The Effect of Early Electric Vehicle Subsidies on the Automobile Market

Substitution of New Technologies

Research in marketing, public policy, and technology management has examined the adoption of innovations for decades (e.g., Fosfuri and Giarratana 2009; Kalish and Lilien 1983; Lin, MacInnis, and Eisingerich 2020; Morvinski, Amir, and Muller 2017; Van Heerde, Srinivasan, and Dekimpe 2010). In doing so, the lion's share of work focuses on factors that can accelerate or attenuate rates at which new innovations are adopted. More recently, researchers have begun exploring how and when new products with emerging technologies substitute for incumbent ones (Adner and Snow 2010; Greenwood et al. 2017). Findings indicate that the emergence of new technologies does not necessarily ensure the abandonment of incumbent technologies, the reason often being heterogeneity in demand and preferences among the consumer base (Adner and Snow 2010; Adner and Zemsky 2006). Inherently, this suggests that the market not only enables incumbent technologies to respond to the emergence of new technologies (Adner and Snow 2010) but also creates competition between old and new. For example, research examining disruptive innovations has widely observed incumbent and emerging technologies sharing the market for nontrivial amounts of time before the incumbent is displaced by the faster-evolving disruptor (Adner 2002; Christensen 1997).

EVs are often cast as a promising innovation and are widely considered disruptive (Adamson 2005). This characterization holds water. EVs offer notable benefits over internal combustion vehicles in terms of GHG emissions (Avci, Girotra, and Netessine 2015), maintenance costs (Sioshansi and Denholm 2009), and operating costs (Lajunen 2014; Simpson 2006). Yet, there are concerns about their viability, including range and charging infrastructure (Carley et al. 2013; Egbue and Long 2012; He et al. 2020; Lim, Mak, and Rong 2014), all of which undermine the competitiveness of EVs. Further, although numerous studies have explored the factors affecting EV adoption, how and when EVs will begin to substitute for and materially displace traditional vehicles is yet unclear. Indeed, do EVs and traditional vehicles even compete in the same market? Does public investment through financial subsidies facilitate the process of technology substitution? Although some scholars (e.g., Adamson 2005) have proposed analytical frameworks to broach these questions, such frameworks have received limited empirical attention.

Governmental Policy Stimulating EV Purchase

In the face of increasing public concern about the ramifications of climate change, numerous governments have sought to accelerate the adoption of EVs to limit the emission of GHGs. To date, three primary policy mechanisms have been used: publicization (Bakker and Trip 2013; Carley et al. 2013), nonpecuniary incentives (Ewing and Sarigöllü 2000; Hackbarth and Madlener 2013; Li et al. 2017), and financial incentives (Huse and Lucinda 2014; Zhang, Xu, and Zhang 2016). We discuss each next.

Regarding publicization, surveys consistently indicate that consumers are wildly misinformed regarding EVs. Nearly two-thirds of people lack basic knowledge (including advantages, disadvantages, and the existence of government incentives) and rarely have direct knowledge of actual vehicle costs (Carley et al. 2013; Lane and Potter 2007). The main thrust of these policies is to provide consumers with information and educate them on the differences between EVs and traditional vehicles, with the long-term goal of stimulating purchase (Bakker and Trip 2013; Krause et al. 2013).

Nonpecuniary incentives (e.g., support for charging infrastructure, access to high occupancy vehicle lanes, free parking) have also been used. In China, for example, EVs are often exempted from registration lotteries (Hao et al. 2014; Zhang, Xu, and Zhang 2016). Yet, extant literature is notably muddy about the effect of these incentives on adoption (Hackbarth and Madlener 2013; Li et al. 2017; Mersky et al. 2016). Although Langbroek, Franklin, and Susilo (2016) observe that incentives like free parking and access to bus lanes are appealing to potential purchasers, Mersky et al. (2016) find that access to bus lanes and toll road waivers are not significant predictors of EV adoption.

The most pertinent stream of research to this work, financial subsidy, is clearer in its effect. Two common incentives, purchase subsidies (Huse and Lucinda 2014; Zhang, Xu, and Zhang 2016) and tax rebates (Beresteanu and Li 2011; Chandra, Gulati, and Kandlikar 2010; Gallagher and Muehlegger 2011; Jenn, Azevedo, and Ferreira 2013), have been extensively used. Overall, evidence lends support for the significant effect of financial incentives on EV adoption (Beresteanu and Li 2011; Chandra, Gulati, and Kandlikar 2010; Gallagher and Muehlegger 2011; Huse and Lucinda 2014; Jenn, Azevedo, and Ferreira 2013), although the effect is sensitive to location and subsidy-specific factors.

What remains notable is that researchers have yet to delve into the source of the increasing demand for EVs after subsidies are put in place. On the one hand, consumers switching to EVs might reduce purchasing of traditional vehicles (i.e., cannibalization). Such an effect is intuitive if customers are single-sourcing vehicles for their own consumption. On the other hand, an expanded EV market may increase the overall market sales (i.e., expansion). To the extent that nonowners might select into ownership, or families might add an additional EV to their resource stocks, it is plausible that the traditional vehicle market may be unaffected. As these competing logics correspond to consumers’ different incentives to purchase EVs, we next discuss EV purchase decision making.

Consumers’ Decisions Making of EV Purchase

To investigate how EV purchasing subsidies influence the automobile market (i.e., market cannibalization or market expansion), we first discuss consumers’ decision making when purchasing EVs. Prior literature has examined how various factors shape consumer EV preferences (Diamond 2009; Gallagher and Muehlegger 2011; Sierzchula et al. 2014). Broadly, this stream of work can be grouped into three categories: financial and operational factors (Beresteanu and Li 2011; Carley et al. 2013; Egbue and Long 2012), consumer characteristics (Barth, Jugert, and Fritsche 2016; Hidrue et al. 2011; Sierzchula et al. 2014), and social influence (Axsen and Kurani 2011, 2013; Zhang et al. 2013).

Table 1 summarizes the key findings of prior studies. Although financial subsidies relieve consumers’ concern of the high sticker price of EVs, other factors (e.g., consumer characteristics, social pressure) still play important roles in influencing consumers’ EV adoption. Prior research on green product adoption provides some insights into how these factors jointly influence consumer purchase behavior. For example, both Ling (2013) and Hsu, Chang, and Yansritakul (2017) show that consumers’ price sensitivity significantly moderates the relationship between their intentions to purchase green products and their environmental consciousness. Similarly, Shao and Ünal (2019) find that consumers’ willingness to pay a price premium for EVs is strongly influenced by information about the environmental impact their decisions may have, as opposed to the social impact of green products. However, these studies have yet to address the tension in consumers’ incentives and resource limitations to invest in EVs as additional vehicles or to purely transition from traditional vehicles to EVs.

OPEN IN VIEWER

Table 1.

Factors That Influence Consumer EV Preferences.

Category	Key Findings
Financial factors	High sticker price of EVs, as compared with traditional vehicles, is a major barrier to adoption (Achtnicht, Bühler, and Hermeling 2012; Carley et al. 2013; Hackbarth and Madlener 2013). The impact of total cost of ownership is mixed: some scholars find that the cost of electricity is not enough to offset the higher purchase price (Prud’homme and Koning 2012; Wu, Inderbitzin, and Bening 2015), while others find the exact opposite (Al-Alawi and Bradley 2013; Propfe et al. 2012). Petrol prices are a key driver of EV adoption (Beresteanu and Li 2011; Diamond 2009; Gallagher and Muehlegger 2011).
Operational factors	The vehicle range is a major consideration delaying EV adoption (Carley et al. 2013; Egbue and Long 2012; Lim, Mak, and Rong 2014). Adequate charging infrastructure alleviates the concern of vehicle range (Lim, Mak, and Rong 2014; Mak, Rong, and Shen 2013). Public charging stations increase EV adoption (Li et al. 2017).
Consumer characteristics	Early EV adopters are likely to be highly educated, environmentally sensitive, interested in new technology, and knowledgeable about fuel economy (Carley et al. 2013; Tran et al. 2013). Conflicting work waffles over whether previous experience with EVs, environmental awareness, gender, or fuel availability affect adoption (Barth, Jugert, and Fritsche 2016; Zhang et al. 2017). Neither education nor income influences adoption (Sierzchula et al. 2014).
Social influence	Vehicle images of intelligence, responsibility, and support of the environment stimulate interest in EVs (Axsen and Kurani 2013). Public awareness of pro-EV policy may positively affect willingness to pay (Zhang et al. 2013). Social acceptability and word of mouth also have a positive effect on market share (Noori and Tatari 2016), while negative word of mouth via social media yields the opposite (Hsu, Li, and Lu 2013). The influence from friends, family, and coworkers plays an important role in participants’ assessment of the technology (Axsen and Kurani 2011).

Market Cannibalization

For consumers affected by purchase subsidies and shifting from traditional cars to EVs, EVs work as substitutes for traditional vehicles. To the extent that the price of EVs is a major barrier to adoption (Achtnicht, Bühler, and Hermeling 2012; Carley et al. 2013; Hackbarth and Madlener 2013), it is plausible that subsidies might help remove this barrier by placing EVs within financial reach. Indeed, research would support such an effect, and some consumers say so (Randall, Halford, and Sam 2019). This process, wherein demand for one product switches to another when two products are substitutes, is known formally as cannibalization (Copulsky 1976; Mason and Milne 1994).

Indeed, prior scholarship has long investigated how green products cannibalize the market of incumbent products from both demand and supply perspectives (Yenipazarli and Vakharia 2015; Zhang et al. 2020). Market cannibalization is also common in the automobile industry (Van Heerde, Srinivasan, and Dekimpe 2010), with price being implicated as a critical factor (Meredith and Maki 2001; Ballardin 2005). Examples are easy to come by. In 2012, U.S. sales of Ford's newly introduced Focus, up 59.7%, largely came at the expense of the older Fiesta model, down 18.0% (Pope 2012). The fact that such dynamics would emerge in the EV market is therefore intuitive. To the degree that a large diversity of cars are forgone when consumers acquire EVs, simply because of economic constraints (Randall, Halford, and Sam 2019), the effect is evident.

The impact of financial incentives on consumers’ decisions to switch from traditional vehicles to EVs may be influenced by other factors. Consumers who are more aware of the benefits of EVs over traditional cars will be more motivated to purchase EVs after financial subsidies. For example, EVs are a widely heralded component of the green technology movement, producing only half the GHG emissions and air pollution of traditional vehicles (Granovskii, Dincer, and Rosen 2006). Consumers who are environmentally conscious will be more likely to respond to subsidies by switching from traditional cars to EVs.

Market Expansion

It is possible that EV purchase subsidies may encourage new purchasers who otherwise would not have purchased a vehicle. If this is the case, the resultant increase in EV purchasing might expand the automobile market instead of cannibalizing the traditional vehicle market. Previous literature lends credibility to such a claim. Huang et al. (2013) show analytically that the introduction of EV subsidies does not lower profits in traditional vehicle markets, suggesting that the demand for EVs may come from new purchasers or additional purchases. Further, research suggests that consumers are more likely to choose EVs when purchasing an additional vehicle, all the while maintaining a traditional automobile as their primary vehicle (Holtsmark and Skonhoft 2014). Thus, it is plausible that purchase subsidies may expand the automobile market overall, rather than cannibalizing existing sales.

Market expansion has been observed and investigated in several contexts, including the automobile industry. Digital distribution channels, for example, expanded the publishing industry (Bounie et al. 2013), ridesharing platforms expanded the automobile market by attracting consumers and stimulating vehicle sales (Gong, Greenwood, and Song 2017), and secondary markets often expand supply chains by stimulating an affinity for cross-product purchasing (Ghose, Smith, and Telang 2006; Ghose, Teland, and Krishnan 2005). Firms also strategically brand, or delay, the introduction of new products in order to expand the overall market (Reddy, Holak, and Bhat 1994; Wang and Hui 2005). In the focal context, little is known about the proportion of customers giving up vehicles versus expanding ownership (Randall, Halford, and Sam 2019). Thus, the subsidy might expand the market in two ways: it may incentivize first-time buyers to purchase an EV (Willems 2019) or push consumers with sufficient means to purchase an additional vehicle (Winegarden 2018).

For the consumers who otherwise would not have purchased a vehicle, their EV purchasing decision making involves comparing ownership of an EV with not owning a vehicle at all, instead of comparing the differences between traditional vehicles and EVs. And prior scholarship suggests that price cuts (either as a result of a subsidy or a pricing change by the firm) are able to attract new consumers without cannibalizing the existing market (McColl, Macgilchrist, and Rafiq 2020; Reimers and Xie 2019).

In the presence of these competing logics, we address the research questions empirically. The two mechanisms proposed previously, market cannibalization and market expansion, are not mutually exclusive. Indeed, both or neither might manifest as a result of purchase subsidies. The purpose of this work is to empirically explore which dominates when subsidies are put in place. Such a determination is critical to the understanding of consumer decision making in the early stage of a nascent market, especially when the market is far from a saturation point in terms of overall car ownership. It is also critical for policy makers who have explicitly invoked EV subsidies as a means to reduce the consumption of traditional vehicles, with the long-term goal of reducing GHG emissions. Finally, given the varied motivations of buyers, empirical exploration of the effect permits us to investigate how different socioeconomic and environmental factors moderate the influence of purchasing subsidies. By examining the moderating role of operating costs, consumer characteristics, and environmental factors, we can better inform policy makers during the crafting of subsequent incentives. Figure 1 contains our framework.

OPEN IN VIEWER

Figure 1.

Theoretical Framework.

Data

To quantify the effect of EV subsidy on the automobile market, we exploit the phased rollout of China's private purchase Electric Vehicle Subsidy Scheme (EVSS) between 2010 and 2015. Like other major automobile markets (e.g., United States, Germany), China views EV adoption as a direct means to reduce traditional vehicle purchases and tackle GHG emissions by facilitating the spread of EVs that can substitute for traditional vehicles (Hao et al. 2014). The goal of the EVSS was to stimulate the diffusion of EVs during the market's infancy before being slowly reduced and eliminated (Perkowski 2018). As of July 2022, the EVSS had three planned phases (see Table 2). ¹ Phase I, from June 2009 to December 2012, covered private purchases in some pilot cities. Phase II, from September 2013 to December 2015, covered both public and private purchases. Phase III ran from 2016 to 2022. During Phase III, the subsidy declined and the directory of subsidized EV models was revised with stricter battery performance standards.

OPEN IN VIEWER

Table 2.

Private Purchase Subsidy in EVSS Phase I and Phase II.

Phase	Phase I		Phase II
Year	2010–2012		2013		2014–2015
Subsidy	PHEV	¥3000/kW h (at most ¥50,000)	PHEV	¥35,000 (R ≥ 50)	PHEV	Reduce 5% on the subsidy amount of 2013 in 2014. Reduce 10% on the subsidy amount of 2013 in 2015.
	BEV	¥3000/kW h (at most ¥60,000)	BEV	¥35,000 (80 ≤ R < 150) ¥50,000 (150 ≤ R < 250) ¥60,000 (R ≥ 250)	BEV	Reduce 5% on the subsidy amount of 2013 in 2014. Reduce 10% on the subsidy amount of 2013 in 2015.
	FCEV	Not covered	FCEV	¥200,000	FCEV	¥200,000

Notes: R denotes electric vehicle range (km). PHEV = plug-in hybrid electric vehicle; BEV = battery electric vehicle; FCEV = fuel cell electric vehicle.

Our empirical investigation focuses on Phase II (note that results are consistent if Phase I cities are included) for several reasons. First, only five cities were treated by Phase I of the EVSS and were considered pilots by the government. Second, there is a nine-month gap between the end of Phase I and the start of Phase II, during which no EVSS policy was active. Third, after excluding cities with a purchase restriction on automobiles (discussed subsequently), only two cities from Phase I remain. In contrast, Phase II expanded the EVSS to 63 cities in November 2013, and another 25 cities were added in February 2014. ² Finally, as the government intended to reduce the subsidy at the outset of Phase III of the EVSS, the theoretical question of how subsidies affect purchasing begins to change. The timing of EVSS for each city is collected from public documents and is compiled in Web Appendix A.

We begin in 2010 because the charging infrastructure was notably accelerated at that time (Bombourg 2012). By the end of 2015, the ratio of public charging facilities to plug-in electric cars in China was .27, higher than that in the United States and Japan in the same period (Ou et al. 2017). Given the importance of charging infrastructure to EV adoption (Lim, Mak, and Rong 2014; Mak, Rong, and Shen 2013), the 2010 to 2015 window enables us to investigate the early diffusion of EVs at a time when a charging infrastructure was not a material obstacle for buyers.

EVs, broadly, can be grouped into four categories: plug-in hybrid electric vehicles (PHEVs), battery electric vehicles (BEVs), fuel cell electric vehicle (FCEVs), and hybrid electric vehicles (Tie and Tan 2013). Phase II covers PHEVs, BEVs, and FCEVs. As indicated in Table 2, there are two changes in Phase II as compared with the previous phase. First, in Phase II, FCEVs are included in the scope of private purchase subsidy. However, there was no available FCEV model for sale during the sample. Second, in Phase II, a range restriction was imposed to increase the substitutability of the adopted EVs (see Table 2). This restriction affects only seven out of the 114 models in our data set, meaning that the scope of EV models covered by Phase I and Phase II did not vary significantly.

To empirically assess the effect of Phase II, we examine new private vehicle purchases between 2010 and 2015. Data on passenger car registration are organized by prefecture-level city ³ at the month level. Due to data availability, imported vehicle purchases are not included. Imported cars account for less than 7.5% of the passenger car market during the sample and are mostly composed of high-end luxury vehicles (Wagner, An, and Wang 2009). Thus, lacking data on imported vehicles is of little concern. Cities from Phase I are also excluded. ⁴ We omit all cities (i.e., Shanghai, Beijing, Guiyang, Guangzhou, Tianjin, Hangzhou, and Shenzhen) with vehicle purchase restrictions (e.g., registration auctions or lotteries), as such restrictions place an artificial limit on purchasing. After omitting cities with no EV sales (where there is no variation in the dependent variable), 222 cities remain.

Variables

Controls

We include a robust set of controls. We collect annual city-level demographic data from the China City Statistical Yearbook, including economic- and transportation-related characteristics. Regarding economic characteristics, we include population, gross region product (GRP) per capita, and average wage. We also include transportation-related controls (i.e., log-transformed metro length, number of buses, number of taxis, and length of paved roads) within the city. Finally, we collect monthly gasoline price data for each city. Prior work suggests that fuel price is strongly correlated with EV adoption (Beresteanu and Li 2011; Diamond 2009; Gallagher and Muehlegger 2011), as it affects the operating costs of traditional vehicles. Gasoline price is lagged to the previous month, to capture the effect of consumers’ perceptions of operating costs, and is log-transformed.

Because the central Chinese Ministry of Industry and Information Technology selected the cities to receive the EVSS, it is worth discussing the selection process. Prior work indicates that city selection was predicated on two factors: a willingness of the city to partially match funding from the central government, and the elimination of license plate controls for EV vehicles (Hao et al. 2014). Priority was then given to cities with more severe air pollution (Hao et al. 2014). Although these factors are largely time invariant and should be absorbed by the location fixed effects (discussed subsequently), they underscore the need to include economic controls (which capture the ability of the city to allocate matching funds to the subsidy) and vehicle use/traffic data (which directly influences air pollution). Summary statistics are in Table 3. Detailed statistics are in Web Appendix B.

OPEN IN VIEWER

Table 3.

Summary Statistics and Correlations.

Variables	Mean	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)
1. ln(NumEVs)	.238	.826
2. ln(NumNonEVs)	7.738	.389	.974
3. Subsidy	.111	.611	.376	.315
4. ln(Buses)	6.566	.221	.718	.239	1.019
5. ln(Taxis)	7.427	.151	.613	.195	.822	.952
6. ln(Population)	6.000	.163	.576	.143	.355	.334	.631
7. ln(GRPPerCapita)	10.435	.235	.44	.281	.490	.349	−.249	.628
8. ln(RoadArea)	7.021	.286	.779	.261	.826	.714	.439	.521	.860
9. ln(Wage)	10.434	.286	.377	.324	.257	.220	.135	.357	.306	.439
10. ln(MetroLength)	.161	.274	.439	.269	.500	.449	.287	.284	.487	.246	.782
11. ln(GasolinePrice)	1.888	−.295	−.015	−.137	.147	.135	.011	.067	.094	−.227	.008	.439

Notes: Standard deviations are on the diagonal.

Market Cannibalization or Market Expansion

To determine the effect of the EV subsidy on both the EV and traditional vehicle markets, we employ a difference-in-differences (DID) estimation. Doing so enables us to exploit the phased rollout of the EVSS into different cities at different times, mimicking an experimental design using observational data (Angrist and Pischke 2009; Bertrand, Duflo, and Mullainathan 2004). In doing so, we leverage the variation in availability and timing of EVSS across different cities to identify the effect of EVSS. Formally,

ln ( NumEVs ) it = α i + θ t + β 1 Subsid y it + β 2 Control s it + ε it ,

(1)

ln ( NumNonEVs ) it = τ i + ϕ t + γ 1 Subsid y it + γ 2 Control s it + ϵ it ,

(2)

Results with EV and traditional vehicle sales as the dependent variables are in Table 4. In each, we start with a base DID and then introduce controls. As can be seen in Column 1, the EVSS had a significant and positive effect on EV sales, corroborating prior work. Results are robust to the inclusion of controls. The coefficient on Subsidy_it in Column 2 suggests that the EVSS is associated with a 241% (= e^1.227 − 1) increase in EV sales. In contrast, we observe no material effect of the EVSS on sales of traditional vehicles (Columns 3 and 4, Table 4). The size of the non-EV coefficient suggests that any effect is so trivial as to be outright dismissed. In summary, results suggest that the EVSS has expanded the EV market without significantly shifting demand from the traditional vehicle market. This is critical, because it indicates that the key objective of the EVSS (i.e., shifting demand) is not being realized.

OPEN IN VIEWER

Table 4.

DID Model.

	(1)	(2)	(3)	(4)
Variables	ln(NumEVs)	ln(NumEVs)	ln(NumNonEVs)	ln(NumNonEVs)
Subsidy	1.254***	1.227***	.0414	.0125
	(.114)	(.114)	(.0298)	(.0292)
ln(Buses)		−.104		.0459
		(.0880)		(.0297)
ln(Taxis)		−.0479		.0146
		(.0511)		(.0252)
ln(Population)		.481		.296
		(.342)		(.171)
ln(GRPPerCapita)		−.243		.0286
		(.172)		(.0465)
ln(RoadArea)		.0461		.0309
		(.0994)		(.0396)
ln(Wage)		−.0769**		.0269
		(.0291)		(.0145)
ln(MetroLength)		.236**		−.00677
		(.0887)		(.0164)
ln(GasolinePrice)(t-1)		−.967**		−.0292
		(.346)		(.111)
Observations	15,762	14,089	15,762	14,089
R-squared	.523	.551	.956	.958
Year-month FEs	Yes	Yes	Yes	Yes
City FEs	Yes	Yes	Yes	Yes

*p < .05.

**p < .01.

***p < .001.

Notes: FEs = fixed effects. Robust standard errors clustered at the city level in parentheses. The number of observations varies by column because some control variables have missing values for some cities during some months. Boldface highlights the main estimators in our study.

Heterogeneous Treatment Intensity

In the main analysis, EVSS treatment is operationalized as a dummy variable. This enables us to estimate the overall effect of EVSS on the automobile market and simplifies the interpretation of any such effect. However, to the degree that the treatment intensity may vary across cities and be correlated with time-varying underlying changes implemented via local policies to complement the EVSS, the potential for an omitted variable bias exists.

To safeguard against this possibility, we next adopt a treatment measure developed by the International Council on Clean Transportation, best exemplified in Cui et al. (2018). This measure captures levels of EVSS intensity (SubsidyIntensity_it), with the goal of quantifying the total benefit EV purchasers in i,t. Thus, the measure takes into account both the fiscal subsidy stemming from the EVSS and local policies. In the measure, all benefits (e.g., traffic control elimination, tax reductions, license plate access, parking fee reductions, charging fee discounts) are transformed into a single index (based on 10,000 Chinese yuan). Details regarding the construction of this measure are in Web Appendix C.

We replicate the main analyses by replacing Subsidy_it with SubsidyIntensity_it in Equations 1 and 2. Results are in Table 5. The coefficient of SubsidyIntensity in Column 1 suggests that if an additional 10,000 Chinese yuan per car is subsidized to buyers, EV sales increase by 13.77%. The coefficient of SubsidyIntensity in Column 2 indicates an insignificant effect of EVSS on traditional vehicles, corroborating overall expansion with no material cannibalization observed.

OPEN IN VIEWER

Table 5.

DID Model: Heterogeneous Subsidy Intensity Across Cities.

	(1)	(2)
Group	Considering Local Policies
Variables	ln(NumEVs)	ln(NumNonEVs)
SubsidyIntensity	.129***	.00132
	(.0120)	(.00308)
Controls	Yes	Yes
Observations	14,089	14,089
R-squared	.551	.958
Year-month FEs	Yes	Yes
City FEs	Yes	Yes

*p < .05.

**p < .01.

***p < .001.

Notes: FEs = fixed effects. Robust standard errors clustered at the city level in parentheses. Boldface highlights the main estimator in our study.

Count Model

In our main analysis, we log-transformed the number of EVs and traditional vehicles as the dependent variables. This allows us to have a distribution that is closer to Gaussian when leveraging an ordinary least squares estimator. Moreover, the log transformation enables us to interpret the coefficient of the subsidy as an elasticity. However, our results may be sensitive to the choice of estimator. To address these concerns, we apply a negative binomial estimation using the raw number of EVs and traditional vehicles as the dependent variables. The benefit of this approach is that we can observe if there is an absolute shift in the market, rather than an elastic shift, thereby making the disparities between the two markets irrelevant. This also enables us to sum EV and non-EV sales and, by extension, examine if the existence of the subsidy is strictly expanding the entirety of the automobile market (i.e., do we see actual expansion of the complete vehicle market?). If the market is not strictly expanding, it would undermine any claim that cannibalization is not occurring.

Results in Table 6 show an increase in EV sales with no change in traditional vehicle sales, consistent with the main results. Further, Column 3 indicates that the overall market grows once the subsidy is introduced (i.e., the sum of EV and non-EV vehicle sales increases). This suggests that expansion strictly dominates cannibalization and refutes the possibility that disparities in market size drive the observed effect.

OPEN IN VIEWER

Table 6.

DID Model: Fixed-Effects Negative Binomial Model.

Variables	(1)	(2)	(3)
	NumEVs	NumNonEVs	NumEVs + NumNonEVs
Subsidy	.748***	.0126	.0171**
	(.0834)	(.00660)	(.00660)
Controls	Yes	Yes	Yes
Observations	13,901	14,089	14,089
Year-month FEs	Yes	Yes	Yes
City FEs	Yes	Yes	Yes

*p < .05.

**p < .01.

***p < .001.

Notes: FEs = fixed effects. Robust standard errors clustered at city level in parentheses.

Additional Checks

We conduct a series of robustness checks and falsification tests to ensure the robustness of the result. Specifically, we apply (1) a relative time model to test the parallel time trends assumption for the DID method, (2) Goodman-Bacon (2018) decomposition to examine whether the staggered treatment timing might bias the estimation of the overall effect, (3) a coarsened matching method to match treated cities with control cities based on observables, (4) alternative samples, (5) models including city linear time trends, (6) a placebo test using data on prior years, and (7) a shuffled treatment test. These tests are summarized in Table 7. These tests corroborate the fact that such subsidies accelerate new technology adoption but yield little effect on traditional markets. A full description is in Web Appendix D.

OPEN IN VIEWER

Table 7.

Summary of Robustness Checks.

Concern	Analysis	Findings
The parallel time trends assumption does not hold	Relative time model	The pretreatment coefficients are insignificant, suggesting the assumption of parallel time trends holds
The different treatment timing might bias the estimation of the overall effect	Goodman-Bacon DID analysis	Our results do not suffer from the “negative weights” problem due to the variation in treatment timing
Bias due to a linear model	Replicating an estimation using the negative binomial model	Our findings remain consistent
Disparity between cities in the treatment and control groups	Replicating an estimation after matching cities	Our findings remain consistent
The information of Phase I is missing	Alternative samples by including Phase I data	Our findings remain consistent
Within-city temporal dynamics are affecting our results	Replicating an estimation including city linear time trends	Our findings remain consistent
The results are an artifact of an idiosyncrasy associated with cities in EVSS	Placebo test using data on prior years	The results are not a function of the structure of the rollout
Serial correlation in standard errors	Shuffled treatment test	Serial correlation in the residuals is unlikely to be driving the effect

Fuel Prices

As discussed, financial considerations are one of the main factors influencing EV adoption (Carley et al. 2013; Diamond 2009). Researchers often focus on the high sticker price of EVs (Carley et al. 2013; Hackbarth and Madlener 2013), but fuel price has also been indicted by the scholarly community (Beresteanu and Li 2011; Diamond 2009) because the price of gas strongly affects the operating costs of traditional vehicles. From a pecuniary perspective, operational costs are a major difference between an EV and a traditional vehicle. It is possible that the EV subsidy will cause cannibalization of traditional vehicles in regional markets with higher gasoline prices, notably if consumers are eager to end their dependence on fossil fuels. To address this question, we explore the price of gasoline at a local level.

To execute these tests, we divide the cities in our sample into three buckets based on tercile gasoline price: GasolinePriceTercileLevel1, GasolinePriceTercileLevel2, and GasolinePriceTercileLevel3. This is done according to the median gasoline price in each city during the sample. Cities in the group of GasolinePriceTercileLevel1 have the lowest median price and those in GasolinePriceTercileLevel3 have the highest median price. We then replicate the DID model while interacting the subsidy indicator with the group indicators. The base terms for each of the gasoline price groups are omitted, as they are perfectly predicted by the city fixed effects.

Results in Table 8 show that there is a significant effect of the Subsidy on EV sales, and no significant effect on non-EV sales. Further, as seen in Column 1, the coefficients of the interaction terms indicate no significant difference between the Level 1 base case and the other two. Intuitively, this suggests that the operating costs of traditional vehicles weigh little on consumers' minds when making the decision to purchase an EV in the presence of a subsidy. Further, in Column 2 there is no significant cannibalization effect, as fuel prices increase for traditional vehicles. Again, this suggests that although consumers might say that operating costs are important when making a purchasing decision (Axsen, Mountain, and Jaccard 2009; Mau et al. 2008), operating costs do not appear to moderate the effect of the EVSS on traditional vehicle sales.

OPEN IN VIEWER

Table 8.

Moderating Effect of City Gasoline Price Level.

	(1)	(2)
Variables	ln(NumEVs)	ln(NumNonEVs)
Subsidy	1.284***	.010
	(.187)	(.035)
Subsidy × GasolinePriceTercileLevel2	−.503	−.089
	(.263)	(.059)
Subsidy × GasolinePriceTercileLevel3	.278	.092
	(.257)	(.048)
Controls	Yes	Yes
Observations	14,089	14,089
R-squared	.560	.958
Year-month FEs	Yes	Yes
City FEs	Yes	Yes

*p < .05.

**p < .01.

***p < .001.

Notes: FEs = fixed effects. Robust standard errors clustered at city level in parentheses. Cities are grouped by the gasoline price level in each city. The cutoffs are quantiles of the median of gasoline price at city level. Higher level number indicates higher gasoline price level. Specifically, we have the following three groups: GasolinePriceTercileLevel1: median of GasolinePrice < 6.8093; GasolinePriceTercileLevel2: 6.8093 ≤ median of GasolinePrice < 6.92; GasolinePriceTercileLevel3: median of GasolinePrice ≥ 6.92. Base is the GasolinePriceTercileLevel1.

Wages, Education, and Private Car Ownership

Several consumer characteristics have also been studied in prior literature; for example, demographic information (i.e., age, gender, education, income), number of cars owned, commute, and personal experience with EVs (Barth, Jugert, and Fritsche 2016; Gallagher and Muehlegger 2011; Zhang et al. 2013; Zhang, Yu, and Zou 2011). Mixed results are found. Whereas Sierzchula et al. (2014) find that neither education nor income influences adoption, Hidrue et al. (2011) observe that income reduces the likelihood to adopt EVs. Other work is conflicted over whether previous experience with EVs, environmental awareness, or fuel availability affect the adoption decision (Barth, Jugert, and Fritsche 2016; Zhang et al. 2017); meanwhile, surveys indicate that early EV adopters are likely to be highly educated, environmentally sensitive, interested in new technology, and knowledgeable about fuel economy (Carley et al. 2013; Tran et al. 2013).

We therefore examine how consumers’ reactions to EV purchase subsidies vary among cities with different levels of wages, educational attainment, and private car ownership. These analyses of moderating effects enable us to see details of the trade-off between the two possible outcomes of the EVSS—market cannibalization and market expansion—and how this trade-off is moderated by consumer characteristics. We again group cities into terciles based on (1) average wage, (2) education attainment (i.e., the ratio of residents with a bachelor's degree or above), and (3) private car ownership (i.e., the average number of owned cars per capita). The sources for the first two variables are the China City Statistical Yearbook and the Chinese 2010 census, respectively. The number of vehicles registered in each city in 2010 is used as a proxy for city-level private car ownership. We then replicate the DID model, interacting the subsidy indicator with these group indicators. Results are in Tables 9 and 10.

OPEN IN VIEWER

Table 9.

Moderating Effect of Consumer Characteristics (Wage and Education).

	(1)	(2)	(3)	(4)
Variables	ln(NumEVs)	ln(NumNonEVs)	ln(NumEVs)	ln(NumNonEVs)
Subsidy	1.137***	.107**	.752***	.107
	(.149)	(.0351)	(.146)	(.0555)
Subsidy × WageTercileLevel2	−.209	−.0632
	(.262)	(.0578)
Subsidy × WageTercileLevel3	.331	−.151***
	(.214)	(.0435)
Subsidy × EducationTercileLevel2			.357	−.119
			(.262)	(.0680)
Subsidy × EducationTercileLevel3			.817***	−.131*
			(.208)	(.0618)
Controls	Yes	Yes	Yes	Yes
Observations	14,089	14,089	12,930	12,930
R-squared	.557	.958	.570	.959
Year-month FEs	Yes	Yes	Yes	Yes
City FEs	Yes	Yes	Yes	Yes

*p < .05.

**p < .01.

***p < .001.

Notes: FEs = fixed effects. Robust standard errors clustered at the city level in parentheses. Boldface highlights the main estimators in our study. Higher level number indicates higher wage level and educational attainment level. Cutoffs of the wage level are 34,102.71 and 40,423.14. Cutoffs of the educational attainment are .019 and .03. Base is the first level.

OPEN IN VIEWER

Table 10.

Moderating Effect of Consumer Characteristics (Private Car Ownership).

	(1)	(2)	(3)	(4)	(5)	(6)
Variables	ln(NumEVs)	ln(NumNonEVs)	ln(NumEVs)		ln(NumNonEVs)
Subsidy	.624***	.106	1.208***	.534	.0312	.00813
	(.180)	(.0740)	(.112)	(.357)	(.0277)	(.0630)
Subsidy × OwnedPrivateCarTercileLevel2	.219	−.0971
	(.257)	(.0826)
Subsidy × OwnedPrivateCarTercileLevel3	.974***	−.121
	(.232)	(.0790)
Subsidy × ProvinceOwnedPrivateCar				6.919†		.236
				(3.662)		(.627)
ProvinceOwnedPrivateCar			9.804**	4.762†	−9.567***	−9.739***
			(3.737)	(2.502)	(1.235)	(1.305)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Observations	14,089	14,089	14,089	14,089	14,089	14,089
R-squared	.568	.958	.553	.557	.960	.960
Year-month FEs	Yes	Yes	Yes	Yes	Yes	Yes
City FEs	Yes	Yes	Yes	Yes	Yes	Yes

^†

p < .10.

*p < .05.

**p < .01.

***p < .001.

Notes: FEs = fixed effects. Robust standard errors clustered at the city level in parentheses. Boldface highlights the main estimators in our study. Higher level number indicates higher level of private care ownership. The cutoff points are 36.2 and 74.16 (unit: number of private cars/10,000 residents). Base is OwnedPrivateCarTercileLevel1.

Column 1 of Table 9 shows that the effect of subsidy on EV sales is not moderated by average wage. Traditional vehicle sales, on the other hand, are more likely to decline in cities with higher wage levels (Column 2), implying that market cannibalization is more likely to occur among consumers with greater income. This suggests that though we do not find an overall market cannibalization effect, traits of cannibalization are more likely to manifest as the wages of consumers rise (at least in the aggregate). The moderating effects of education are similar (Columns 3 and 4, Table 9). Results again corroborate the positive effect of subsidy on EV sales, and also suggest a stronger effect among consumers with higher levels of education. Moreover, although the subsidy has little effect on aggregate traditional car sales, we do witness a modest cannibalization effect among more educated populations. Given that the purpose of EV purchase subsidies is to encourage consumers to switch from traditional cars to EVs, our findings suggest that the effectiveness of the EV purchase subsidy is positively associated with educational attainment.

Results for the moderating role of private car ownership are in Table 10. Column 1 indicates that the effect of subsidies on EV sales is stronger in cities with higher levels of private vehicle ownership than in cities with lower levels of ownership. The impact of subsidies on traditional vehicle sales, however, is not moderated by private car ownership (Column 2). One possible explanation for these findings is that consumers are likely buying EVs as a second car. Results in Column 1 support this notion. To further examine the impact of private car ownership on the sales of EVs and traditional vehicles, we collect the province-level private car ownership data from 2010 to 2015 and include ProvinceOwnedPrivateCar, a continuous variable in the models of Equations 1 and 2. Results are in Columns 3 and 4 of Table 10, and they indicate that EV sales are significantly and positively correlated with ProvinceOwnedPrivateCar, whereas traditional vehicle sales are significantly and negatively correlated (Columns 5 and 6). This strongly suggests that consumers’ motivations for buying EVs and traditional cars are different. This distinction may explain why the implementation of EVSS results in market expansion rather than market cannibalization.

Environmental Factors

Our final moderators are environmental factors, which serve as an important nonpecuniary aspect of consumers’ purchase decision making of green products (Chen and Chai 2010). Vehicle images of intelligence, responsibility, and support for the environment often stimulate interest in EVs (Axsen and Kurani 2013), and prior work has found that early ecoproduct consumers are often environmentally conscious (Carley et al. 2013; Garvey and Bolton 2017; Tran et al. 2013). According to the daily air pollution index collected from the National Meteorological Information Center, China's air quality is among the global worst. ⁵ Because air quality, especially severe pollution, can influence the environmental awareness of consumers (Palmer et al. 1998), it is plausible that the effect of the EVSS might be stronger in cities with more severe air pollution given that these consumers are confronted with the issue of air pollution on a daily basis.

To investigate whether the severity of local air pollution moderates the effect of the EVSS, we again group cities into terciles based on the median-city-level air pollution index: AirPollutionTercileLevel1, AirPollutionTercileLevel2, and AirPollutionTercileLevel3. Cities in AirPollutionTercileLevel1 have the best air quality while those in AirPollutionTercileLevel3 have the worst. Results are in Table 11.

OPEN IN VIEWER

Table 11.

Moderating Effect of City Air Pollution Level.

	(1)	(2)
Variables	ln(NumEVs)	ln(NumNonEVs)
Subsidy	.882***	.036
	(.138)	(.047)
Subsidy × AirPollutionTercileLevel2	.563*	−.084
	(.260)	(.054)
Subsidy × AirPollutionTercileLevel3	.597*	−.001
	(.252)	(.061)
Controls	Yes	Yes
Observations	14,066	14,066
R-squared	.559	.958
Year-month FEs	Yes	Yes
City FEs	Yes	Yes

*p < .05.

**p < .01.

***p < .001.

Notes: FEs = fixed effects. Robust standard errors clustered at the city level in parentheses. Boldface highlights the main estimators in our study. Cities are grouped by the AirPollutionIndex level in each city. The cutoffs are quantiles of the median of AirPollutionIndex at city level. Higher level number indicates higher pollution level. Specifically, we have the following three groups: AirPollutionTercileLevel1: median of AirPollutionIndex < 65.2; AirPollutionTercileLevel2: 65.2 ≤ median of AirPollutionIndex < 77; AirPollutionTercileLevel3: median of AirPollutionIndex ≥ 77. Base is the AirPollutionTercileLevel1.

We continue to see a positive relationship between the implementation of the subsidy and EV purchase, with no observed relationship between the subsidy and non-EV purchase. Further, and strikingly, the positive relationship between the subsidy and EV purchase accelerates as pollution increases (as indicated by the interactions of Subsidy with AirPollutionTercileLevel2 and AirPollutionTercileLevel3). This supports the intuitive notion that EVSS has a greater effect on EV sales in cities with severe air pollution. Interestingly, we see in Column 2 that even under the most extreme circumstances (i.e., AirPollutionTercileLevel3), the traditional vehicle market remains unaffected by the EVSS. This underscores that these are largely different markets, with no discernable change in non-EV purchase behavior.

Contributions to Literature

Several contributions stem from this study. First, we contribute to the literature on technological substitution during the diffusion of de novo innovations. Prior research emphasizes that substitution is a dynamic process, and the emergence of new technologies does not ensure the abandonment of antiquated or outmoded ones (Adner 2002; Adner and Snow 2010). Yet, while the impact of subsidies on technology adoption has been widely examined (Kalish and Lilien 1983), research on how subsidies affect substitution is scarce. This is striking given the frequency with which subsidies are used, both within context and globally. Our findings underscore the fact that although subsidies can greatly accelerate the diffusion of new technologies, as would be expected, they do not ensure the abandonment of older ones, at least in the short term or within developing markets (e.g., Chinese autos).

We also contribute to the literature on policies affecting EV diffusion, specifically in terms of financial incentives. There is general empirical support for the positive effect of subsidies in simulating EV adoption (Beresteanu and Li 2011; Chandra, Gulati, and Kandlikar 2010; Gallagher and Muehlegger 2011; Huse and Lucinda 2014; Jenn, Azevedo, and Ferreira 2013; Zhang, Xu, and Zhang 2016). However, how consumers react to the subsidies on the broader automobile market is unclear. Our findings support the notion, at least within the Chinese context, that the purchase subsidies stimulate the EV diffusion with little change in the demand for traditional vehicles, suggesting that the growing EV demand results from market expansion rather than market cannibalization.

Further, we extend research on factors affecting the diffusion of EVs by investigating factors that moderate the observed response. Whereas extant research has examined many micro-level issues, including financial factors (Diamond 2009; Egbue and Long 2012; Lim, Mak, and Rong 2014), consumer characteristics (Bennett and Vijaygopal 2018; Hidrue et al. 2011; Sierzchula et al. 2014), and social influence (Axsen and Kurani 2011, 2013; Zhang et al. 2013), research has yet to consider the broader environment that bolsters or undermines the effect of subsidies. Findings suggest that the goal of subsidies is more likely to be realized among consumers in cities with higher wages and educational attainment. For cities with more severe air pollution, consumers are more likely to be attracted by the subsidy, though the cannibalization effect remains insignificant. Moreover, operating costs like gasoline price do not seem to moderate the effect. These moderating effects suggest that consumers’ motivation to purchase an EV vary across groups, resulting in the heterogeneous effect of subsidy on the automobile market.

Implications for Public Policy

Many governments have adopted and implemented EV subsidies, along with a suite of other measures, as a means to facilitate the diffusion of green technologies and reduce GHG emissions (Chandra, Gulati, and Kandlikar 2010; Gass, Schmidt, and Schmid 2014). Such measures are costly, requiring either the levy of additional taxes or the accrual of debt, neither of which is politically popular. It is therefore critical to assess whether subsidies result in a shift from traditional vehicles to EVs.

We find that the goal of reducing GHG emissions by introducing subsidies in developing markets is not being achieved. These subsidies do promote new technology adoption, but do not cannibalize the traditional market in the short term; this indicates that the effect of purchase subsidies is different than expected. Although it would be inappropriate to conclude that such subsidies should immediately be scrapped, as the policies might be inducing changes in the longer-term market by shifting the Bass curve forward, we believe these results should push policy makers to refine such policies to encourage such a demand shift early on. For example, the subsidy could be applied only when a less efficient vehicle is traded in (e.g., the Obama administration's successful Cash for Clunkers program).

Results of the moderating effects provide additional implications to policy makers on how to implement such measures more efficiently. And although our results are relegated to early-stage adoption in a single context, China, they remain telling. As consumers living in cities with higher education attainment are more likely to increase EV purchasing and decrease traditional car purchasing, spreading knowledge of green technology is likely an important supplement to financial incentives. Our findings show that consumers are more likely to purchase EVs under subsidy policies if they live in more polluted cities, though their incentives to purchase traditional vehicles do not decrease. It is possible that consumers with low environmental consciousness consider the behavior of purchasing EVs instead of replacing traditional cars with EVs as a way of reducing GHG emissions, which echoes Garvey and Bolton’s (2017) findings that consumers might decrease proenvironmental behavior after purchasing ecoproducts. Public policy makers should consider combining the financial incentives of adopting green technologies with publicizing environment protection knowledge to realize the goal of reducing GHG emissions.

Managerial Implications for Automobile Manufacturers

This research also yields implications for automobile manufacturers on their marketing strategies. Clearly, the emergence and adoption of new innovations have the potential to usurp established markets. However, when subsidies are introduced, it is critically important to understand which customer segments are more likely to respond to them.

Our results that the growing demand for EVs comes from market expansion rather than cannibalization of the traditional vehicle market suggest that marketing efforts from EV manufacturers should target two independent consumer segments: those likely to purchase an additional vehicle and those who would not otherwise purchase a vehicle (Holtsmark and Skonhoft 2014), and especially those with higher environmental awareness in both segments. One ACNielsen report on China's EV market indicates that nearly 60% of the high-end EV purchasers in China are also traditional vehicle owners, and 41% of the EV purchasers state that the lower GHG emissions is one of the reasons they chose an EV. ⁶ Marketing and manufacturing efforts should target these segments, with a special focus on consumers who might be subject to more severe air pollution. By targeting these groups, instead of trying to fold them into the traditional vehicle market segments, EV manufacturers can more carefully cater to their preferences and motivations.

Concluding Remarks

This work is subject to certain limitations that offer rich opportunities for future research. First, we examine the dynamics of the Chinese automobile market in the early stage of EV diffusion. Although subsidies are usually adopted to accelerate the diffusion of pioneering technologies (Kalish and Lilien 1983), our work cannot provide a comprehensive picture of how subsidies influence the evolution of the EV market across different stages or different markets throughout the world. Insofar as China is a developing automobile market, which is far from the point of ownership saturation, subsidies may plausibly yield different impacts in other markets (e.g., United States, Canada, Europe) where ownership is more diffuse. As our results have identified some level of cannibalization in cities with higher income and education attainment, it would be meaningful to examine the generalizability of the finding by replicating the study in developed markets.

Second, in this study, the identification of the policy effect is based on the phased rollout of the EVSS to different cities using observational data. Although we have controlled for city-invariant heterogeneity using city fixed effects, seasonality using time fixed effects, city-specific trends using city-specific splines, and included an extensive set of controls, we are unable to observe the direct reasoning behind activating the EVSS in any given city beyond the discussed preference for polluted cities and partial subsidy matching (Hao et al. 2014). Relatedly, the DID estimates only permit us to estimate the average treatment effect of the EVSS in China. This is simply a limitation of secondary data. It is probable that certain market sectors experience some degree of partial cannibalization (e.g., consumers who have access to public/private charging stations). Our analyses of moderating effects underscore the need for future research to expand the knowledge regarding subsidies.

Next, even though our study focuses on the effect of the EV private purchase subsidy on EV sales, we are unable to observe tax incentives or manufacturing subsidies that are made to manufacturers or the long-term changes in the public's perception of the viability of EVs; this information is vital to a broader understanding of market evolution (Beresteanu and Li 2011; Chandra, Gulati, and Kandlikar 2010). It is not clear whether subsidies for public vehicles (e.g., buses) would work the same way as the private purchase subsidy. Recall that we only observe private sales. We believe it is critical to take an expansive view of the nature of paternalism in the form of market subsidy. Future work should examine other subsidies and their effects on emerging and traditional markets.

Finally, this work studies only the overall effect of the EVSS in the automobile market and observes which mechanism dominates: market expansion or market cannibalization. Still, it is possible that both mechanisms coexist. We hope that future research can build on this and develop a comprehensive understanding of the automobile market by disentangling the two mechanisms and noting when either or both might operate.