progress need to be factored in when estimating the cost of climate policies.

The Norwegian market is the most mature electric vehicle market in the world (IEA 2022), so this is a good test bed for the future of the market for electric cars worldwide. The market share for battery electric vehicles has been above 10% of new car purchases since 2014 and was 82% in 2023 (OFV 2024).¹ We use a novel data set from the largest web platform for secondhand vehicles in Norway, finn.no, accounting for close to 90% of the market, over the years 2011-June 2021. At this platform, the electric vehicle share of the market for vehicles that are 10 years or younger has been above 1% since 2013 and is 23% as of June 2021. The time period 2011–2021 covers the years where electric vehicles have been an important part of the Norwegian market.

The technology shift in the Norwegian car market is a prediction of what may happen in other countries if batteries continue to fall in price and policies promoting electric vehicles and charging infrastructure continue. Other important car markets are less than 10 years behind Norway in the electric vehicle adoption.

We find that used electric vehicles decline faster in price than used gasoline vehicles. This finding is robust to various specifications. The difference in the price fall seems to be driven by electric vehicles with driving range below 200 km (close to the median range in our sample). The price pattern of electric cars with the highest driving ranges are more similar to those of gasoline cars.

To understand the difference in the price decline between the two technologies, we first develop a theoretical framework based on the model of Stolyarov (2002) of the resale pattern of durable goods. As electric cars are based on less mature technologies than cars based on combustion engines, technological progress is faster. This reduces the utility of used cars relative to that of a new car, and hence leads to reduced demand and increased supply of secondhand cars. To maintain equilibrium in the market of used cars when the technology is improving, the price of used cars has to decrease. Hence the price of electric vehicles drops faster compared to gasoline vehicles.

This paper contributes thematically to three different strands of the literature, namely, technological development, durable goods and the car market. In growth theory there is a literature on how technological progress evolves endogenously as resources are spent on R &D (e.g. Aghion and Howitt 1998). This can also be extended to the interaction between technological change and environmental policy (e.g. Acemoglu et al. 2012). In this paper, however, we take the technological process as exogenously given. Indeed, our work relates more closely to the work on vintage capital where technological progress only affects the economy through new capital—and where every vintage of capital is marked by the period in which it was produced (Solow et al. 1966). As technology is embedded in the car, this leads to questions of optimal replacement of capital (Rust 1987). In the case of rapid technological change, as is the case for electric vehicles, there is also a question of how expectations about future technological improvements affect the choice between adopting technology today or waiting for improved technology (Rosenberg 1976; Balcer and Lippman 1984). De Groote and Verboven (2019) a case resembling our case, household’s investment decision in solar power, where both quality is increasing and prices decreasing over time.

Cars are one of the prime examples of durable goods. Waldman (2003) reviews much of the early literature on durable goods, including theories of optimal durability of goods, the effects of a secondhand market on the new goods market, and information problems in the secondhand market. A study that in many ways is close to ours is the one by Fudenberg and Tirole (1998). They consider a monopolist selling goods in two periods with improvements in quality between the two periods. Their focus, however, is on the behavior of the monopolist selling the new good and taking the secondhand market for given whereas we focus on the secondhand market. In a seminal contribution close to our theoretical approach, Stolyarov (2002) develops a model of the resale pattern of durable goods. Goods are purchased new at a fixed price and the quality of the product deteriorates over time. He studies more complex deterioration patterns and markets with many vintages. To focus on the effect of different types of technology, we only include old and new varieties.

There is a vast literature on the demand for cars. One starting point is Akerlof’s (1970) seminal theoretical study, which in itself is not very relevant to our study. This study was followed up by Bond (1982, 1983), who studied the demand for used pickup trucks empirically. To see whether bad cars are driving out good ones from the secondhand market, he investigates whether secondhand trucks require more maintenance than otherwise similar trucks bought new. He does not find any difference between first- and secondhand bought trucks. This is one of the first empirical studies of car price behavior.

The newer empirical literature on the demand for automobiles is heavily inspired by the study by Berry et al. (1995). In their study, however, they only consider the static demand for new cars. A number of extensions have been suggested in the literature.² Our empirical approach is most closely related to the approach followed by Purohit (1992). He estimates the effect of changes such as major styling changes, downsizing and changes in horsepower, in new cars on the price of used cars and finds that enhanced features of new cars increases the obsolescence effect of used cars. The changes he studies, however, are minor compared to the technological changes electric vehicles have experienced the last decade.

Our paper is also closely related to the literature on the evolution of the price of used durable goods, particularly cars, over time.³ Wykoff (1970) studies the declining price of selected car models to study the shape of the depreciation function. A number of other studies has corroborated the declining price pattern of both secondhand and new cars over time (Porter and Sattler 1999; Esteban and Shum 2007; Copeland et al. 2011). But as should be obvious from the date of these studies, they only study cars with combustion engines.

Schloter (2022) investigates the depreciation rate of electric vehicles compared to gasoline vehicles, and finds that electric vehicles depreciate faster than gasoline vehicles. He draws on web scraped data on prices of nine car makes from multiple countries collected in 2020 and 2021. In comparison we cover the whole used car market on finn.no (which again covers around 90% of the used car market in Norway), while Schloter (2022) ends up only looking at electric vehicles in Norway. Moreover, as we have the whole used car market for ten years in a country that is far ahead on the electric vehicle transition, it gives a more complete picture of the evolution of the market for electric cars although we are limited to a single country. Reassuringly, Schloter (2022) finds the same pattern of prices of electric vehicles decline faster than the price than gasoline vehicles that we find. We also have data on the range of the vehicle models and find that the fast price decline of electric vehicles seem to be driven by vehicles with below median range.

Also related to our study, Sallee et al. (2016) use used car prices in their analysis, but investigate whether consumers recognize the value of fuel economy. Also Strittmatter and Lechner (2020) use similar data, investigating whether there is sorting in the secondhand car market based on environmental quality.

Finally, there is an emerging literature on the demand for electric and “green” cars.⁴ Muehlegger and Rapson (2022) study the effect of a subsidy of electric vehicles in California and find a quite high price elasticity where low- and middle-income households benefit the most from the subsidy. There is also a literature on provision of incentives to replace polluting cars with newer varieties.⁵

Our first contribution is to present empirical results that indicate that the price paths of goods that have a large technological improvement each year is different than for goods that do not have large technological improvements from year to year. Second, we contribute to the state of knowledge on the electric vehicle market. Investigating the price path of electric vehicles during the lifetime with information from the most mature electric vehicle market in the world has not been done before and is relevant since this is a technology that probably will be phased into the car market globally the coming decade. Third, we inform the estimation of the cost of climate policy since valuation of the used low-carbon technology is part of the cost.

The structure of the rest of the paper is as follows. First, we give background information about the car market in Sect. 2. In Sect. 3 we present our theoretical framework that we use to make a hypothesis about the cause of the findings. Further, we present the data, descriptive statistics and the empirical strategy in Sect. 4. In Sect. 5 we present and discuss the results and in Sect. 6 we conclude.

2 Background on the Car Market

An extraordinary technology shift is expected to happen in the car market. Both the EU and California aim to only sell zero emission passenger cars from 2035 and on wards. A car lasts for 15–20 years. For the car fleet to transition to zero emission vehicles, the market share for zero emission vehicles among the new cars sold need to increase first. A large part of this shift has already happened in the Norwegian car market over the last decade. The share of different types of vehicles sold in Norway from 2011 to 2021 can be seen in Fig. 1a. From having 96% market share in 2011, gasoline and diesel vehicles have plummeted to a 10% market share. Electric vehicles have increased from a 1.5% market share in 2011 to 57% by June 2021.⁶

Fig. 1

Sale shares by technology for new and second hand cars in the Norwegian market 2011–2021. Data for 2021 covers January–June

The high sale volumes of new electric vehicles extends into the secondhand vehicle market, but naturally there is some delay. The fuel share of the secondhand vehicle market in Norway in 2011–2021 can be seen in Fig. 1b, where we for each year show the fuel share of all cars that are below 10 years old that are advertised on the web platform finn.no.⁷ The share of electric vehicles hits 1% in 2014, and reached 23% by June 2021. The left part of the graph for new vehicles (Fig. 1a) from 2011 to 2017 looks similar to the right part of the graph for used vehicles from 2014 to June 2021 (Fig. 1b).

The sale shares of electric vehicles in the USA, Korea and Canada in 2021 are equal to the sale shares of new electric vehicles in Norway about ten years ago (IEA 2022).⁸ Globally the sale share of electric vehicles in 2021 is equal to the sale shares in Norway in 2013, while China and Europe are ahead of the USA and have sale shares for electric vehicles in 2021 equal to the sale shares in Norway in 2013–2014 (IEA 2022). The Netherlands have sale shares for electric vehicles in 2021 that are equal to the Norwegian sale shares in 2017 (IEA 2022) while Iceland has sale shares for electric vehicles in 2021 that match the Norwegian 2018 numbers (The Norwegian Electric Vehicle Association 2022b). Thus, these markets are less than 10 years behind Norway in the electric vehicle roll-out.

Driving range has increased considerably since the market of electric vehicles started to grow in the beginning of the 2010s. In addition, comfort of electric vehicles has increased and the variety of vehicle models available has expanded. In Fig. 2 we see the number of electric vehicle models available in the Norwegian market until June 2021. The black line is the total number of electric vehicle models available, while the dotted line is the number of large electric vehicle models available. We see that around 50 electric vehicle models are available in the Norwegian car market. As a comparison there are 153 different models available in 2020 for gasoline vehicles. Thus, the electric vehicle market is maturing, but it is not fully mature yet.

Fig. 2

Number of available electric vehicle models for sale. Data for 2021 is as by June 2021. Large vehicles are defined as station wagon, SUV, pickup and multipurpose vehicles. Small vehicles are sedan, hatchback, coupe and cabriolet. These are models that are imported by the car manufacturers official seller network. Imported models from other actors are not included here

3 Theoretical Framework

There are several explanations for why prices of cars decline over time. One obvious explanation is depreciation. Our study encompasses this effect, but the main focus is on the effect of technological progress. Gasoline vehicles is a mature technology where we expect technological progress to be slower than for electric cars. To fix ideas on the effect of these two factors, we present a simple version of Stolyarov’s (2002) model tailored to the secondhand market for cars. This is a dynamic model with an infinite time-horizon for the consumers. We focus on the steady state of the model, i.e. the situation where the number of buyers and sellers of new and used cars is constant over time.

A car has a characteristic x we refer to as its quality. The market consists of a continuum of consumers each described by a parameter h which indicates their valuation of cars. The parameter h follows a distribution over some interval $\Omega \subseteq \mathbb {R}_+$ with cumulative distribution function F. We assume no holes and no mass points. A consumer with characteristic h has an instantaneous utility over cars of quality x and consumption c given by

$$\begin{aligned} U=xh+c, \end{aligned}$$

(1)

where consumption is the income not spent on car purchases. For simplicity and without loss of generality, utility from other consumption is disregarded in what follows. Consumers have a common discount factor $\beta \in (0,1)$.

Cars last for two periods. In the first period, they provide a quality $x_n$. The second period, they deteriorate and provide quality $x_u=\mu x_n$ with $\mu \in (0,1)$. After two periods, the car provides no further benefit.

3.1 The Market for Cars Without Technological Progress

Consider first the market for cars that do not have technological progress. This can be seen as conventional cars. The technology in this market is mature, and a new car has the same utility characteristic $x_n$ every period. New cars are sold at some exogenously given price $p_n$. This could correspond to a small country that acts as a price taker in the global market for cars or a competitive car industry with constant returns to scale. Used cars are traded in the secondhand market at a price $p_u$.

Assume initially that everybody needs a car. Consumers in the market have three options. They can each period buy a new car and sell it after one period, they can buy a new car and drive it for two periods, or they can every period buy a used car. Let $U^N$, $U^K$, and $U^O$ denote the utility of these options. Notice that the K group has no impact on the market for secondhand cars except from being an outside option. The main reason for keeping a car throughout it’s lifetime rather than either always buying new or secondhand cars may be transaction costs. To avoid having to model transaction costs explicitly, we assume that consumers can commit to a two period strategy. Then consumers with moderately strong preference for new cars will chose the intermediate strategy, which acts as a mixing strategy.

We have

$$\begin{aligned} U^N&=x_n h-p_n+\beta (x_n h-p_n +p_u) \end{aligned}$$

(2)

$$\begin{aligned} U^K&=x_n h-p_n +\beta x_u h \end{aligned}$$

(3)

$$\begin{aligned} U^O&=x_u h-p_u + \beta (x_u h-p_u) \end{aligned}$$

(4)

The utilities as function of h are illustrated in Fig. 3.

Fig. 3

Utility of various car ownership strategies by h

Consumers, differentiated though their index h, then choose the option N, K, or O that suits their preferences the best. We solve the model by letting consumers make a plan for two periods. The outcome is, however, dynamically consistent so no consumer would like to change her mind in the second period. Whenever the price of secondhand cars is below the price of new cars ($p_u<p_n$), consumer with $h=0$ prefers O over K and K over N. As the slope of the utility functions is steeper for the latter two, however, consumers with high valuation of car quality (high h) chooses option N, consumers with intermediate h chooses option K, and low valuation consumers choose option O. Specifically, there are levels $\underline{h}$ and ${\bar{h}}$ so consumers with $h<\underline{h}$ chooses O, consumers with $\underline{h}<h<{\bar{h}}$ chooses K, and consumers with $h>{\bar{h}}$ chooses N. The parameters are defined by indifference, so

$$\begin{aligned} x_u \underline{h}-p_u + \beta (x_u \underline{h}-p_u)&= x_n \underline{h}-p_n +\beta x_u \underline{h} \\ x_n {\bar{h}}-p_n+\beta (x_n {\bar{h}}-p_n +p_u)&= x_n {\bar{h}}-p_n +\beta x_u {\bar{h}} \end{aligned}$$

yielding

$$\begin{aligned} \underline{h}&= \frac{p_n-(1+\beta )p_u}{x_n-x_u} \\ {\bar{h}}&= \frac{p_n-p_u}{x_n-x_u}. \end{aligned}$$

Equilibrium in the market for secondhand cars requires that $p_u$ is chosen so the number of sellers and buyers of secondhand cars, i.e. the groups N and O, are equal. Using the assumption that the quality of cars deteriorate by a factor $\mu$ so $x_u=\mu x_n$, this yields the equilibrium condition

$$\begin{aligned} F\left( \frac{p_n-(1+\beta )p_u}{(1-\mu )x_n} \right) = 1-F \left( \frac{p_n-p_u}{(1-\mu )x_n} \right) \end{aligned}$$

(5)

Implicit differentiation of this expression yields

$$\begin{aligned} \frac{\partial p_u}{\partial \mu } = \frac{f\left( \underline{h} \right) \underline{h} + f\left( {\bar{h}}\right) {\bar{h}} }{(1+\beta ) f\left( \underline{h}\right) + f\left( {\bar{h}}\right) } x_n > 0. \end{aligned}$$

(6)

Hence, as we should expect, if the quality of a car retained after a period increases, the price of secondhand cars also increases. It could also be the case that some parameters of the model are unknown to the consumer. Particularly the quality of a used car may be challenging to observe, as explored e.g. in the famous market for lemons (Akerlof 1970). This would imply that a risk averse consumer achieves lower ex ante utility from a secondhand vehicle, and this could be modelled as a higher quality degradation factor $\mu$.

For the case of electric cars, there is a worry that the battery capacity is declining over time. Usually, batteries will lose one to two percent of their capacity each year (Nichols 2022), but this depends on how the car is used and charged. There may also be a popular belief that battery degradation is stronger than it actually is. In the model, we can model battery degradation as $\mu$ being lower for electric cars than for gasoline cars, and hence having a price that declines faster. With risk averse consumers, this would also be the case if there is higher uncertainty about the degradation of electric vehicles than gasoline vehicles.

So far in the analysis, we have assumed that all consumers need a car, hence implicitly that the utility of not owning a car is sufficiently low. In many settings, other means of transportation can be a suitable alternative. To include this in the model, we can introduce an outside opportunity of not owning a car yielding a utility level $\underline{U}$. A full analysis of this case can be found in Appendix A.1. This change to the model does not change any qualitative insights from the model, so to maintain simplicity we disregard the presence of an outside opportunity for the rest of the paper.

The solution studied above is a steady state where the price of secondhand cars remains stable. This is a reasonable description of a mature market. However, at the first introduction of cars to the market, there were of course no secondhand car initially. Subsequently the economy converged to an economy where the three groups N, K, and O obtained stable sizes and hence that prices stabilized. We do not think that this dynamic is particularly enlightening to study formally.

3.2 The Market for Cars with Technological Progress

Consider now a market for cars with technological progress, i.e. that the utility of a new car is increasing over time. This market can for example be the case of electric cars. The technological progress can be the general comfort of driving the car, but maybe more relevant is the car’s driving range and battery capacity. We use this extension to study the market for electric cars and compare the market for secondhand electric and conventional cars. For analytical tractability, we disregard interaction between the two markets.

To account for technological progress, we assume that the utility characteristic $x_n$ of a new car increases by a rate $\gamma > 1$ every period. Seen from a specific period where new cars have quality $x_n$, next period’s new cars have quality $\gamma x_n$ whereas new cars last period were sold with quality $\frac{x_n}{\gamma }$. We still maintain the assumption of a constant price $p_n$ of new cars and study how the price of secondhand cars depend on the rate of technological progress $\gamma$.

This implies two changes for the consumers’ choice of car ownership option. Consumers always buying new cars (N) face an improved car technology next period, increasing their willingness to sell their used car on the secondhand market. Consumers always buying used cars (O) face an obsolete used car in every period, reducing their utility of the strategy. The new utilities of the three options become

$$\begin{aligned} \left\{ \begin{aligned} U^N&=x_n h-p_n+\beta [\gamma x_n h-p_n +p_u] \\ U^K&=x_n h-p_n +\beta \mu x_n h \\ U^O&=\frac{\mu }{\gamma } x_n h-p_u + \beta (\mu x_n h-p_u). \end{aligned} \right. \end{aligned}$$

(7)

This yields the cut offs

$$\begin{aligned} \underline{h}&= \frac{p_n-(1+\beta )p_u}{\left( 1-\frac{\mu }{\gamma }\right) x_n} \\ {\bar{h}}&= \frac{p_n-p_u}{(\gamma -\mu )x_n} \end{aligned}$$

and the equilibrium condition becomes

$$\begin{aligned} F\left( \frac{p_n-(1+\beta )p_u}{\left( 1-\frac{\mu }{\gamma }\right) x_n} \right) = 1-F\left( \frac{p_n-p_u}{(\gamma -\mu )x_n} \right) \end{aligned}$$

(8)

An increase in the rate of technological progress initially reduces the utility of the O option and increases the utility of the N option as illustrated in Fig. 4. The utility profile of options N and O shifts to $U^{N'}$ and $U^{O'}$. This leads to reductions in both $\underline{h}$ and ${\bar{h}}$ to levels $\underline{h}'$ and ${\bar{h}}'$. The first effect leads to reduced demand for secondhand cars, the second effect to increased supply. To maintain equilibrium, the price of used cars has to decrease. Specifically, we show in Appendix A.2 that

$$\begin{aligned} \frac{dp_u}{d\gamma } = - x_0 \frac{ f(\underline{h})\underline{h} \frac{\mu }{\gamma } +f({\bar{h}}) {\bar{h}} }{ f(\underline{h}) (1+\beta )(\gamma )+f({\bar{h}}) } <0 \end{aligned}$$

(9)

To see how the speed of technological progress $\gamma$ affects the size of the secondhand market, we can study $\frac{d\bar{h}}{d\gamma }$. As shown in Appendix A.2, the total effect of increased technological progress on the cut-off for always purchasing a new car can be written as

$$\begin{aligned} \frac{d{{\bar{h}}}}{d\gamma }&= -\frac{f(\underline{h})}{(\gamma -\mu )} \times \frac{ (1+\beta )\gamma {\bar{h}} -\frac{\mu }{\gamma } \underline{h} }{ f(\underline{h}) (1+\beta )\gamma +f({\bar{h}}) } \end{aligned}$$

As $(1+\beta )\gamma {\bar{h}} > \frac{\mu }{\gamma } \underline{h}$, we get $\frac{d{{\bar{h}}}}{d\gamma }<0$.

Fig. 4

The effect of technological progress on the utility of car ownership strategies. The dashed lines indicate the shift in utility resulting from an increase in the rate of technological progress $\gamma$. Subsequent price effects are not shown

The increase in $\gamma$ increases the utility of new cars and hence tends to reduce ${\bar{h}}$ as more consumers enter into the N segment. This is the effect illustrated in Fig. 4. To maintain equilibrium, the fraction of consumers who always buy used cars (U) increase equivalently, whereas the fraction of consumers who buy new cars and stick with them (K) declines. This is partially offset by a reduction in $p_u$ as it increases the relative price of a new car. This effect, however, is a consequence of the first effect and hence cannot dominate.

This finding is intuitive because the utility of owning a new car increases with technological progress and therefore more people than without technological progress decide to not stick with the car they bought for two periods, but rather decide to buy a new car each period. These are the consumers with quite high values of h, the utility weight on the car’s quality.

With the price decline for used cars, consumers who in the market without technological progress would buy a new car in the first period and stick with it for the next, decide to rather buy used cars in both periods. These are the consumers with quite low levels of h. The model therefore predicts larger turnover and a larger secondhand market for cars when there is large technological progress.

3.3 Multiple Vehicle Technologies in the Same Market

The market we study empirically consists of cars using different fuel types, where our main focus is the difference between electric and gasoline cars. In the model above, we model these as cars where technological progress is still ongoing (immature technologies) and models where technological progress to a larger extent has stagnated (mature technologies). Hence, the consumer has a choice not only between a new and a secondhand car, but also between models where technological progress differs. We do not present a full model of this case but discuss how the preceding framework can be used to understand it.

If the models with technological progress maintain a constant rate of technological improvement, they will eventually completely fill the market at the expense of cars without technological progress. A complete model would have to take this into account, modelling the technological progress as a transition towards a steady state with mature technology that remains at a constant level. We limit ourselves to an informal discussion.

To model that there are consumers preferring both types of cars, we extend the model to the utility function of driving a car with immature technology to

$$\begin{aligned} U=xh+c+\epsilon , \end{aligned}$$

where $\epsilon$, the net utility of the immature technology compared to the mature technology, has a distribution in the population. In general, $\epsilon$ can take on any real value indicating either a preference in the direction of mature or immature technologies. This term can be related to technology specific preferences, characteristics such as noise or emission level as well as social image associated with the technology. The utility of driving a car with mature technology remains the same. The prices of new cars of both types are given exogenously. We now find six choice sets for consumers. Consumers can purchase a new car every period and sell it in the secondhand market at the end of the period (N), purchase a used car in the secondhand market every period (O), or purchase a new car every second period and keep it for two periods (K). In all three cases, the consumer can buy a car with mature (m) or immature ($\mathop {m}\limits ^{\lnot }$) technology.

As above, an interesting experiment is to study the effect of increased technological progress $\gamma$ on the various prices. As for the case with only immature technologies, an increase in $\gamma$ makes it more attractive to purchase an $\mathop {m}\limits ^{\lnot }$ type car, moving consumers from $K_{\mathop {m}\limits ^{\lnot }}$ to $N_{\mathop {m}\limits ^{\lnot }}$. Second, the increase in $\gamma$ makes new $\mathop {m}\limits ^{\lnot }$ cars more attractive than new m cars, moving some consumers from $N_m$ to $N_{\mathop {m}\limits ^{\lnot }}$. Both effects increases the supply of secondhand $\mathop {m}\limits ^{\lnot }$ cars. The second effect also reduces the supply of secondhand m cars. To maintain equilibrium in the markets for secondhand cars, we need increased demand of secondhand $\mathop {m}\limits ^{\lnot }$ cars. Hence, the price of secondhand $\mathop {m}\limits ^{\lnot }$ cars goes down.

We should also see an increase in the price of secondhand m cars as their supply is reduced. Hence as long as electric cars undergo more rapid technological progress than gasoline cars, we expect the price of secondhand electric cars to decrease relative the price of secondhand gasoline cars.

4 Data, Descriptive Statistics and Empirical Strategy

4.1 Data

In our analysis we use a novel data set from the largest web platform for secondhand vehicles in Norway, finn.no, which accounts for close to 90% of the secondhand car market. This is the first time the data is used for this kind of analysis. Finn.no provided us with the data.

The data cover the period January 2011 to June 20 2021. We start our analysis in 2011 because the Nissan Leaf was introduced to the Norwegian market at the end of this year.⁹ Nissan Leaf being the first modern electric five-seat vehicle to be produced for the mass market from a major manufacturer, this corresponds to the start of the mass market for electric cars in Norway.

We analyze passenger vehicles sold as new in 2011 or later. This means that we only have vehicles that are 10 years or younger in our analysis. The total number of observations in the data is 5.5 million, yielding on average 500,000 ads for secondhand vehicles on finn.no each year. Around 2.9 million cars are older than 10 years and around 1.2 of the 2.6 million that are left have been advertized with less than 3 month in between, and we define them as duplicates and only use the newest ad. Then the total number of observations is 1.4 million. The vehicles that are gasoline or electric vehicles and 10 years or younger are in total around 545,000 observations. The number and shares of vehicles of the different fuels can be seen in Appendix Table 6.

In addition to the data from the online platform, we have list prices for new vehicles from The Norwegian Road Federation (OFV). The Norwegian Road Federation is a membership organization for car importers, transport companies and other actors within transportation, and have a role as producer of statistics about the car sale in Norway. They also gather price information from the car importers and delivers the official list prices to the tax authorities that use them to calculate wealth tax. The data from The Norwegian Road Federation also categorizes the vehicles into different body style categories.¹⁰

Further, we have information about the driving range of the electric vehicles, collected from the Norwegian Electric Vehicle Association’s website that informs about the different electric vehicle models (The Norwegian Electric Vehicle Association 2022a).¹¹ We also have access to a survey of Norwegian electric vehicle owners conducted by the Norwegian Electric Vehicle Association. There is around 15,000 respondents in the survey.¹² Additional details about the data and the variables can be found in Appendix B.

Notice that information on some variables is incomplete. From the platform data, we have the ask prices, but not the price the vehicle is actually sold for as there can be haggling over the prices. However, measurement error in the outcome variable is generally unproblematic as long as this is not a systematic measurement error. Further, we don’t know whether a transaction has actually taken place, but for most it probably has. Except for rather unlikely events, e.g., seller changes their mind and suddenly wanting to keep the car, a transaction is by far the most likely reason for the ad being removed.

Each ad is one observation each month. The average time to sell a car on finn.no is less than a month (The Norwegian Electric Vehicle Association 2021a). We remove duplicate data entries based on a combination of the chassis number, dealer type, effect, engine volume, fuel, main color, make, model, municipality of the seller, transmission wheel drive, year model, the year the vehicle is first registered and number of seats. If the same car is advertized with 3 months or less interval, it is defined as a duplicate. This means that if the same car is advertized two months in a row, only one of the ads is part of the analysis. We use the ad that is newest in the analysis, because the price of the newest ad is probably the closest to the deal price.

If the car is advertised with the same price more than three times, even if it is advertized with some time interval it is likely that the car is not sold at all. Therefore, observations that are duplicates and the price is the same for three ads are taken out of the analysis. Note that we have already taken out the observation of the analysis if a car is advertized with less than 3 months in between each ad, no matter if the price is the same or not.

Most cars come in different trims, which are different versions for each models with different equipment level. We do not have consistent data on trim levels of the model that is advertised. As the price of the new vehicle can vary a lot depending on trim, but also e.g. battery capacity, engine size and body style, the price of the new vehicle can have a large interval. We take this into account in one empirical specification by removing models where the relative difference between the maximum and the minimum price of one model exceeds 1.5. This does not change the result much.

4.2 Descriptive Statistics

In the analysis we compare electric vehicles and gasoline vehicles. Cars of both fuel types are maximum 10 years old. There are roughly 367,000 gasoline vehicles and 179,000 electric vehicles in the sample, which means that the share of electric vehicles in the sample is around 1/3 and gasoline vehicles around 2/3 (see Table 1). Among the gasoline vehicles, about 2/3 are small and 1/3 is large, while for electric vehicles there are fewer large vehicles available (11% of the sample), which is natural when we recall that the supply of large electric vehicles has been limited up until recently.¹³ In 2011 there are 32 times more gasoline vehicles as electric vehicles. By 2016, this number is reduced to 4, while there are more electric vehicles than gasoline vehicles sold used in 2021 (see Table 7 in Appendix C).

Table 1 Number of vehicles based on size

Table 2 The electric vehicle sample

Electric vehicles in the sample are younger than the gasoline vehicles. Average age is 2.7 for electric vehicles, while for gasoline vehicles the average age is 3.5, see more details in Table 8. The number of secondhand vehicles that are from 2011 or later that are advertised on finn.no every year can be seen in Table 7 in the Appendix. 56% of the electric vehicle sample has range below 200 km and only 16% 400 km or above (Table 2). The mean driving range each year is lower for the used vehicles than the new vehicles, which underlines the fast technological development, see Appendix Table 9.

The market for gasoline vehicles are more varied than the market for electric vehicles, which is natural when we recall that there are fewer electric models available than gasoline models. The secondhand electric vehicle market is dominated by Nissan Leaf with about 1/5 of the market. In Appendix Table 11 we see the top ten models in the secondhand market for electric and gasoline vehicles.

4.3 Empirical Strategy

A car i is sold new (time $t=0$) at price $P_{i0}$ which we treat as exogenous. We assume that the resale value follows a process so that at time t it is

$$\begin{aligned} P_{it}=A_i B_i^t P_{i0}^\mu e^{\epsilon _{it}}, \end{aligned}$$

(10)

where $A_i$ is a car specific characteristic to be detailed below. We assume that secondhand vehicle prices decay exponentially with an annual rate $1-B_i$. The parameter $\mu$ is the pass-through of the new price, and $\epsilon _{it}$ is the stochastic error term.

In order to find the annual percentage drop in prices, we take the logarithm of equation (10) which leads us to the following equation where lower case letters denote logs of variables:

$$\begin{aligned} p_{it}=a_i + \beta _i t + \mu p_{i0} + \epsilon _{it} \end{aligned}$$

(11)

The key parameter of interest is $\beta _i$, the annual percentage decline in prices.¹⁴ As we only observe each vehicle at one point in time (when it is sold in the secondhand market), the specific $\beta _i$ for each car is not observable. Instead we exploit that vehicles are sold at different ages in the secondhand market and estimate a fuel-specific $\beta _f$. The rate is potentially specific within other categories than the fuel, such as the make, the driving range of the electric car, the price category of the new vehicle, the size and the vintage.

The variable $a_i$ measures the technological level of cars of make i. Technological level and technological progress are difficult to measure directly. Rather we include a composite measure of several characteristics, including range for electric vehicles, mileage, make reputation, fuel and size. First, by including a fuel specific coefficient ($\gamma _f$) on log mileage, $m_{it}$, at the time the vehicle is sold in the secondhand market. Second, we add fixed effects for make ($\theta _m$), size (measured by body style or weight, $\zeta _s$) and fuel times year ($\alpha _{t\times f}$).¹⁵ As the market for electric vehicles are changing each year, it makes sense to have a different control for each fuel each year. For instance is the supply of fast charging stations increasing each year and we control for that with fuel times year fixed effects.

We also control for type of dealer by adding dealer fixed effects ($\omega _d$). There are three different dealer categories: private seller, professional importer of that specific brand and professional car sellers, and there might be selection into what cars are sold from which type of seller.¹⁶ Cars that are similar in age, fuel, make, model and mileage, but are sold by different types of dealers, might be different and we control for this. There is a statistically significant difference between the price fall of the different sellers, as shown in Table 22 in Appendix D, so we believe it is useful to control for the dealer type.

A vehicle i of make m with fuel type f and size category s sold in year t from a dealer of type d that was new in year n gets a price

$$\begin{aligned} p_{i,t}=\beta _f t + \alpha _{t\times f} + \theta _m + \zeta _s + \omega _d +\mu p_{i,n} + \gamma _f m_{i,t} + \epsilon _{i,t} \end{aligned}$$

(12)

The standard errors are clustered on the make (m).

We assume a linear annual percentage fall in prices in the main specification, but we also investigate the price fall with a squared age term. In addition we investigate the secondhand prices non-parametrically in Fig. 10.

If price at time n, $p_{i,n}$, is unknown for some reason, we introduce a dummy for missing variables. This implies simply following the price path downward without information of the initial price. We check whether the results are the same without the dummies (and therefore without the observations without new vehicle price) in Table 26.

We could also use the price on the same car model as new at the same time as when the used car is sold, rather than when the used car was new. However, the correlation coefficient between the price when the car was new and the price of the new car when the car was sold is above 0.9. Therefore we use the price when the used car is new.

5 Results

In this section we first estimate the effect age has on secondhand prices, both year-by-year price decline and average annual decline. Then we look at range on electric vehicles interacted with age. These are our main findings. In addition we have estimated quantiles instead of the mean for used vehicle prices. Further we have investigated different size categories, new vehicle price percentiles and vintage effects. We also compare electric vehicles with diesel and other fuels, before we turn to other factors affecting the price path. At the end of the chapter we discuss robustness.

5.1 The Price Fall of Electric Vehicles Compared to Gasoline Vehicles

In Fig. 5 we show the results from a regression with age dummies interacted with the fuel type. We control for mileage interacted with fuel, new vehicle price, a dummy on those with missing on mileage and new vehicle price, year $\times$ fuel fixed effects and dealer fixed effects. We see that the overall picture is that while the price of secondhand gasoline cars drops by about 60% after 10 years, the price fall for electric cars is even stronger at about 85%.

Fig. 5

The price path for gasoline and electric vehicles. The grey areas are 95% confidence intervals. The standard errors are clustered on make. There are 62 clusters. The regression includes year times fuel fixed effects and dealer fixed effects, and we control for mileage, vehicle price when new and whether the new vehicle price or mileage is missing

In Table 3, we show that this holds across a range of specifications. Here, different versions of regression equation (12) are shown. Gasoline vehicles is the baseline, and we see that the annual price drop for gasoline vehicles is approximately 10–11%. The price on electric vehicles fall between 4.7 and 6.4 p.p. more than gasoline vehicles, which is 16–17% annually. The difference between gasoline and electric vehicles is statistically significantly at least on a 5% level in all specifications, giving a clear indication that electric cars fall more quickly in price than gasoline cars.

In column (1) we control for year $\times$ fuel fixed effects, which absorb influences of all omitted variables that differ from one year to the next for each fuel type but are constant over all vehicles of the same fuel type in each year. This can for instance be development of fast charging stations or other year specific events for electric vehicles or for gasoline vehicles. In addition, we control for dealer fixed effects in all specifications, except in column (5), where we show the results without dealer fixed effects.

In column (2) we add body style fixed effects to year $\times$ fuel fixed effects.¹⁷ The difference between gasoline and electric vehicles does not change much compared to column (1).

In column (3) we add make fixed effects to the year $\times$ fuel fixed effects. The dummy variables for the makes absorb the influences of all omitted variables that differ from make to make, but are constant over time. The difference between gasoline and electric vehicles reduces by $1.3$  percentage points compared to column (1), indicating that the price fall varies between different makes and when only looking at the price fall within each make, the price fall on electric vehicles is lower (but slightly higher for gasoline vehicles). However, the difference in price fall between electric and gasoline vehicles is still statistical significant. We investigate the price fall for specific makes further in Appendix Fig. 14 and in Appendix Sect. D.1.

Table 3 Comparing gasoline and electric vehicles

In column (4), we take out the models that have a large variation in new vehicle prices between the different variants of the model. This reduces the sample from around 550,000 to around 340,000. It could be that the models that vary a lot in prices as new also have other characteristics that influence the price decline, so this is not our main specification, but a robustness test. We see that the coefficients do not change a lot and that the difference is still statistically significant.

In column (5), we show the results without controlling for dealer fixed effects. The difference between electric and gasoline vehicles stay in the same magnitude and remain statistically significant.

We see that mileage is not an important factor for the price of used gasoline vehicles, and even less important for electric vehicles, when age is taken into account. Each percentage increase in mileage is associated with a price fall of around 0.07% for gasoline vehicles. This might be due to age and mileage being closely correlated (see Appendix Fig. 9). For electric vehicles mileage is not statistically significant in any of the specifications in Table 3, but in other specifications shown in the Appendix.¹⁸ However, the magnitude of the effect is small.

Finally, the price decline appears to increase over time. The price fall curve depicted in Fig. 5 seems to be concave and in Table 12 in Appendix D we investigate whether the concavity is statistically significant by adding an age squared term. The concavity is statistically significant for both technologies, and electric vehicles are statistically significant more concave than gasoline vehicles.

5.2 Range and Age

One factor that has increased significantly over the 10 years we look into, is driving range. Range can be seen as a proxy for technological development. We investigate whether the price decline is different for different ranges. Figure 6 shows a version of Eq. (12) where the coefficient on age is modelled as a fourth grade polynomial interacted with range. The figure shows clearly that it is vehicles with range below 200 km that have the largest price decline. The secondhand prices on vehicles with range over approximately 200 km decline at the same level as gasoline vehicles (around 10% annually, see Table 3). However, over half of the electric vehicle sample is vehicles with range below 200 km. The vehicles with range from 200 km and upwards have mean age of 1.74 years, median of 1 year and the highest age is 8 years. We therefore do not have a sample where those with the longer range than 200 km have been in the car market for many years. This is therefore an important question for further investigation when the electric vehicle market has developed further.

Fig. 6

The price decline for electric cars with different driving range. This is the regression in Eq. (12) with a fourth grade polynomial of range interacted with age. Range is WLTP standard. If NEDC is the only available range, we reduce the range by a factor of 0.65 (Power 2020). The grey areas are 95% confidence intervals. We start the graph where the sample range start: 76 km range. There are 28 observations with 663 km range. The rest of the sample has a maximum of 564 km. Therefore the graph ends at 564 km. We do a robustness check for the factor converting between NEDC and WLTP and increase it to 0.75. This can be seen in Figure 11

In Fig. 7 we see that when we take out the electric vehicles with range below 200 km of the analysis, the difference in price decline between gasoline and electric vehicles is no longer present. In Appendix Table 13 we have the same sample, and the electric vehicles actually decline less in price than gasoline vehicles. However, the comparison between electric and gasoline vehicles might not be of same-sized cars, as electric vehicles with range below 200 km are small cars, while for gasoline vehicles the sample is all sizes.

Fig. 7

The price path for electric vehicles with range from 200 km and upwards and gasoline vehicles. The range is measured with WLTP standard. If NEDC is the only available range, we reduce the range by a factor of 0.65 (Power 2020). The grey areas are 95% confidence intervals. The standard errors are clustered on make. There are 59 clusters. The curve for electric vehicles stop at 8 years because there are no older electric vehicles with range from 200 km and more. The regression includes year times fuel fixed effects and dealer fixed effects, and we control for mileage, vehicle price when new and whether the new vehicle price or mileage is missing

5.3 Additional Findings

Our main study is on the marginal effect of age on prices by fuel type. It is also interesting to look at the whole distribution of prices conditional on age. To do so, we show conditional quantile regression on the 10th, 25th, 50th, 75th and 90th percentile in Appendix Table 14. The estimates show that the distribution of prices for gasoline cars conditional on age shifts downward with age. But as the coefficients at all percentiles are fairly similar, it seems the distribution of prices remain similar. For electric cars, the conditional distribution shifts downward and more strongly than for gasoline cars. Moreover, we see that the coefficients on age are numerically larger for the lower percentiles than the higher percentiles. This indicates that the distribution of prices is widening in age. As each of the reported estimates have non-overlapping confidence intervals at conventional levels of confidence, we can also reject statistically that the distribution remains unchanged. Notice that price and range are also highly correlated, but as gasoline cars don’t have a meaningful range this is not included in these regressions.

We compare different size categories based on either body style or weight in Appendix Table 15. Large electric vehicles do not have a statistically significant larger price fall than gasoline vehicles. Depending on how size is defined, it might seem that large electric vehicles have a smaller price decline than large gasoline vehicles. However, this is uncertain as an electric vehicle with the same size as a gasoline vehicle weigh more due to the batteries, but it is uncertain exactly how much more.

We investigate whether there are vintage effects. With vintage effects we mean that there are different price decline pattern for the vehicles that are new in the first part of the period (2011–2015) than the vehicles that are new in the second part (2016–2021). The results are shown in Table 16. The price decline of electric vehicles compared to gasoline vehicles in the first period is around 1 percentage point higher than in the second period (column 1), and within each make the difference between vintages is 1.5 percentage points.

We split the sample into different subgroups in order to investigate whether and how the price decline difference between the two technologies are present in different subgroups. We compare vehicles in different price categories when they were new in Appendix Table 17.¹⁹ For the vehicles that are priced when new equal to or above the median or equal to or above the 75th percentile the electric vehicles decline less in price than the gasoline vehicles. For vehicles that are below the median in new vehicle price the electric vehicles decline more than gasoline vehicles. This can be due to range, but also other factors. We see in Table 9 that there is a large difference in range for the cars that are above and below the median.

We also group the sample into different time periods based on when they were new (Appendix Table 18), when they were sold (Appendix Table 19) and different age groups (Appendix Table 20). The difference in price fall between gasoline and electric vehicles is statistically significant for all subgroups, except for those new in 2014–2016 and those sold in 2018–2021.

5.4 Diesel and Other Fuels

We compare electric vehicles with gasoline vehicles which is a mature technology. Diesel vehicles could also be included in the analysis, but the price path of diesel vehicles will not just represent vehicle technology that are mature, but a vehicle technology that policymakers want to phase out, to a larger degree than gasoline vehicles. Diesel cars had a large advantage from the Norwegian car tax system in 2007 to 2011, and the diesel share of the new car market was above 70% during this period. From 2012 the car tax system was changed in order to reduce the diesel share. Because other things than being a mature technology is important for the price path of diesel vehicles, they are not in the main analysis.

We present the results for diesel vehicles and the other fuels in Table 21 in Appendix D. This shows that electric vehicles fall more in price than all other fuel types, including diesel vehicles. However, the difference in price drop between diesel and electric vehicles is only statistically significant in some specifications.

5.5 Other Factors Affecting the Price Path

5.5.1 Battery Degradation

Degraded batteries or a worry of degraded batteries may be a reason for the price fall on electric vehicles. Degradation of batteries was a concern when the modern electric vehicles market had just started, but anecdotal evidence in the media and correspondence with experts in the sector point in the direction that rapid degradation, rendering batteries useless, is not a large problem (Kelleher Environmental 2019).²⁰ However, this is a question that warrants further investigation.

The warranty that are given by the car manufacturers can give some indication of the life length of batteries. Nissan has a warranty of 8 years for the vehicles sold in 2016 and onward,²¹ guaranteeing the batteries are 75% of its original state. From 2014 Tesla has a warranty that the battery maintain 70% of its original state for 8 years or a kilometer stand of 160,000–240,000 km depending on the model. Hyundai Ioniq guarantees 8 years/200,000 km. This indicates that the car sellers are confident about the battery lifetime.

Even though the batteries last long, a 25–30% degradation of the batteries might be enough to reduce the price of the vehicle, especially if the range is already quite low. Whether this is the case cannot be found with the data from finn.no. To investigate the battery question further, we have looked into a survey among Norwegian electric vehicle owners conducted by the Norwegian Electric Vehicle Association.²² For the years 2019–2021 they were asked whether the battery capacity in their car has become considerably lower since it was new.²³ Notice that the survey does not provide a clear definition of the term “considerably lower”. Still it gives an indication of the state of the battery. Respondents answer on a 1–5 Likert scale range, and the higher the score the better experience do the consumer have with the battery.

In Fig. 8 we see the share of respondents who experience that the battery capacity is considerably lower since the car was new (answering 1 or 2 on the survey question). When the car is between 0 and 3 years the share is around or below 10%. Between 4 and 6 years the share is lower than 30%. Between 7 and 8 years-old the share is between 20 and 45%. The 9 year-olds have shares between 50 and 60% and 10 year-olds down to 37%. However, the sample size is low for the 9 and 10 year-olds (below 100 for each year).

Fig. 8

Reported battery degradation by vehicle age. The figure shows the fraction of respondents who report that they experience that the battery has degraded (score 1 or 2 out of 5) since the car was new

The mean score for 5 year-old electric vehicles is 3.4 in 2021, 3.5 in 2020 and 3.8 in 2019, which means that the mean score has gone down from 2019 to 2021. This can be seen in Fig. 13 in Appendix D. The mean score for 8 year-old electric vehicles is 3.3 in 2021 and 2.9 in 2019 and 2020, which means that the mean score has gone up from 2019 to 2021. For nine year old cars the score is lower with 2.5 in 2021 and 2.8 in 2020, but again: the sample size is low.

From this information we cannot exclude the possibility that some of the price fall for electric vehicles is due to some degradation of the batteries, but the information does not indicate that degradation of the batteries drives the whole price fall on electric vehicles. However, we can not be conclusive on this matter. Another important point is that cars with short range has to be charged more often than cars with longer range. Therefore, as charging reduces the quality of the battery, the cars with short range might have more degraded batteries and as the battery technology improves, battery degradation might also reduce.

5.5.2 Maintenance Cost

Gasoline vehicles typically have higher maintenance cost when they are older. Besides the battery degradation, this is not the case for electric vehicles, though.²⁴ Therefore, without technological change and battery degradation, electric vehicles should have a higher price on the secondhand car than gasoline vehicles if the two technologies are otherwise equal.

5.5.3 Supply Side Constraints

There have been—and still are—supply side constrains on electric vehicles, both that the models that fit the preferences of different consumers are not available and that there are waiting lists on popular electric vehicles. The number of individuals on the waiting list that actually end up buying the car when given the possibility is not public information. The waiting lists might also be part of a marketing strategy, as simply raising the price on the car would seem to be a more straightforward approach. We don’t know if and how the supply side constrains affect the secondhand car market, but we can assume that it had a positive effect on the price in the secondhand electric car market. Hyundai Kona, for instance, has been advertised with higher price on finn.no than the price on the new vehicle, but we do not know whether the vehicle is sold for this price. If waiting lists have influenced the used car market, the effect is probably that the price fall on electric vehicles is lower than without the waiting lists. This means that without supply side constraints the price fall on electric vehicles would probably be higher, especially on large electric vehicles.

5.6 Robustness

We check whether the main results hold under other specifications than in Table 3. The difference in price fall between gasoline and electric vehicles is statistically significant in all robustness checks, except with absolute values instead of log values when we include make fixed effects. We use all variables available, such as transmission, wheel drive and main color. In addition, instead of using the price when the vehicle was new as control, we use the new car price the year the vehicle was sold used. Especially for electric vehicles the price of new vehicles might change in a non-linear way because of new models entering the market and reducing prices on existing models (The Norwegian Electric Vehicle Association 2021b). The price fall difference when we do not control for dealer fixed effects and at the same time control for new vehicle price when the vehicle was sold is very high (Table 23, column 5), indicating that the private sale of electric vehicles and gasoline vehicles is different. We also test whether the main results hold when we drop the observations that are 0 years old when they are sold on finn.no. We also take out the dummies on those that miss the control variables new vehicle price and mileage (which result in a smaller sample). In addition we do the regression with absolute values, instead of log values. We also show the results when clustering the standard errors differently, both with make $\times$ fuel and make $\times$ the year the vehicle is new. The standard errors are as expected somewhat smaller with more clusters, but the statistical significance stay at the same level. The robustness checks are reassuring for the main result. See more details in Appendix E.

6 Conclusion

In this article we have investigated whether the price of electric vehicles decline faster than the price of gasoline vehicles. We have found empirical evidence that they do, but only those with range below 200 km. Our hypothesis is that the faster price fall on electric vehicles compared to gasoline vehicles is mostly due to faster technological development. As the adoption of electric vehicles is increasing, this is relevant information for more consumers in more markets. The valuation of electric vehicles that are ten years or younger in 2021 is a snapshot that might change as the improvements of the technology of electric vehicles flattens out. Comparing the results of this analysis with an analysis that is done 2–10 years later will be interesting. It will especially be interesting to see how the price path of the electric vehicles with longer range than 200 km develops. As the electric vehicles with range from 200 km and upwards is still young, we need to wait some more years to get the full picture of the price path of electric vehicles with longer range. In addition, it would be interesting to investigate which factors influence the vehicle price and the price fall, for instance using causal forest approaches. Further, the price path of other technologies with rapid improvements would be interesting to investigate, for instance solar power and smart mobile phones.

Our findings do not directly answer whether buying an electric vehicle is an economic sensible choice. Even if electric vehicles with less range than 200 km have fallen more in price the last decade than gasoline vehicles, there are other factors than value loss that influence the total cost of ownership of a vehicle. Electric vehicles have lower maintenance cost and lower fuel cost compared to vehicles with internal combustion engine. The batteries can be reused for other purposes and the car can be recycled (Kelleher Environmental 2019). In addition, electric vehicles are subsidized in many countries, making new technology cheaper for consumers (IEA 2021).

The maturity of a technology might depend on the adoption rates in combination with other factors (Pathak et al. 2022, p. TS-128-129). Therefore the findings of our paper should not be used as an argument for delaying the deployment of new low-carbon technologies. Rather, the findings may be used to inform about the cost of implementing climate policies. The value of the secondhand products matters for the total cost of ownership of the good, which again matters for the cost of climate policy. The Norwegian government, for instance, has a goal that “the purchase of zero-emission cars should be more economically favourable than the purchase of conventional cars” (Meld. St. 33:2016–2017). When comparing whether it is favourable to buy an electric vehicle compared to a gasoline vehicle, it is important to take into account also the difference in the value of the used car, not just the price of the new car. This point can be transferred to other low-carbon technologies with fast technological progress. At the same time, many actors are working on finding new business models for batteries that are no longer in use in a car (see for instance Evyon 2022). This indicates that even if products with fast technological development deteriorates faster, they are not without value.

Appendices

Appendix

The Effect of Technological Progress

1.1 Opting Out of Car Ownership

In the main analysis we assume that all consumers need a car, hence implicitly that the utility of not owning a car is sufficiently low. In many settings, other means of transportation can be a suitable alternative. To include this in the model, we can introduce an outside opportunity of not owning a car yielding a utility level $\underline{U}$. This alternative is most relevant for consumers with a low car preference parameter h. Hence this alternative is chosen for consumers with $U^O<\underline{U}$. This yields a new cut-off level $\mathop {h}\limits _{\sim }$ defined by

$$\begin{aligned} x_u \mathop {h}\limits _{\sim }-p_u + \beta (x_u \mathop {h}\limits _{\sim }-p_u) = \underline{U} \Rightarrow \mathop {h}\limits _{\sim } = \frac{\underline{U}}{(1+\beta )x_u} + \frac{p_u}{x_u} \end{aligned}$$

As the number in the N and O groups still have to be equally sized in equilibrium, this yields a new equilibrium condition

$$\begin{aligned} F\left( \frac{p_n-(1+\beta )p_u}{(1-\mu )x_n} \right) - F\left( \frac{\underline{U}}{(1+\beta )x_u} + \frac{p_u}{x_u} \right) = 1-F \left( \frac{p_n-p_u}{(1-\mu )x_n} \right) \end{aligned}$$

This change to the model does not change any qualitative insights from the model, so to maintain simplicity we disregard the presence of an outside opportunity for the rest of the paper.

1.2 Proof of Theoretical Results

To see the effect of technological progress on the market, we can differentiate equation (8) to get

$$\begin{aligned}{} & {} f(\underline{h}) \left[ -\frac{1+\beta }{\left( 1-\frac{\mu }{\gamma }\right) x_n} dp_1 - \frac{[p_n-(1+\beta )p_u] \frac{\mu x_n}{(\gamma )^2} }{\left[ \left( 1-\frac{\mu }{\gamma }\right) x_n \right] ^2} d\gamma \right] \nonumber \\{} & {} \quad = -f({\bar{h}}) \left[ -\frac{1}{(\gamma -\mu )x_n} dp_1 -\frac{(p_n-p_u)x_n}{\left[ (\gamma -\mu )x_n \right] ^2} d\gamma \right] \end{aligned}$$

(13)

where the left hand side are the demand effects and the right hand side the supply effects. Solving, we find

$$\begin{aligned} \frac{dp_1}{d\gamma }&= - x_0 \frac{ f(\underline{h})\underline{h} \frac{\frac{\mu }{(1+\gamma )^2}}{1-\frac{\mu }{\gamma }} +f( {\bar{h}}) {\bar{h}} \frac{1}{(\gamma -\mu )} }{ f(\underline{h}) \frac{1+\beta }{1-\frac{\mu }{\gamma }} +f({\bar{h}}) \frac{1}{\gamma -\mu } } \end{aligned}$$

(14)

$$\begin{aligned}&= - x_0 \frac{ f(\underline{h})\underline{h} \frac{\mu }{\gamma } +f({\bar{h}}) {\bar{h}} }{ f(\underline{h}) (1+\beta )(\gamma )+f({\bar{h}}) } <0 \end{aligned}$$

(15)

Moreover, we have

$$\begin{aligned} \frac{d{{\bar{h}}}}{d\gamma } = -\frac{(p_n-p_u)x_n}{(\gamma -\mu )^2x_n^2} - \frac{1}{(\gamma -\mu )x_n} \frac{dp_u}{d\gamma } \end{aligned}$$

Using equation (8), we find

$$\begin{aligned} \frac{d{{\bar{h}}}}{d\gamma }&= -\frac{1}{(\gamma -\mu )} \left( {{\bar{h}}} - \frac{ f(\underline{h})\underline{h} \frac{\mu }{\gamma } +f({\bar{h}}) {\bar{h}} }{ f(\underline{h}) (1+\beta )\gamma +f({\bar{h}}) } \right) \\&= -\frac{f(\underline{h})}{(\gamma -\mu )} \times \frac{ (1+\beta )\gamma {\bar{h}} -\frac{\mu }{\gamma } \underline{h} }{ f(\underline{h}) (1+\beta )\gamma +f({\bar{h}}) } \end{aligned}$$

As $(1+\beta )\gamma {\bar{h}} > \frac{\mu }{\gamma } \underline{h}$, we then get $\frac{d{{\bar{h}}}}{d\gamma }<0$.

Data

In this Appendix, we present details of the data we use in our analysis.

1.1 Variables from finn.no

The data from finn.no forms the basis of our analysis. It includes a hashed vehicle id number, chassis number, the listed secondhand price of the vehicle the seller wants, the year the vehicle is first registered, year model, fuel type, mileage, make and model, number of seats, municipality of the seller, dealer type, engine effect, engine volume, transmission, wheel drive, main color, as well as the month and year of an ad on finn.no.

We will now go through the variables that we use in the analysis.

1.1.1 Vehicle Id Number

Each vehicle has an id number which is unique and encrypted to keep the data anonymous. We combine this variable with the chassis number, dealer type, effect, engine volume, fuel, main color, make, model, municipality of the seller, transmission wheel drive, year model, the year the vehicle is first registered and number of seats to check whether a vehicle has been advertised for more than a month (and therefore becomes a duplicate observation) and to a lower price.

1.1.2 Vehicle Age

To find the age of a vehicle, we can use two variables to tell us the year the vehicle was new: The year model and the year the vehicle was first registered. The year model in itself is an imprecise term. Some manufacturers start selling year model t early in year $t-1$, and some maintain the year on the model until they change the model.

However, year of first registration can also be wrong. Although our data ends in 2021, there are observations with year of first registration date after 2021. There are also numerous observations with year of first registration before 1940. Although some of these may be antique cars, some are clear mistakes. There are for instance multiple Nissan Leaf presumably registered before 1940, which is unreasonable as the model entered the market after 2010.

Our solution is to use the lowest of the year of first registration and the model year whenever both are present. Moreover, we assume that vehicles first registered before 1940 is a mistake, and in this case we use the year model unless the year of first registration and year model is the same year, we keep the year of first registration.

As we have data on the year the vehicle is advertised on finn.no, we find the age of the vehicle by subtracting the year the vehicle is advertised with the year the vehicle is new.

1.1.3 Fuel Type

The data from finn.no has data on fuel type. However, there may be mistakes by the seller, for instance claiming that a vehicle is a gasoline vehicle when it is a hybrid gasoline vehicle or stating that it is an electric vehicle when it is a hybrid or plug-in hybrid. When there is only one observation of a vehicle model one year that is electric, the electric model is not listed by OFV, and we do not recognize the model name as an electric vehicle (there are so few electric vehicle models that it is possible to know most of them), we take them out of the analysis, because it is likely that they are erroneously classified as electric vehicles. This is the case of 41 observations.

Gasoline vehicles that are not found on the OFV list are treated differently, see more elaboration on this in Sect. B.2.1.

The finn data do not distinguish between plug-in hybrid and hybrid. To distinguish between the two, we also use data from OFV. If a vehicle is listed as gasoline-hybrid in the finn data and plug-in gasoline hybrid or a gasoline vehicle in the OFV data, we code it as a plug-in gasoline hybrid. If a vehicle is listed as hybrid in the finn data and hybrid and plug-in hybrid in the OFV data, we do not change it. As most of our analyses compare electric and gasoline vehicles this coding scheme has a minor impact on our results. In Table 21, however, we compare all fuel types.

1.1.4 The Price of the Secondhand Vehicle

We only have the price the seller lists on the finn platform, not the price the vehicle is actually sold at. There can be cases where the buyer and the seller agree on a lower price than the listed one.

Moreover, the price is without re-registration fee which varies through the period of investigation and for different vehicles, depending on age and fuel.

1.1.5 Mileage

We use mileage as a control variable because this is a variable that probably affect the price of the vehicle. Finn has rounded the number to the nearest thousand in order to anonymize the data. Observations where mileage is missing are controlled for with a dummy variable. See Table 5 to see the number of missing on mileage.

1.1.6 Number of Seats

This variable is used to distinguish between passenger vehicles and utility vehicles with the same model name.

1.1.7 Municipality of the Seller

We use the municipality of the seller as one of the variables to detect duplicates.

1.2 Variables from OFV

The Norwegian Road Federation (OFV) data are merged into the finn data set by fuel, model, make and the year the vehicle is new. The variables from OFV that we use are the price on the new vehicle and body style category.

The data from OFV are combined with the data from finn.no matching on the following variables: make, model, fuel and the year the vehicle is new.

1.2.1 Price on the New Vehicle

Norwegian Road Federation (OFV) collects prices on new vehicles from all importers that are formally linked to the producer. The price list of June every year is used to find the price on the new vehicle. When not found in the June list, we use the November list or the December list. If not found on either June, November or December list, the vehicle is marked as missing price on the new vehicle (see below for more information on the reason for the vehicle missing on the price list). See Table 5 to see the number of missing on new vehicle price.

There are many different equipment levels within the same model and the information from finn.no does not tell us which equipment level each vehicle has. Some equipment levels within a model are a lot more expensive than the others. Therefore, we take the median price of the make, model, fuel and year. For robustness check we take out those models where the ratio between the maximum price and the minimum price is above 1.5. This can be seen in Table 3, column 4.

Body style category should also be one of the variables that the median price is based on, but since finn.no do not have this variable, we can not merge on this variable. Some models are more than one type of category. For instance Ford Focus can be bought as a hatchback or station wagon. Mercedes-Benz A-class is either sedan or hatchback. Mercedes-Benz C-class is either sedan or station wagon. Then the median price is the median of all the models in the two types of vehicles body style categories.

Notice that the new prices we use are listed new prices. Actual prices of new cars could be lower due to e.g. haggling, rebates or campaigns.

Those observations with make or model “others” or “N/A” and fuel “N/A” when the model can be both electric and conventional vehicle are not merged and therefore lack the price on the new vehicle.

Some observations on finn.no are not found on the OFV list of new vehicle prices. There are four main explanations: (1) privately imported new vehicles from the factory, i.e. not through the importer that is related to the producer, (2) models from earlier years, (3) import of used vehicles, and (4) mistakes in the OFV list or at finn.no, for instance the advertiser is not aware that the vehicle is a hybrid and list it as a gasoline vehicle.

An example of explanation (1) is Fiat 500 Plug-in Hybrid which was produced for the American market. Fiat decided not to sell these vehicles in Europe, but some were imported through channels that were not related to the producer’s importer network.

The price on the new vehicle in category (2) is difficult to know. We can find the price on the model in earlier years, but the price probably changes from year to year.

For category (1), the price on finn.no is strictly speaking the price of the new vehicle in the Norwegian market if the car is 0 years-old. All vehicles that are not part of the OFV list get a dummy and the dummy is controlled for in the regression in Table 3. We also have the same specifications as in Table 3 without a dummy on those missing new vehicle price and that can be seen in Table 26.

1.2.2 OFV Listed Fuel Type

In the beginning of the period of investigation there were some models that were flexible fuel vehicles (cars that are able to use both gasoline and ethanol). OFV have them on their list, but as the market for these vehicles never took off in Norway, finn.no do not distinguish between gasoline vehicles and flexible fuel vehicles. We define all flexible fuel vehicles in OFV data as gasoline vehicles.

1.2.3 Size Categories

There is no clear definition of different size categories. The EU Commission lists different segments in the car market, but also state that “the exact market definition was left open”(Commission of the European Communities 1999, p.2).

In the USA the Environmental Protection Agency (EPA) has different size classes based on body style combined with passenger and cargo volume (fueleconomy.gov; n.d.). In contrast, the National Highway Traffic Safety Administration (NHTSA) in the USA categorize by class and weight (National Highway Traffic Safety Administration; n.d.). A third option in the USA is by The Insurance Institute for Highway Safety (IIHS) and The Highway Loss Data Institute (HLDS) which both are organizations funded by auto insurers and insurance associations. They use a combination of weight and length times width (IIHS and HLDI 2020).

1.2.3.1 Body style

Most models have different variants. Some models are in two different body style categories. For instance, Volkswagen Golf can be both a hatchback or a station wagon. However, for electric vehicles the different variants of a model are mostly in the same body style categories. For gasoline there are several models that can be both a hatchback and a station wagon. We categorize them based on which body style categories there are most model variants of.

We use the body style categories that The Norwegian Road Federation (OFV) categorize different models into.²⁵ We have translated their definition.

“Cabriolet are cars with roofs that can be opened or removed, either by folding it into the body style, or being completely removed. Roadsters should have body style category cabriolet.

Hatchbacks are cars with a large tailgate/rear door that include the rear window. The rear edge of the rear window ends where the body ends, without any horizontal surface/trunk lid. The hedge on the hatchback may have different inclination. The trunk is relatively small compared to the total volume of the car. The station wagon has larger volume, while the body shape may otherwise be quite similar.

Coupes are cars that have prioritized appearance at the expense of space and practical considerations. They are characterized by what most people will perceive as "elegant lines". Often, the coupes are lower with a flatter rear window than ordinary hatchbacks and sedans, which the space, especially from the front seats and backwards, suffers underneath. The tailgate is hinged under the rear window. Coupes can have 2 and 4 doors.

Pickups are cars with an open loading plane behind a cab. The cab can contain several rows of seats.

Sedan are cars with preferably 4 doors and separate trunk and marked trunk lid that are hinged under/behind the rear window.

Station wagon are cars with full height backwards to the tailgate. The tailgate can be somewhat sloping and the trunk is open towards the cabin. They are characterized by large trunk in relation to the total volume of the car.

Multipurpose car are cars that are designed for maximum interior space utilization and flexibility. The body style shape is similar to that of the station wagon or hatchback, but the multipurpose cars are usually higher than these, albeit with normal ground clearance. They have a minimum of 5 seats in addition to luggage space.”

SUV is not defined by OFV.

1.2.3.2 Weight

Weight can be used as a proxy for size. We take the median of the different trims of the same model. As electric vehicles are heavier than gasoline vehicles due to the batteries, we use the gasoline vehicles as the basis to find the cutoff between large and small vehicles. We use the 75th percentile of the weight for all gasoline vehicles in the sample as the cutoff. This is 1382.5 kg. How much more an electric vehicle weigh due to the batteries is difficult to know, but the tax authorities use 15% extra weight for plug-in hybrids from 1st of January 2022 and before that they used 23% (The Norwegian Tax Administration; n.d.). We therefore add 10%, 20% and 30% to the electric vehicle weight when distinguishing between large and small vehicles based on weight in Table 15.

1.3 Variables from other Sources

1.3.1 Range

Range is taken from the webpage of the Norwegian Electric Vehicle Association (The Norwegian Electric Vehicle Association 2022a). When the same model the same year has different range alternatives, we construct the median of the range. The range we use is the WLTP, which is the new standard. If the old standard, NEDC, is the only available, we reduce the driving range by a factor of 0.65 (Power 2020). Some models are for sale with different year models with different range at the same time and therefore we do not know for sure the range of the car the year the vehicle is new. We use the median range of the year model the year the vehicle is new.

1.4 Sample Selection

Vehicles that are new in 2011 or after is part of the analysis. We therefore do not include the observations for vehicles that are new earlier than 2011. See Tables 4 and 5 for an overview of the number of observations that are not part of the analysis.

In order to not let outliers influence the result, we analyze vehicles with a secondhand price range between 10 000 NOK and 2 million NOK, see Table 4 for the number of observations. In addition, we take out Think and Buddy cars as these can not be characterized as modern vehicles. Observations where the recorded vehicle age is negative at the time of transaction are also removed.

Table 4 Observations not part of the analysis

Table 5 Number of observations with some of the variables missing

Descriptive Data

See Tables 6, 7, 8, 9, 10 and 11, Fig. 9

Table 6 Number of vehicles based on fuel

Table 7 Number of vehicles that are new in 2011 or later advertised each year

Table 8 Number of vehicles based on how old they are when they are for sale

Table 9 The mean range in the year the vehicles are new and sold in the secondhand market. The median refers to the price of the vehicle as new

Table 10 Share of vehicles based on seller

Table 11 Top 10 secondhand models

Fig. 9

Binned scatterplot of the relationship between mileage and age

More Detailed Results

The relationship between vehicle age and secondhand price is illustrated in Fig. 10. As expected, the price is declining in age. We also see that this decline is clearly stronger for electric vehicles than for gasoline cars (Tables 12, 13, 14, 15, 16, 17, 18, 19, 20 and 21, Figs. 11, 12 and 13).

Fig. 10

Secondhand vehicle prices by age and fuel type. Each point represent an equal number of observations. Results control for mileage, the log price of the new car and dummies for missing variables. The y-axis is on a logarithmic scale

Fig. 11

The price decline for electric cars with different driving range, using 0.75 instead of 0.65 as converting rate between the range standard NEDC and WLTP

Table 12 Table 3 with age squared

Table 13 Table 3 without the electric vehicles with range below 200 km

Table 14 Quantile regression

Table 15 Different size categories based on either body style or weight

Table 16 Investigating vintage effects

Table 17 Different price categories

Fig. 12

The price path of electric vs gasoline vehicles, both with price of the new vehicle over the median. The grey area is the 95% confidence interval. The standard errors are clustered on make $\times$ fuel

Table 18 The cars grouped into when they where new

Table 19 Comparing 0–3-year-old used vehicles sold in different time periods

Table 20 Comparing different age groups

Table 21 Table 3 with all fuels and electric vehicles as the baseline

Fig. 13

Mean of the answers to the question about whether the battery has degraded since the car was new. The higher the better experience of the battery from 1 to 5

1.1 Makes

There is a large difference in the average price fall between the different makes, which can be seen in Fig. 14. This can be due to different vintage or age of the cars, since electric Nissan entered the market in 2011, while electric Jaguar entered the market in 2018, but there can also be other reasons. There is a pattern where makes with large price fall have small cars with four seats (Smart, Fiat, Mitsubishi, Peugeot, Citroën).²⁶ Further, the majority of Ford’s and Renault’s models cannot fast charge, and the price fall of these cars is also above average of electric vehicles.²⁷

The majority of the used cars sold from Tesla, Opel, Jaguar, Audi and Polestar have range above 400 km, and these makes have less than 10% price fall. Polestar has a price increase. Hyundai, which has a price fall in the same magnitude as Opel, has a model with range above 400 km (Hyundai Kona with range 449 km), but the sample also include Hyundai Ioniq which has 180 km²⁸-311 km.

We also make this plot controlling for the new vehicle price when the car was sold, instead of when the car was new, in Appendix Fig. 15. This changes the estimate for some of the makes, for instance Tesla.

Fig. 14

The average price fall for electric vehicles of different makes. The red lines mark the average price fall for electric vehicles ($-$0.173) and the average price fall for gasoline vehicles ($-$0.106), based on the result in column 1 in Table 3. The grey line mark 0

Fig. 15

Price fall for different makes controlling for the price of the new model when the car was sold, not when the car was new. The red lines mark the average price fall for electric vehicles ($-$0.173) and the average price fall for gasoline vehicles ($-$0.106), based on the result in column 1 in Table3. The grey line mark 0

We compare the price path of some specific models and makes. The most sold secondhand electric model, Nissan Leaf, compared to the most sold secondhand gasoline model, Volkswagen Golf, (see Table 11) can be seen in Fig. 16.

Fig. 16

Price path of the most sold used electric vehicle Nissan Leaf, compared to the most sold used gasoline vehicle, Volkswagen Golf. The grey area is the 95% confidence interval. The standard errors are robust

In Fig. 17 we compare gasoline vehicles where the new car price more than the median price of new vehicles with Tesla in the same price range. We see that Tesla fall less in price than gasoline vehicles.

Fig. 17

The price path of Tesla vs gasoline vehicles. The grey area is the 95% confidence interval. Here we compare the Teslas and gasoline vehicles that both have new vehicle price over the median. The standard errors are robust

Robustness

See Tables 22, 23, 24, 25, 26, 27, 28 and 29, Fig. 18

Fig. 18

Binned scatter plot of the data with prices in absolute values

Table 22 Robustness check 1: controlling for many variables

Table 23 Robustness check 2: Table 3 with new vehicle price the year the vehicle is sold, not the year the vehicle is new

Table 24 Robustness check 3: Table 3 with new vehicle price and fuel interacted

Table 25 Robustness check 4: Table 3 without those with age=0

Table 26 Robustness check 5: Table 3 without dummy on those that have missing new vehicle price and mileage, meaning that the sample is smaller

Table 27 Robustness check 6: Table 3 with absolute values instead of ln values

Table 28 Robustness check 7: Table 3 with clusters based on make $\times$ fuel

Table 29 Robustness check 8: Table 3 with clusters based on make $\times$ year the vehicle is new

.