In this article, I develop a structural framework to evaluate how tiered subsidy adjustment coefficients shape the technology choices of electric vehicle manufacturers. The context of my study is China’s electric vehicle purchase subsidy program, which has utilized a tiered structure based on battery energy density since 2017. My research objective is to determine whether such tiered subsidy designs can effectively induce manufacturers to adopt higher-performance batteries and thus promote technological progress in the electric vehicle industry. Using a dataset of 2,571 pure electric vehicle models from the Chinese Catalogue of Recommended Models for New Energy Vehicle Demonstration and Promotion from 2017 to 2021, I estimate a profit-maximizing discrete choice model of battery technology adoption. My identification strategy exploits variation in subsidy levels across battery energy density categories, driving ranges, and time periods. The empirical results indicate that subsidy amounts have a statistically significant positive effect on manufacturer profits, suggesting that subsidies are capitalized into firm payoffs. However, my counterfactual simulations reveal an important nuance: while the existing tiered subsidy adjustment coefficients do increase the number of new electric vehicle models adopting batteries with energy density at or above 160 Wh/kg by over 14 percent compared with a uniform subsidy coefficient design, the overall impact on average energy density is modest, at approximately only 1 percent. Notably, I find that a quasi-linear subsidy coefficient schedule—where subsidy rates increase linearly with energy density thresholds—could promote approximately a 10 percent increase in average energy density of newly launched electric vehicle models. This finding highlights a fundamental design trade-off in attribute-based subsidies for electric vehicles: tiered thresholds create strategic bunching at eligibility cutoffs, whereas linear schedules provide continuous incentives for marginal technological improvement. My study contributes to the literature on attribute-based regulation by offering micro-level evidence from the supply side of the electric vehicle market, and my analytical framework provides a practical tool for policy pre-assessment in the absence of randomized policy variation.
My interest in this topic stems from the observation that subsidies are commonly employed to correct market failures associated with positive externalities, particularly in the electric vehicle sector. The fundamental design question is how governments should structure subsidy schemes to achieve maximum social benefit under fiscal constraints. Performance-based tiered subsidies represent a common approach, wherein products meeting higher performance thresholds receive larger subsidy amounts. This policy design logic presumes that by making better-performing products more competitive, firms will be incentivized to invest in research and development and continuously improve product attributes. Such policies are deployed in many contexts, including electric vehicle purchase subsidies, energy-efficient appliance subsidies, and green building incentives in China. Many countries have adopted analogous mechanisms, such as Japan’s special depreciation for energy-saving equipment and the United States’ rebates for high-efficiency appliances.
Existing literature has primarily focused on estimating the demand-side effects of these attribute-based subsidy policies on product adoption. However, empirical evidence on how such policy designs influence producer behavior and technology adoption at the intensive margin remains scarce. If tiered subsidy designs can change the strategic product choices of manufacturers and encourage them to research, develop, and produce higher-performance products, then such policies can also accelerate technology diffusion by reshaping the consumer choice set. Whether tiered subsidy designs achieve this supply-side objective remains an open empirical question, and answering it is essential for evidence-based policy optimization.
Policy Background and Institutional Details
China’s electric vehicle market development has been strongly shaped by government intervention. From the establishment of the Electric Vehicle Major Project under the national 863 Program in 2001 to the China Manufacturing 2025 initiative in 2015 and the more recent framing of electric vehicle development as a strategic element of peak carbon and carbon neutrality goals, an extensive policy apparatus has supported this sector. The purchase subsidy program has attracted substantial attention from researchers and practitioners alike. Under this program, the central government subsidizes consumers who purchase battery electric and plug-in hybrid vehicles that qualify for the Recommended Models Catalogue. Beginning in 2016, the program was scaled to cover all regions of the country, and local governments could supplement central subsidies, occasionally pushing combined subsidies to as much as 60 percent of vehicle prices. According to the China Association of Automobile Manufacturers, 331,000 electric vehicles were sold in 2015, a figure that quadrupled to 1.256 million units by 2018. More recently, annual sales grew from 1.367 million units in 2020 to 6.887 million in 2022, representing over a quarter of all new passenger car sales in that year.
The subsidy framework incorporates two interconnected tiered designs. The first tier is based on driving range, with higher-range electric vehicle models qualifying for larger base subsidies. For example, in 2016, a pure battery electric vehicle with a range between 100 and 150 kilometers received RMB 25,000, while models with 150-250 kilometers of range received RMB 45,000, and those exceeding 250 kilometers were eligible for RMB 55,000. This design intentionally links subsidy payments to the electric vehicle’s driving range because achieving higher range typically requires adopting more advanced technologies, which increases production costs. By subsidizing these higher-range models, the policy aimed to make longer-range electric vehicles more affordable to consumers, thereby encouraging manufacturers to produce them.
The second tier, which is central to my study, involves battery energy density adjustment coefficients. The rationale for adding this refinement was to address a potential distortion: under range-based subsidies alone, manufacturers could potentially increase an electric vehicle model’s certified test range by reducing vehicle weight or motor power without meaningfully advancing battery technology. Such gaming behavior could result in low-quality electric vehicles that are manufactured merely to receive subsidies, ultimately harming the long-term development of the industry. In 2017, four Chinese ministries jointly issued updated subsidy rules requiring that pure battery electric passenger vehicles achieve a minimum battery system energy density of 90 Wh/kg to qualify for subsidies, with qualifying vehicles receiving a 1.1 multiplier if their battery energy density exceeded 120 Wh/kg. In 2018, the policy was revised to require a minimum of 105 Wh/kg, with models achieving 105-120 Wh/kg receiving 0.6 times the base subsidy, 120-140 Wh/kg receiving the full subsidy, 140-160 Wh/kg receiving 1.1 times, and models at or above 160 Wh/kg receiving 1.2 times the base amount. The thresholds were raised again in 2019 when the minimum was set at 125 Wh/kg, with models between 125 and 140 Wh/kg receiving 0.8 times the base subsidy, 140-160 Wh/kg receiving 0.9 times, and those above 160 Wh/kg receiving the full base subsidy.
These repeated policy adjustments illustrate an escalation in technological requirements that mirrors the industry’s actual progress. Figure 1, shown below, depicts the distribution of newly launched electric vehicle models from 2017 to 2021 against the relevant energy density thresholds, with vertical dashed lines representing the eligibility boundaries for various subsidy tiers. The figure reveals that the density of newly launched models tends to be highest just around the cutoffs—specifically, among models with energy density between 140 and 160 Wh/kg, the number of models with energy density barely exceeding 140 Wh/kg is greater than the total number of models in other ranges within the same category. This pattern provides descriptive evidence that tiered subsidy adjustment coefficients incentivize strategic behavior among electric vehicle manufacturers, leading them to select batteries that qualify for the next higher subsidy bracket.
Conceptual Framework and Identification Strategy
To answer whether attribute-based subsidy designs influence the technology choices of electric vehicle manufacturers, I must compare the costs and benefits associated with different technology levels. In practice, however, manufacturers typically treat technology costs as proprietary business information, which poses an obstacle to direct empirical analysis. To address this challenge, I first estimate the marginal contribution of different battery technologies to electric vehicle model profits and then use the derived estimates to perform counterfactual simulations of policy alternatives.
Consider a manufacturer deciding which battery type \(j\) to install in a newly designed electric vehicle model \(i\) at time \(t\). The expected profit \(\pi_{ijt}\) is a linear combination of observable vehicle attributes \(X_{ijt}\), marginal costs \(MC_{ijt}\), and subsidy amount \(S_{ijt}\). I express this as:
\[
\pi_{ijt} = X_{ijt}\gamma + \alpha(S_{ijt} – MC_{ijt})
\]
In this specification, \(X_{ijt}\gamma\) captures the price premium generated by observable vehicle characteristics, while \(\alpha(M)\) represents the influence of cost and subsidy pass-through on final profitability. The actual profit realized by the manufacturer includes both this deterministic component and an idiosyncratic shock:
\[
R_{ijt} = \pi_{ijt} + \varepsilon_{ijt}
\]
I make three key assumptions to make the model tractable. First, the random utility shock \(\varepsilon_{ijt}\) is independently and identically distributed according to a Type I extreme value distribution. Second, every new electric vehicle model must be equipped with one battery chosen from a set of available alternatives; hence, there is no outside option of not choosing a battery. Third, conditional on other model characteristics, the same battery choice set is available to all manufacturers, and battery prices do not vary across buyers. Under these assumptions, the probability that model \(i\) at time \(t\) is equipped with battery \(j\) from the feasible set \(L_t\) is:
\[
P_{ijt} = \frac{\exp(\pi_{ijt})}{\sum_{k \in L_t} \exp(\pi_{ikt})}
\]
Substituting the linear profit function and noting that terms common across battery choices cancel out, I obtain the estimation equation by taking log differences in market shares relative to a benchmark battery choice \(j=0\). For vehicles in range category \(r\), the difference in log market shares between battery \(j\) and the benchmark battery at time \(t\) is:
\[
\ln s_{rjt} – \ln s_{r0t} = \beta'(k_{rjt} – k_{r0t}) + \alpha(S_{rjt} – S_{r0t}) + (b_{jz(t)} – b_{0z(t)}) + \lambda_r + \lambda_t + u_{rjt}
\]
Here, \(k_{rjt}\) is a battery-related vehicle attribute—specifically, the driving range associated with the electric vehicle model—and the parameter vector \(\beta’\) captures both the production cost effect and the valuation effect of range. The term \((b_{jz(t)} – b_{0z(t)})\) is a battery type-by-year fixed effect that absorbs any unobserved, time-varying differences in battery pricing and consumer preferences for specific battery technologies. The specification also includes range fixed effects \(\lambda_r\) and time fixed effects \(\lambda_t\). Because the battery selection decision is made at the electric vehicle design stage, and because battery manufacturers typically negotiate supply contracts based on anticipated volumes, the scope for simultaneity bias is limited.
Data Sources and Variable Definition
My empirical analysis draws on a unique dataset assembled from multiple sources. The core database is the Catalogue of Recommended Models for the Promotion and Application of New Energy Vehicles, published by China’s Ministry of Industry and Information Technology. This catalogue contains the essential technical parameters for each electric vehicle model, including battery energy density, driving range, battery type, battery capacity, and vehicle weight. All models included in the catalogue must pass a rigorous series of inspections and field tests, and listing in the catalogue is a necessary condition for eligibility for government subsidies. Hence, the catalogue provides a reliable census of mainstream electric vehicle models in China’s market.
To construct a comprehensive dataset with additional model characteristics, I matched each catalogue entry to its corresponding marketing name using the Ministry’s vehicle purchase tax exemption lists. This step permitted me to extract the manufacturer’s suggested retail price from an online automotive database and cross-validate the technical parameters reported in the catalogue. After dropping observations with missing key variables, duplicates, and light commercial vehicles, pickup trucks, and micro vans, my final analysis sample includes 2,571 unique pure battery electric vehicle models marketed in China between 2017 and 2021. Table 1 presents the distribution of battery chemistries across sample years.
| Year | NCM models | Share | LFP models | Share | Other models | Share |
|---|---|---|---|---|---|---|
| 2017 | 255 | 72.23% | 35 | 9.78% | 68 | 18.99% |
| 2018 | 356 | 75.58% | 21 | 4.46% | 94 | 19.96% |
| 2019 | 585 | 82.05% | 74 | 10.38% | 54 | 7.57% |
| 2020 | 392 | 71.66% | 133 | 24.31% | 22 | 4.02% |
| 2021 | 258 | 53.53% | 210 | 43.57% | 14 | 2.90% |
I classify battery types according to two dimensions: energy density and technological route. In total, the classification yields 27 distinct battery categories. Energy density intervals are constructed to align with the prevailing subsidy eligibility thresholds and the empirical distribution of the data. Table 2 reports descriptive statistics for the key variables used in my analysis.
| Variable | Mean | Std. Dev. | Min | Max |
|---|---|---|---|---|
| Energy density (Wh/kg) | 143.22 | 20.39 | 91.1 | 206 |
| Subsidy (10,000 RMB) | 1.69 | 1.32 | 0 | 6 |
| Driving range (km) | 348.43 | 117.58 | 100 | 1,008 |
| Battery capacity (kWh) | 46.26 | 18.70 | 9 | 144.4 |
| Manufacturer suggested retail price (10,000 RMB) | 15.85 | 9.90 | 2.68 | 75.43 |
| NCM battery indicator | 0.72 | 0.45 | 0 | 1 |
| LFP battery indicator | 0.18 | 0.39 | 0 | 1 |
The variable subsidy takes different base values depending on an electric vehicle model’s driving range. The amount of subsidy an electric vehicle model can receive is the product of the range-based subsidy level and the energy density adjustment coefficient. Table 3 summarizes the subsidy schedule by year and range category.
| Range category | 2016 | 2017 | 2018 | 2019 | 2020 | 2021 |
|---|---|---|---|---|---|---|
| 100–150 km | 2.5 | 2.0 | — | — | — | — |
| 150–200 km | 4.5 | 3.6 | 1.5 | — | — | — |
| 200–250 km | 4.5 | 3.6 | 2.4 | — | — | — |
| 250–300 km | 5.5 | 4.4 | 3.4 | 1.8 | — | — |
| 300–400 km | 5.5 | 4.4 | 4.5 | 1.8 | 1.6 | 1.3 |
| ≥400 km | 5.5 | 4.4 | 5.0 | 2.5 | 2.3 | 1.8 |
Empirical Results: Baseline Parameter Estimates
I estimate the baseline model using ordinary least squares with fixed effects and clustering of standard errors at the range category level. Table 4 presents the main results. The dependent variable is the difference between the log market share of the battery category and the log market share of the benchmark category. Column (1) includes only battery-type-by-year fixed effects, and the coefficient on subsidy difference is positive and statistically significant. In Columns (2) and (3), I add batch and range effects; the coefficient remains positive. The specification in Column (3) suggests that an increase in the subsidy difference of RMB 10,000 is associated with an increase in the relative log market share of about 0.96. Columns (4) through (6) report estimates that include the log of driving range as an additional control for attribute differences. Across all specifications, the coefficient on the log of driving range is positive and significant, implying that consumers value longer driving range, which translates into profitability for manufacturers that adopt battery configurations enabling higher range.
| Dependent variable: ln(s) − ln(s₀) | (1) | (2) | (3) | (4) | (5) | (6) |
|---|---|---|---|---|---|---|
| Subsidy difference (10,000 RMB) | 1.828*** | 0.960** | 0.960*** | 1.683*** | 0.888** | 0.888*** |
| (0.371) | (0.382) | (0.143) | (0.359) | (0.373) | (0.136) | |
| ln(driving range) | — | — | — | 2.412*** | 1.484** | 1.484** |
| — | — | — | (0.661) | (0.580) | (0.413) | |
| Observations | 540 | 540 | 540 | 540 | 540 | 540 |
| Adjusted R² | 0.360 | 0.575 | 0.574 | 0.377 | 0.580 | 0.579 |
| Battery×Year FE | Yes | Yes | Yes | Yes | Yes | Yes |
| Batch FE | No | Yes | Yes | No | Yes | Yes |
| Range FE | No | Yes | Yes | No | Yes | Yes |
| Clustering level | None | None | Range | None | None | Range |
In Column (6), my preferred specification, I find that a one-unit increase in the difference in subsidy amounts—measured in ten thousand yuan—between an electric vehicle model adopting battery technology \(j\) and a model adopting the benchmark technology increases the relative profitability of the electric vehicle model by 0.888 units. In terms of economic magnitude, this coefficient indicates that public subsidies are substantially passed through to manufacturer surplus. Since the subsidy amount is determined by the base subsidy schedule and adjustment multiplier for each battery and range combination, the identifying variation emerges from policy reforms that alter the multiplier structure.
Notably, the estimated coefficient on the logarithm of driving range of 1.484 implies that a 1 percent longer driving range is associated with an increase in manufacturer profits of roughly 1.5 units of the log-share metric. This relationship reflects both consumer willingness to pay for range and the additional engineering costs of accommodating larger battery packs.
Robustness Checks and Sensitivity Analysis
To ensure that the baseline results are not an artifact of particular modeling choices, I performed a series of robustness tests. The first test replaces the benchmark battery category. In the main specification, the benchmark is the most widely adopted technology in each year; however, to investigate whether this choice matters, I reran the model using a single benchmark across all years—the NCM battery with energy density of 140 Wh/kg. The second robustness test trims extreme values from the sample by dropping observations with market shares below the fifth percentile or above the ninety-fifth percentile. The third test restricts the analysis sample to the two mainstream battery chemistries, NCM and LFP, while excluding others. The fourth test adds range-by-year fixed effects, which allow the marginal value of additional driving range to differ across years, potentially due to changes in the availability of charging infrastructure or consumer preferences. The fifth test clusters standard errors simultaneously by both the range category and batch to account for any two-dimensional correlation in the error term. Finally, the sixth test replaces the log of driving range with its absolute level.
| Specification | Subsidy coefficient | SE | Observations |
|---|---|---|---|
| (1) Alternative benchmark category | 0.782** | (0.234) | 468 |
| (2) Trimmed sample | 0.817*** | (0.122) | 499 |
| (3) Excluding other chemistries | 0.969*** | (0.162) | 414 |
| (4) Adding range×year fixed effects | 0.630** | (0.141) | 539 |
| (5) Two-way clustering | 0.888* | (0.332) | 540 |
| (6) Range in absolute terms | 0.882*** | (0.141) | 540 |
Consistent with the baseline results, the coefficient on subsidy is positive and statistically significant in every specification, ranging between 0.63 and 0.97. These estimates provide strong evidence that subsidy differentials across battery energy density tiers influence the expected profitability of electric vehicle model designs. The stability of the coefficient across various specifications adds credibility to the structural interpretation that the subsidy effect on manufacturer profits is real and economically meaningful.
Battery Types and Their Marginal Contribution to Vehicle Profits
Using the estimated fixed effects for battery category by year, I can summarize how different battery technologies have contributed to electric vehicle model profits over time relative to the benchmark category in each year. The estimated contributions, expressed in ten-thousand-yuan units, are presented in Table 6. The benchmark technologies are the energy-density 120 Wh/kg NCM battery for 2017, the 140 Wh/kg NCM battery from 2018 to 2019, and the 160 Wh/kg NCM battery for 2020 to 2021.
| Energy density | 2017 | 2018 | 2019 | 2020 | 2021 |
|---|---|---|---|---|---|
| 100 Wh/kg | 0.123 | — | — | — | — |
| 110 Wh/kg | −0.190 | −1.189 | 0.730 | — | — |
| 120 Wh/kg | −0.066 | −1.169 | −0.916 | −0.418 | 0.233 |
| 130 Wh/kg | −0.356 | −1.340 | −1.336 | −0.646 | 0.510 |
| 140 Wh/kg | −0.153 | −0.428 | −0.500 | −0.212 | 0.319 |
| 150 Wh/kg | −1.071 | −1.139 | −1.271 | −1.146 | −0.180 |
| 160 Wh/kg | — | −1.220 | −0.820 | −0.243 | −0.106 |
| 170 Wh/kg | — | — | −2.217 | −0.767 | 0.699 |
| 180 Wh/kg | — | — | −1.737 | −0.964 | 0.211 |
Three salient patterns emerge from Table 6. First, the maximum energy density level observed in each year increased steadily during the period, moving from 150 Wh/kg in 2017 to 180 Wh/kg in 2019 and thereafter. Second, in 2017, lower energy density batteries made a stronger positive contribution to electric vehicle model profits than their higher-density counterparts, which reflects the higher cost and limited availability of advanced battery systems at the time. Third, this pattern reversed gradually thereafter as the economics of high-density battery packs improved. In 2020 and 2021, high-density batteries (above 170 Wh/kg) showed positive marginal profit contributions, indicating that the battery technology frontier had shifted significantly. These results suggest that battery cost reduction and performance improvement over time have fundamentally altered the trade-offs facing electric vehicle manufacturers.
Counterfactual Simulations of Policy Alternatives
With the structural parameters estimated from the discrete choice model, I conduct counterfactual simulations to quantify the effects of alternative subsidy designs on the technology choices of electric vehicle manufacturers. I define two counterfactual policy scenarios. The first is a uniform subsidy policy, under which every electric vehicle model receives the same adjustment coefficient regardless of its battery energy density, subject to a minimum energy density threshold for eligibility. The second is the actual existing tiered policy under which higher energy density yields a larger adjustment multiplier. To make the comparison meaningful, I keep the total subsidy budget identical across scenarios by adjusting the base subsidy level proportionally.
For each policy scenario, I compute the implied subsidy amount for each potential electric vehicle design and use the estimated model to calculate the probability that a profit-maximizing manufacturer would adopt each battery technology. The model probabilities are converted into expected counts of new electric vehicle models by multiplying them by the number of newly launched models observed in each range category and batch. For the actual tiered policy, I use the observed counts directly; for the uniform policy scenario, I use my model to simulate the counterfactual choices of manufacturers.
Table 7 reports the differences in the predicted number of new electric vehicle models adopting batteries at each energy density level under the two policy schemes. A negative value indicates that the tiered subsidy adjustment coefficient design reduces the number of models at the corresponding energy density, relative to the uniform subsidy scheme. Conversely, a positive value indicates an increase.
| Energy density | Uniform policy predicted count | Tiered policy observed count | Percentage change |
|---|---|---|---|
| 100 Wh/kg | 58 | 51 | −12% |
| 110 Wh/kg | 66 | 56 | −16% |
| 120 Wh/kg | 266 | 257 | −3% |
| 130 Wh/kg | 195 | 174 | −11% |
| 140 Wh/kg | 909 | 880 | −3% |
| 150 Wh/kg | 155 | 150 | −3% |
| 160 Wh/kg | 418 | 477 | +14% |
| 170 Wh/kg | 133 | 145 | +9% |
| 180 Wh/kg | 97 | 106 | +9% |
The counterfactual analysis reveals that the tiered subsidy adjustment coefficient design does significantly shift the distribution of technology choices in the direction the policy intended. There is a substantial reduction of 16 percent for the 110 Wh/kg group and 11-12 percent reductions for the 130 Wh/kg and 100 Wh/kg groups. Most notably, the number of new electric vehicle models adopting batteries at 160 Wh/kg increases by 14 percent, which corresponds to roughly 59 additional models over the sample period. The models with 170 and 180 Wh/kg also experience about 9 percent increases in count. This result indicates that the tiered subsidy adjustment coefficient successfully induces more manufacturers to choose battery products with an energy density above 160 Wh/kg. It is important to observe, however, that the largest induced increase is concentrated among the models that just satisfy the highest threshold at 160 Wh/kg, and not among the very highest categories.
Next, I decompose these policy effects by battery chemistry to understand the differential response of manufacturers using NCM and LFP technologies. Table 8 presents the results by technology route. The effect of the tiered subsidy adjustment coefficient is most concentrated in the NCM market: models with NCM batteries in the low-density categories face a reduction of over 20 percent, while models with NCM at 160 Wh/kg increase by 14 percent. For LFP batteries, the policy effect is a reduction in the number of models at lower energy densities, and no increase is observed in the highest categories, because virtually no LFP products in the sample reach the high-density levels.
| Energy density group | NCM change | NCM % change | LFP change | LFP % change | Other change | Other % change |
|---|---|---|---|---|---|---|
| 100 Wh/kg | −3 | −12% | −3 | −13% | −1 | −12% |
| 110 Wh/kg | −8 | −20% | −2 | −9% | 0 | 0% |
| 120 Wh/kg | 0 | 0% | −9 | −13% | 0 | 0% |
| 130 Wh/kg | −11 | −11% | −8 | −13% | −1 | −3% |
| 140 Wh/kg | −19 | −3% | −8 | −4% | −2 | −2% |
| 150 Wh/kg | −4 | −4% | 0 | 0% | 0 | 0% |
| 160 Wh/kg | 55 | +14% | 0 | 0% | 3 | +14% |
| 170 Wh/kg | 12 | +9% | 0 | 0% | 1 | +13% |
| 180 Wh/kg | 9 | +9% | — | — | 0 | 0% |
The overall economic effect of the tiered policy on average energy density is small, however. I compute the model-weighted average energy density under the actual tiered policy and under the counterfactual uniform policy. Under the uniform policy, the average energy density over the entire sample period would be roughly 142 Wh/kg, whereas under the tiered policy it is approximately 143.2 Wh/kg. This corresponds to a difference of only around 1 percent. The reason for this modest direct effect is twofold. First, the tiered design disproportionately induces the choice of battery products near the threshold, with many firms selecting the least-cost product that qualifies for the next tier. Second, the battery-type fixed effects estimated for high energy density levels were mostly negative for most of the sample period, which offsets the policy’s pull for products with higher energy density. Thus, although the tiered subsidy adjustment coefficients are directionally effective, their overall magnitude in terms of the average energy density of newly launched electric vehicles has been limited.
Dynamic Counterfactuals: Alternative Designs of Subsidy Schedules
I next explore the implications of altering the design of the tiered subsidy schedule. My objective is to understand what schedule characteristics are relevant to incentivize the adoption of electric vehicles with higher energy density battery technologies without expanding the overall subsidy budget. I analyze two alternatives to the existing tiered policy: policy A increases the subsidy spread between energy density tiers, and policy B introduces a quasi-linear schedule in which the subsidy adjustment coefficient increases proportionally with the energy density of the battery, without discrete jumps.
In policy A, I increase the maximum adjustment coefficient for the highest energy density bracket in each year to 1.2, and lower the adjustment coefficient for the lowest energy density products to 0.6. This increases the subsidy difference between high and low energy density. In policy B, I set the minimum energy density product as receiving zero subsidy, and I calculate the subsidy amount as a linear function of the energy density of the battery. More formally, within a given batch, the subsidy rate for energy density bracket \(d\) is:
\[
S_{drt} = \frac{(d – d_{\min})}{(d_{\max} – d_{\min})} \cdot S_{rt}
\]
where \(S_{rt}\) is the total subsidy budget allocated for vehicles in range category \(r\) in batch \(t\), \(d_{\min}\) is the lowest energy density, and \(d_{\max}\) is the highest energy density available in the market at that time. By construction, both policy A and policy B have the same total subsidy expenditure as the observed actual policy in every range category and batch. I use the estimated model parameters to simulate electric vehicle manufacturers’ technology choices under each alternative policy.
| Energy density | Actual policy | Policy A (wider spread) | Policy B (quasi-linear) |
|---|---|---|---|
| 100 Wh/kg | 51 | 46 | 28 |
| 110 Wh/kg | 56 | 52 | 34 |
| 120 Wh/kg | 257 | 234 | 165 |
| 130 Wh/kg | 174 | 155 | 180 |
| 140 Wh/kg | 880 | 835 | 653 |
| 150 Wh/kg | 150 | 179 | 221 |
| 160 Wh/kg | 477 | 553 | 628 |
| 170 Wh/kg | 145 | 173 | 308 |
| 180 Wh/kg | 106 | 147 | 234 |
Table 9 reports the number of newly launched electric vehicle models selecting each battery energy density level under the actual and both counterfactual policy designs. Policy B, the quasi-linear schedule, markedly outperforms the actual tiered policy in shifting the technology distribution toward high energy density. Under policy B, the number of models using 180 Wh/kg or above grows to 234, which is more than double the count of 106 under the existing policy. Even if I compare policy A with the actual policy, widening the subsidy spread has a visible impact on the number of models with the highest energy density, though the numbers do not fully reach the levels seen under policy B.
Figure 2 illustrates the effect of the alternative subsidy designs on the average energy density of newly launched electric vehicles. Under the actual tiered subsidy adjustment coefficient, the average energy density is around 143 Wh/kg. Under policy A, the average energy density increases to approximately 146 Wh/kg. Under policy B, the average energy density reaches about 158 Wh/kg, an increase of roughly 10 percent relative to the actual policy. The rationale behind this result is that policy B does not produce local bunching at discrete thresholds. In the actual tiered system, as well as under policy A, the discrete jumps in subsidy payments create strong incentives for manufacturers to select the minimum technology level that qualifies for the higher payment bracket. In contrast, in a quasi-linear system, each additional increment of energy density is compensated continuously, so the profit-maximizing electric vehicle manufacturer decides on the optimal technology level by trading off the additional battery cost against the extra subsidy amount along the full support of feasible energy density values.
The results of my counterfactual experiments highlight the central role of subsidy schedule design in an attribute-based subsidy framework. The discrete tier structure commonly used in many public programs may be an administratively expedient approach, but it creates unintended incentives for product-level strategic manipulation at the thresholds. A quasi-linear structure is likely to produce stronger technological progress for a given level of subsidy expenditure, despite being more challenging to design because it requires specifying detailed energy density brackets.
Extensions and Heterogeneity
To gain deeper insight into manufacturer behavior, I examine whether the response to subsidy differences varies systematically with market or vehicle attributes. I split the sample between vehicles produced by manufacturers that specialize in electric vehicles and those produced by traditional automobile manufacturers. The battery supply market’s evolution might also affect responses, because when many battery suppliers compete, the cost premium of high energy density products declines, making manufacturers more sensitive to subsidy spreads. I therefore estimate refined versions of the structural model by product segments. While the baseline model uses the manufacturer as the unit of analysis for entry decisions, each manufacturer offers multiple models in the sample, so I also examine the count of models per manufacturer as an alternative measure.
To explore the role of market heterogeneity and to assess whether the response to the tiered subsidy design is uniform, I re-estimated the model using the subsamples of NCM and LFP battery chemistries. For the NCM battery market, my point estimates of the subsidy coefficient are larger, reflecting the fact that NCM products have historically covered a broader range of energy density levels and have therefore been more responsive to energy density-based incentives. By contrast, LFP products in my sample are typically anchored in the low-to-intermediate energy density region and are less responsive to subsidy differences, since their sales appeal is linked more to cost and safety than to high energy density.
The model’s linear structure assumes that the marginal effect of the subsidy on profit is constant. I relax this assumption and find preliminary evidence that the marginal effect of subsidies may diminish at higher total subsidy amounts. When manufacturers consider adopting the highest energy density batteries, they may be sensitive to the risk of battery supply bottlenecks. Using the point estimates from the linear model, I nevertheless find robust evidence that a positive subsidy differential matters.
Discussion and Policy Implications
My analysis yields several policy-relevant insights. The first is that attribute-based tiered subsidies, targeted at the energy density of electric vehicle batteries, do influence the technology selection strategy of manufacturers. However, in the current Chinese electric vehicle subsidy policy, the discrete adjustment coefficient system created an incentive for manufacturers to choose products barely above a particular threshold rather than pursuing markedly higher technological performance. According to my estimates, the selected energy density of newly introduced electric vehicle models was approximately only 1 percent higher than it would have been under a uniform subsidy distribution. In comparison, a quasi-linear scheme, with the same total government expenditure, would have lifted the average selected energy density by about 10 percent.
From a policy-design perspective, this evidence supports the view that while tiered structures are a step forward from non-differentiated subsidies, they are still suboptimal if the goal is to maximize technological advancement. If a government’s primary objective is to promote the adoption of frontier technology products, then the subsidy should be concentrated on the highest feasible technology standards, with a large enough incentive gap to overcome the manufacturers’ economic reluctance to select the most expensive and most advanced batteries.
Alternatively, a quasi-linear subsidy scheme, in which the subsidy paid to manufacturers is proportional to the energy density of the battery technology embodied in the electric vehicle model, is preferable because it offers continuous incentives for marginal technological improvement and avoids the bunching problem. Although implementing such a schedule is potentially more complex than a coarse tiered system, the additional administrative cost may be justified by the substantial social gains from a faster pace of technological development in the electric vehicle industry.
It is also crucial to regularly update the eligibility thresholds and subsidy schedules to reflect the state of technology and market conditions. The electric vehicle industry is characterized by fast innovation cycles, and a subsidy rule designed for one stage of technology may be irrelevant or harmful at a later stage. For example, the minimum energy density thresholds increased from 90 Wh/kg in 2017 to 125 Wh/kg in 2019, and this periodic tightening of standards tracked market developments. My empirical analysis demonstrates that battery technology in the sample improved remarkably over a short period, from average energy density of 117 Wh/kg in 2017 to 150 Wh/kg in 2021. This technological progress underscores both the value and the limitations of static subsidy rules.
The lessons of my paper extend beyond electric vehicle subsidies. Many other new energy industries, including the solar photovoltaic and wind power sectors, rely on government subsidy schemes that are differentiated by product performance. These include feed-in tariff structures that pay higher rates for higher-efficiency installations. The evidence presented here suggests that program designers in those industries should pay close attention to how the subsidy schedule is defined; they should be mindful of potential bunching around eligibility thresholds and should carefully consider linear or quasi-linear designs, which may deliver faster technological progress for the same amount of government spending.
Conclusions
In this article, I examine whether attribute-based tiered subsidies affect the design choices of electric vehicle manufacturers. My empirical framework uses a structural model of manufacturer profit maximization to estimate the value of adopting different types of electric vehicle batteries, and my counterfactual experiments simulate the technology choices that manufacturers would make under alternative policy regimes. I find that the tiered subsidy adjustment coefficient in the Chinese electric vehicle market produced a moderate shift toward higher energy density batteries among newly launched electric vehicle models, but this shift was primarily concentrated among products that just crossed the highest subsidy threshold.
My findings demonstrate that more sophisticated, quasi-linear subsidy schedules would likely accelerate technological progress in the electric vehicle sector without requiring an increase in the total government subsidy budget. By extending the earlier empirical literature, which has mainly focused on the demand-side impacts of attribute-based subsidies, my study contributes important supply-side evidence and provides a practical framework that can be used for ex ante evaluation of policy proposals. As governments around the world design clean energy industrial policies, my analysis emphasizes that the structural details of the policy design, rather than simply the total subsidy amount, should be a primary consideration for achieving technological goals.
My study also opens avenues for further research. An interesting direction is to explore the dynamics of innovation in battery technology itself, tracking how manufacturers respond over longer observation windows to past subsidy incentives. In my current design, I treat the set of electric vehicle battery technologies as predetermined in each period, which is a reasonable approximation for the short sample interval. Future research should attempt to model firms’ innovation investments in battery products and their product adoption decisions jointly, using a richer dynamic structural model.
In conclusion, the paper provides robust evidence that manufacturers of electric vehicles respond strongly to the economic incentives embedded in the design of public subsidy schemes, and it suggests that more effective policy designs are available. Attribute-based subsidy design constitutes a powerful instrument for improving the technological composition of the electric vehicle fleet, yet its effectiveness depends greatly on how eligibility thresholds and payment schedules are chosen.