Exploring the Impact of Government Subsidies on R&D Cost Behavior in the Chinese New Energy Vehicles Industry

Abstract

This study investigates whether government subsidies promote R&D cost stickiness in the new energy vehicle (NEV) industry in China—that is, whether public funding encourages firms to retain R&D resources even during periods of declining sales. While prior literature primarily explores the relationship between subsidies and R&D investment levels, it often overlooks firms’ financial position and dynamic cost behaviors. Given that R&D investment has high adjustment costs and is sensitive to cash flows, reductions in R&D spending during downturns may reflect managerial cost asymmetry rather than a crowding-out effect of subsidies. Moreover, government subsidies may serve as a signal of long-term market optimism, motivating managers to retain R&D resources during economic downturns. Using a panel dataset of 573 listed new energy vehicle (NEV) firms in China’s A-share market from 2007 to 2021, we construct a model based on the asymmetric cost behavior framework to empirically assess the impact of government subsidies on R&D cost stickiness. The results show that government subsidies significantly increase the degree of R&D cost stickiness. Serving as a signal of future market optimism, subsidies raise managerial expectations and incentivize decisions to retain R&D-related costs during economic downturns. This positive relationship is more pronounced in firms with high levels of green innovation, large-scale enterprises, and non-state-owned firms. These findings suggest that public funding alleviates managerial pressure to cut R&D expenses amid revenue declines, thereby supporting firms’ long-term innovation strategies. Our study contributes to the cost management literature by highlighting a novel channel through which subsidies influence managerial discretion under uncertainty. It also provides policy implications for the future phase-out of subsidies, emphasizing the need for complementary market mechanisms to sustain innovation investment, particularly for small, young, and financially constrained firms.

1. Introduction

1.1. Cost Stickiness and Managerial Discretion

Traditional cost accounting theory generally assumes that costs adjust symmetrically in response to changes in sales. However, Anderson et al. (2003) [1] point out that certain types of costs, including selling, general, and administrative (SG&A) expenses and research and development (R&D) expenditures, exhibit a phenomenon known as cost stickiness—that is, the increase in costs when sales rise tends to be greater than the decrease in costs when sales decline. This asymmetric cost behavior highlights the critical role of managerial discretion in cost adjustment decisions.

Existing studies suggest that the mechanisms behind cost stickiness are diverse and complex. They may arise from high adjustment costs [1,2,3,4], managerial expectations about future demand [5,6,7], or agency problems that distort resource allocation [8,9]. Moreover, cost stickiness is not a static phenomenon but rather a dynamic cost response. On the one hand, the negative association between cost stickiness and profitability can drive managers to reduce sticky costs or even exhibit anti-stickiness in order to avoid reporting losses (e.g., [7,10,11,12]). On the other hand, external stakeholders such as investors and creditors may exert pressure on managers to reduce the degree of cost stickiness, aiming to improve resource efficiency and mitigate the financial distress (e.g., [13,14,15]).

Recent literature increasingly shifts the focus to the external institutional environment as a determinant of managerial discretion over cost behavior. For instance, Kim et al. (2022) [16] find that political uncertainty leads managers to retain R&D resources, thereby increasing R&D cost stickiness. Similarly, Chang et al. (2022) [4] show that union strength induces firms to preserve the stickiness of labor costs while decreasing the level of SG&A cost stickiness. Exploring how institutional variables influence such behavior contributes to a deeper understanding of the economic consequences and governance of asymmetric cost adjustments.

1.2. R&D Cost Behavior and Its Unique Dynamics

Among various types of costs, R&D expenditures are particularly prone to exhibiting sticky behavior due to their long-term nature and the characteristics of knowledge accumulation. Firms often avoid sudden cuts in R&D spending to prevent disruptions in the innovation process that could hinder technological breakthroughs [17]. Moreover, the high adjustment costs associated with R&D—such as the difficulty of rehiring technical personnel or restarting suspended projects—further reinforce this cost stickiness [18].

Nevertheless, existing research has paid limited attention to the external factors influencing R&D cost stickiness. Most related literature regards the retention of R&D resources as a reflection of a firm’s future competitiveness [19,20]. While the role of internal expectations in shaping sticky cost behavior has been widely discussed, how institutional interventions—such as government subsidies—affect managerial decisions regarding R&D investment remains underexplored.

1.3. Government Subsidies: Conflicting Views and New Questions

Government R&D subsidies are designed to compensate for market failures, particularly the spillover effects and long gestation periods associated with innovation [21]. However, the literature presents conflicting views on their effectiveness. Some studies argue that subsidies encourage private R&D investment by alleviating financial constraints and signaling technological potential [22,23], while others find evidence of crowding out or substitution [24,25]. More recent research emphasizes conditional effects, suggesting that firm characteristics moderate the impact of public funding [26].

Our study seeks to contribute to this debate by investigating a novel mechanism through which subsidies may influence firm behavior: R&D cost stickiness. We propose that subsidies not only provide financial relief but also act as external signals of innovation value and policy endorsement, influencing managerial confidence and long-term resource allocation decisions.

1.4. Theoretical Contribution: Signaling and Agency Perspectives

Based on signaling theory [27], we argue that government subsidies serve as credible signals of a firm’s innovation potential. They reduce information asymmetry between managers and external stakeholders and, internally, reinforce managerial commitment to long-term strategic goals, thereby helping to mitigate agency problems [28]. Managers are typically risk-averse and subject to short-term performance pressures. When subsidies legitimize R&D expenditures and lower the perceived risks associated with them, managers may be more inclined to preserve R&D resources.

Within this framework, cost stickiness reflects the managerial decision to retain slack resources in support of future strategic objectives. We propose that public funding can function as a signal that reduces information asymmetry and alleviates agency concerns, thereby enhancing managerial confidence and encouraging greater investment in long-term projects.

1.5. Empirical Context: China’s NEV Industry

We examine our hypotheses in the context of China’s new energy vehicle (NEV) industry for three reasons. First, NEV firms, like other R&D-intensive sectors, face high innovation costs and long payback periods. Second, the NEV sector is crucial for addressing climate change, and its development is a policy priority under international agreements such as the Paris Accord (2021). Third, China’s economic transition provides a natural setting to explore the effectiveness of public funding in promoting technological upgrading and firm-level innovation.

NEV technologies follow a learning curve that requires continuous investment to realize productivity gains [29]. However, due to high initial costs and risks of imitation, firms may underinvest in R&D absent government support. Subsidies, in this setting, not only provide resources but also mitigate market failures and guide technological direction [30].

1.6. Key Findings

To empirically test the relationship between subsidies and R&D cost stickiness, we extend the ABJ (2003) [1] cost model by incorporating government subsidy data. Our baseline results indicate a positive association between subsidy levels and R&D cost stickiness. Furthermore, using Banker et al.’s (2014) [5] conditional cost model, we explore how subsidies influence managerial responses to sales declines. We find that subsidized firms are more likely to retain R&D resources despite negative revenue shocks, suggesting that subsidies enhance managers’ confidence in future sales. Finally, we examine the moderating role of green innovation and find that the positive effect of subsidies on R&D is more pronounced among firms with higher levels of green innovation.

1.7. Contributions

Our study makes three contributions: First, for cost stickiness theory, we introduce government subsidies as an institutional determinant of managerial discretion, showing that sticky R&D costs may reflect rational, forward-looking decisions under policy support. Second, for agency and signaling theories, we show that public funding reduces managerial risk aversion by enhancing information transparency and aligning managerial incentives with long-term innovation goals. Third, for innovation policy, we provide evidence that R&D cost stickiness under subsidies does not imply inefficiency but rather strategic smoothing behavior in response to external volatility. This highlights the role of subsidies not just in promoting R&D levels but also in shaping firms’ adjustment dynamics. Finally, our research provides insights for future NEV policy guidance and forecasting. We present evidence that large firms are better positioned to leverage government subsidies for technological upgrading. As subsidies are gradually phased out after 2022, it is important to monitor how managers adjust R&D investments and whether increased market competition may lead to technology monopolization by dominant firms.

2. Related Literature and Hypotheses Development

2.1. Asymmetric Cost Behavior

Anderson et al. (2003) [1] find that traditional cost accounting models underestimate the response of costs to increases in activity and overestimate the response of costs to decreases in activity. The cost stickiness means that managers deliberately retain costs when sales decrease. This asymmetric cost response shows that a firm’s resources decrease less when sales fall than when sales rise and a deliberate adjustment by managers to weigh economic consequences. In a follow-up paper, Anderson et al. (2007) [31] revisit the cost stickiness and conclude that it is not a mechanical cost adjustment. That means it is important to find the reasons that induce managers to asymmetrically adjust their costs. That is, the motivation of managerial discretion towards overt cost management in response to changes in activity.

It has been argued that this discretion is based on managers’ expectations of future sales. Contrary to the traditional cost view that an increase in the SG&A cost ratio implies inefficiency, Anderson et al. (2007) [31] interpret a high increase in the SG&A cost ratio as positive information conveying managerial optimism toward future sales. They find that firms with high increases in SG&A costs tend to earn abnormal positive returns. Banker, Ciftci, and Mashruwala (2008) [32] consider the manager’s optimistic expectation of demand in the market as a key source of asymmetric cost behavior. However, when managers are pessimistic about future sales, they may choose not to retain costs, thus reducing anti-cost stickiness [33]. Alternatively, when the prior sales decline, managers will choose not to retain resources in the face of successive sales declines. Banker et al. (2014) [5] provide evidence that the existence of cost stickiness requires a certain condition, that is, a rise in prior sales. His argument further emphasizes that the cost structure of a firm can be affected by managerial discretion and we need to focus on the motivation for it.

2.2. Government Subsidies and R&D Expenditures

Cost stickiness reflects managerial decisions to retain committed resources in the face of declining sales, anticipating future returns and avoiding the high costs of resource adjustment. This behavior directly affects firm profitability, and prior studies demonstrate that stickiness is sensitive to earnings pressures. On one hand, managers may avoid reporting losses and refrain from maintaining resources during downturns. When facing continuous sales declines, managers tend to be pessimistic about future demand and thus reduce stickiness [5,6,10]. Additionally, profit-oriented managers may even engage in anti-sticky behavior, aggressively cutting resources rather than retaining them. On the other hand, external investors also discourage sticky cost decisions due to their potential negative impact on profitability. Firms exhibiting sticky cost behavior are often associated with lower firm value, reduced dividends, and fewer institutional investors [13,14,34]. As such, maintaining resources during economic downturns may impose financial burdens and undermine credit, thereby exacerbating financing constraints.

Government subsidies can mitigate financial pressure and incentivize firms to retain R&D resources even during downturns, promoting long-term technological development. On the one hand, a substantial body of empirical literature demonstrates that subsidies can effectively stimulate firms’ investment in innovation. Xu et al. (2021) [35], through an analysis of Chinese listed firms, found that subsidies significantly influence both the likelihood and intensity of firms’ R&D investment. Chung et al. (2023) [36] argue that subsidies not only have a positive impact on private R&D activities but also enhance the overall social spillover effects of R&D. For high-risk firms in particular, subsidies can markedly improve both the level and efficiency of R&D investment [37]. On the other hand, the nature of high-tech industries is characterized by high initial costs and long investment horizons. Bointner (2014) [38] noted that, due to high uncertainty and expense, it is difficult for private firms to invest in new energy technology. According to Khezri et al. (2021) [39], subsidies can guide capital flows toward the renewable energy sector and help offset the financial risks associated with R&D in this field. Given the capital-intensive nature of new energy technologies, sufficient financial support is essential, and public fiscal resources can lay the groundwork for energy structure transformation. For developing countries, such transformation is more challenging and often depends on open trade platforms to attract financial support from developed economies. Moreover, in developing countries, energy transitions are more challenging due to limited fiscal capacity. Chang et al. (2020) [40] demonstrate that public financial support can significantly enhance the efficiency of new energy technology diffusion. Shrestha et al. (2022) [41] emphasize that energy transition is vital for economic development in developing countries and constitutes a fundamental component of their broader economic transformation.

H1. 

In the new energy vehicle (NEV) industry, government subsidies are positively associated with R&D cost stickiness.

2.3. The Signaling Effect of Subsidies on R&D Cost Behavior

Government subsidies exert a signaling effect by reducing agency costs and encouraging asymmetric cost behavior. When managerial and shareholder interests align, lower agency problems are associated with higher cost stickiness, as managers are more inclined to retain R&D resources for future value creation [19,31,42]. Subsidies also alleviate R&D-related risk aversion and reduce information asymmetry between managers and investors. As public signals, they help bridge the perception gap caused by selective disclosure and R&D spillovers. Prior research highlights their certification role—subsidized firms are perceived as government-endorsed [23,43], and government-disclosed information is often deemed more credible than corporate reports [44]. In high-tech sectors like new energy vehicles, such signals enhance investor confidence, attract external resources, and reinforce managerial commitment to R&D. Empirical evidence confirms that subsidies signal future market potential, attracting talent, capital, and technology, thereby improving firm performance and innovation efficiency [45,46,47,48].

A complementary stream of research highlights the role of managerial expectations in shaping cost stickiness. Banker et al. (2008) [49] first introduced the idea that managerial expectations—proxied by changes in sales—are key drivers of cost stickiness. They argued that managers may retain committed resources in anticipation of future benefits when they hold optimistic expectations about market conditions. Building on this, Banker et al. (2014) [5] incorporated forward-looking indicators, such as order backlog and GDP growth, to capture external market optimism. Their empirical findings revealed that positive macro signals increase managerial expectations, thereby strengthening cost stickiness. Notably, even when firms experienced continuous sales declines—suggesting managerial pessimism—upbeat signals from the external environment still prompted some managers to retain resources rather than cut costs, highlighting the role of external optimism in moderating cost asymmetry. Subsequent studies have examined other determinants of managerial optimism. For instance, Kama and Weiss (2013) [10] suggest that profit pressure can lead even pessimistic managers to engage in stronger asymmetric cost cutting. Ciftci and Zoubi (2019) [6] argue that the magnitude of past sales changes influences managerial expectations and thus cost behavior. Costa and Habib (2021) [14] further demonstrate that even optimistic managers tend to reduce cost stickiness when facing financial constraints, showing how resource availability conditions the optimism–behavior link.

Building on this literature, we propose that government subsidies may serve as a positive market signal that affects managerial cost decisions. Subsidies can be interpreted as endorsements of firm potential and sectoral support for the new energy vehicle industry, thereby fostering managerial optimism even during downturns. In particular, we aim to examine whether government subsidies moderate the relationship between past sales declines and cost stickiness, acting as an external affirmation of future prospects.

H2a. 

As a positive signal of the NEV industry’s growth potential, government subsidies lead subsidized firms to exhibit greater R&D cost stickiness than unsubsidized firms.

H2b. 

Government subsidies, by fostering more optimistic managerial expectations, increase the degree of R&D cost stickiness following prior sales increases and reduce the degree of cost anti-stickiness following prior sales decreases.

2.4. The Moderating Role of Green Innovation

We argue that government subsidies exert a stronger positive effect on R&D cost stickiness in firms with higher green innovation intensity, as such firms are more likely to interpret subsidies as signals of long-term market potential and are therefore more motivated to sustain innovation investment. Innovation should be evaluated not only by R&D inputs but also by the outcomes of these investments. Assessing innovation from the perspective of outcomes allows researchers to examine the efficiency of R&D efforts while also capturing the incentive effects of innovation performance on managerial behavior [50,51,52]. In the context of the new energy vehicle (NEV) industry, technological development often follows a “learning-by-doing” trajectory, where firms improve through accumulated experience over time. This learning curve, which integrates product life cycle and technological progress, necessitates long-term and stable R&D investment [29,30]. As a result, asymmetric cost behavior—particularly cost stickiness in R&D—is essential for firms to sustain technological advancement. Recent research on green innovation further confirms a strong positive correlation between R&D input and green patents, reinforcing the importance of sustained investment in innovation [53,54].

However, due to the high initial investment costs and delayed returns, the NEV sector often suffers from market failures, necessitating government intervention. Innovation in this field involves substantial risk, and firms frequently face financial constraints that hinder long-term R&D commitment [55]. Empirical studies have shown that government subsidies can effectively ease these constraints, whether in direct technology investments or across the broader supply chain, thereby enabling firms to maintain R&D intensity [56,57,58]. We therefore hypothesize that, in firms with higher levels of green innovation, government subsidies have a more pronounced positive effect on R&D cost stickiness. This suggests that the signaling and resource effects of subsidies are amplified when firms are already engaged in high-quality, policy-aligned innovation efforts.

H3. 

The positive relationship between government subsidies and R&D cost stickiness is more pronounced for firms with higher levels of green performance.

3. Research Design

3.1. New Energy Vehicle Sector

Previous studies about the relationship between subsidies and R&D resources mainly focus on manufacturing or R&D-intensive firms. We argue that the new energy vehicle (NEV) sector shares key characteristics with these industries, particularly in the importance of public funding in facilitating knowledge accumulation and technological advancement. However, NEVs also offer distinct behavioral additionality—namely, their contributions to reducing carbon emissions and enhancing social welfare. These factors are crucial not only for long-term firm productivity but also for the broader goal of ensuring energy security and supporting sustainable development.

The NEV sector is an important industry for developing countries like China in terms of economic transformation. Emerging countries transit their traditional energy sources to renewable energy sources, which will contribute to the stable development of national economies and the solution of global public problems. In the International Energy Agency’s (IEA) annual summary of electric vehicles, China accounts for half of the world’s electric vehicles and has the world’s largest cumulative number of related patent applications. We believe that public funding targeting energy transition in emerging countries is of great significance for exploring the relationship between government intervention and private cost decisions.

As a high-tech technology industry, new energy vehicles are characterized by “learning by doing”, which means that this kind of enterprise needs to invest in R&D personnel continuously and projects to accumulate the stock of technological knowledge. We estimate the NEV innovation inputs from the perspective of R&D cost stickiness because scholars have argued that firms’ heterogeneous resources should be immobilized to stay competitive [59]. In other words, R&D cost stickiness is a managerial decision to maintain resources during economic downturns which is vital to the firm to sustain its competitive advantages. This implies that the investment in R&D is stable and positive for knowledge-generating outputs. In addition, uncertain output and high initial costs of R&D programs reduce the investors’ incentives to invest, and we argue that government subsidies play a significant role in avoiding “lemon product” in the market and stimulating private investment in emerging industries such as NEVs.

3.2. Data and Sample Selection

Similar to Banker et al. (2014) [5], we need valid sales data for years t to t − 2 and valid R&D data for years t to t − 1. We choose valid subsidy data for years t to t − 1 as signals represent whether the policy is tightening or relaxed. We first identify data from Shanghai and Shenzhen A-share markets, in which there are 573 listed NEV firms. The new energy policy was spread by the government in 2007 in China and the grant data we could obtain start from 2007. The selected data consist of available financial data in the CSMAR from 2007 to 2021. We trim the top and bottom 0.5 percent of observations with extreme values. We drop observations with missing data on economic variables and the full sample contains 3825 firm-year valid observations.

3.3. Variable Measurement

3.3.1. Independent Variable

Government Subsidies (Sub): We measure subsidy intensity as the ratio of government subsidies to total assets, lagged by one period to account for delayed effects. To explore differences in cost behavior, we also create a binary variable indicating whether a firm received subsidies. Missing values are not replaced with zeros to avoid potential measurement bias.

3.3.2. Dependent Variable

The ABJ model improves cross-firm comparability and mitigates heteroskedasticity by employing a linear-log specification. Following this approach, we measure cost changes using the logarithmic difference in R&D expenditures between the current and previous period, i.e., log-change in R&D cost expenditure in year t relative to year t − 1.

3.3.3. Moderator Variable

To address the lag between R&D input and patent grants, we use the number of green patent applications—including both invention and utility model types—as a proxy for green innovation. Specifically, we measure it as ln (green patent applications + 1).

3.3.4. Control Variables

Among the control variables, asset intensity (AI) and employee intensity (EI) imply the adjustment of the firm’s resources. Firms generally choose to retain capital and personnel resources because of the high cost of adjusting assets and the increased cost of hiring new employees or the reduction in corporate reputation that comes with firing personnel. Controlling for GDP implies keeping employees optimistic when the macroeconomic environment rises will influence managers’ resource adjustment decisions. And Sdec implies that a continuous declining market downturn will make managers pessimistic and make adjustments to resources. The leverage of the firm will influence the managers’ intention to invest in R&D resources.

3.4. Empirical Model

Empirically, cost stickiness is modeled by regressing the change in cost on the change in sales, conditional on the direction of the change in sales in the current period [1]. We follow this method and interpret R&D cost stickiness as the phenomenon where the reduction in R&D costs for a one percent drop in sales is less than the increase in R&D costs for a one percent increase in sales.

Firstly, we use the basic model of cost stickiness proposed by Anderson et al. (2003) [1] as follows:

$${\mathrm{l}\mathrm{o}\mathrm{g}(RD}_{i,t}/{RD}_{i,t-1})={\beta }_0+{\beta }_1{\mathrm{l}\mathrm{o}\mathrm{g}(Sales}_{i,t}/{Sales}_{i,t-1})+{\beta }_2 {Dec}_{i,t}\times {\mathrm{l}\mathrm{o}\mathrm{g}(Sales}_{i,t}/{Sales}_{i,t-1})+{\epsilon }_{i,t}$$

(1)

$${\beta }_2={\alpha }_0+{\alpha }_1{Sub}_{i,t-1}+{\alpha }_2{AI}_{i,t}+{\alpha }_3{EI}_{i,t}+\alpha 4{Sucdec}_{i,t}+{\alpha }_5{LEV}_{i,t}$$

(2)

$${\beta }_1={\lambda }_0+{\lambda }_1{Sub}_{i,t-1}+{\lambda }_2{AI}_{i,t}+{\lambda }_3{EI}_{i,t}+{\lambda }_4{Sucdec}_{i,t}+{\lambda }_5{LEV}_{i,t}$$

(3)

In Equation (1), Dec is a dummy variable that equals 1 when sales revenue falls in year t and 0 otherwise. β1 indicates the extent to which costs increase when sales revenue increases. If β2 is negative and thus β1 + β2 is less than β1, it means that the change in R&D costs when sales fall is less than the change in R&D costs when revenues increase.

To investigate the impact of government subsidies on the R&D cost behavior of NEV firms, we substitute Equations (2) and (3) into Equation (1) to expand the standard cost-stickiness model as follows:

• Model A

$$\begin{array}{cc}{\Delta RD}_{i,t}={\beta }_0\hfill & +\left\{ {\alpha }_0+{\alpha }_1{Sub}_{i,t-1}+{\alpha }_2{AI}_{i,t}+{\alpha }_3{EI}_{i,t}+{\alpha }_4{Sucdec}_{i,t}+{\alpha }_5{LEV}_{i,t}+{\alpha }_6{GDP}_{i,t}\right\}\times {Dec}_{i,t}\times {\Delta Sales}_{i,t}\hfill \\ & +\left\{ {\lambda }_0+{\lambda }_1{Sub}_{i,t-1}+{\lambda }_2{AI}_{i,t}+{\lambda }_3{EI}_{i,t}+{\lambda }_4{Sdec}_{i,t}+{\lambda }_5{LEV}_{i,t}+{\lambda }_6{GDP}_{i,t}\right\}\times {\Delta Sales}_{i,t}+{\epsilon }_{i,t}\hfill \end{array}$$

• Mode B

$$\begin{array}{cc}{\Delta RD}_{i,t}={\beta }_0\hfill & +I_{i,t-1}\times ({\alpha }_1 \Delta {Sales}_{i,t}+{\alpha }_2\Delta {Sales}_{i,t}\times {Dec}_{i,t}+{\alpha }_3\Delta {Sales}_{i,t}\times {Sub}_{i,t-1}+{\alpha }_4{Dec}_{i,t}\times {\Delta Sales}_{i,t}\times {Sub}_{i,t-1})\hfill \\ & +D_{i,t-1}\times ({\lambda }_1\Delta {Sales}_{i,t}+{\lambda }_2\Delta {Sales}_{i,t}\times {Dec}_{i,t}+{\lambda }_3\Delta {Sales}_{i,t}\times {Sub}_{i,t-1}\hfill \\ & +{\lambda }_4{Dec}_{i,t}\times {\Delta Sales}_{i,t}\times {Sub}_{i,t-1}){+\alpha }_5{AI}_{i,t}\times {\Delta Sales}_{i,t}+{\alpha }_6{EI}_{i,t}\times {\Delta Sales}_{i,t}+{\alpha }_7{GDP}_{i,t}\times {\Delta Sales}_{i,t}\hfill \\ & +{\alpha }_8{LEV}_{i,t}\times {\Delta Sales}_{i,t}+{\epsilon }_{i,t}\hfill \end{array}$$

where:

ΔSalesi,t = log-change in sales revenue in year t relative to year t − 1;

ΔRDi,t = log-change in R&D costs in year t relative to year t − 1;

Sub i,t − 1 = logarithm of government subsidies to total assets in the last year (for year t − 1);

Subdummy = a dummy variable which equals 1 firm with subsidies in the last year (for year t − 1), 0 otherwise;

Deci,t = a dummy variable which equals 1 when sales revenue decreases for year t − 1 to t, and 0 otherwise;

Di,t − 1 = a dummy variable which equals 1 when sales revenue decreases for year t − 2 to t − 1, and 0 otherwise;

Ii,t − 1 = a dummy variable which equals 1 when sales revenue increases for year t − 2 to t − 1, and 0 otherwise;

GDP i,t = the percentage of GDP growth in year t;

EI i,t = the number of employees × 100,000/sales revenue;

AI i,t = logarithm of the ratio of total assets to sales revenue;

Sdec i,t = a dummy variable, which equals 1 when sales have decreased in two consecutive years, and zero otherwise;

Lev i,t = logarithm of the ratio of total debts to total assets;

GI1 i,t = natural logarithm of (1 + the number of green patent applications);

GI2i,t = natural logarithm of (1 + the number of green utility patent applications).

We use Model A to test the first hypothesis, i.e., whether the R&D cost stickiness of new energy vehicle companies is affected by government subsidies. To examine the positive effect of subsidies on R&D cost stickiness, we follow the ABJ model framework (Anderson et al., 2003) [1] and incorporate subsidy variables into the cost stickiness model, referencing Kama and Weiss (2011) [33], Golden et al. (2019) [3], and Chang et al. (2022) [4]. To ensure model robustness, we control for interaction terms between ΔSales and other control variables.

We expect to see a negative value for α₁, which implies that managers increase the degree of asymmetric R&D cost decisions under the influence of subsidies. But if α1 is positive or insignificantly negative, it means that government subsidies do not affect R&D cost stickiness.

To test the signaling effect of subsidies, we divide the sample into two groups—firms that receive government subsidies and those that do not. This allows us to examine whether subsidies have behavioral additionality, acting as a qualitative signal that encourages firms to sustain their R&D spending even under negative performance pressure. We argue that subsidies convey positive expectations about the future, incentivizing firms to maintain R&D investment. In contrast, non-subsidized firms may show no stickiness or even anti-stickiness in R&D costs.

Furthermore, following Banker et al. (2014) [5], we extend the traditional ABJ model to distinguish between conditional stickiness and anti-stickiness. In periods of sales increase, managers may retain resources in anticipation of continued growth, yielding a negative coefficient on lagged sales change (α₂ < 0), consistent with cost stickiness. Conversely, during sales decline, pessimistic expectations may lead to aggressive cost cuts, producing a positive coefficient (λ₂ > 0), indicating anti-stickiness [6,10,14]. By applying this model, we isolate the role of prior-period sales in shaping cost behavior and examine whether subsidies function as a forward-looking signal that alters managerial expectations. Specifically, in Model B, if subsidies reverse the effect of sales decline on managerial expectations, we expect the coefficient on the interaction term (λ₄) to be negative.

3.5. Descriptive Statistics

Table 1. Descriptive statistics (Currency: Chinese Yuan).

Variable	N	Mean	SD	Min	Max	Median
ΔRD	3829	0.203	0.382	−0.915	1.811	0.139
R&D/Sales	3825	0.042	0.035	0	0.475	0.039
R&D/Employee	3825	41,231	46,389	0	860,172	30,675
ΔSales	3829	0.138	0.274	−0.678	1.211	0.121
Sub	2914	0.007	0.011	0	0.217	0.005
Sdec	3825	0.11	0.313	0	1	0
Dec	3825	0.257	0.437	0	1	0
AI	3825	0.552	0.538	−0.738	2.336	0.533
EI	3825	0.142	0.097	0.009	0.551	0.123
GDP	3825	0.098	0.04	0.027	0.184	0.101
Lev	3825	0.442	0.19	0.069	0.956	0.438
SOE	3825	0.266	0.442	0	1	0
Size	3825	22.12	1.221	19.52	25.68	21.96
GI1	3430	0.527	0.918	0	5.958	0
GI2	3430	0.513	1.007	0	6.339	0

ΔRD = log-change in R&D costs in year t relative to year t − 1; R&D/Sales = R&D costs relative to sales revenue in year t; R&D/Employee = R&D costs relative to the number of employees; ΔSalesi,t = log-change in sales revenue in year t relative to year t − 1; Sub = logarithm of government subsidies to total assets; Sdec = a dummy variable, which equals 1 when sales have decreased in two consecutive years, and zero otherwise; Dec = a dummy variable which equals 1 when sales revenue decreases for year t − 1 to t, and 0 otherwise; AI = logarithm of the ratio of total assets to sales revenue; EI = the number of employees×100,000/sales revenue; GDP = the percentage of GDP growth in year t; Lev i,t = logarithm of the ratio of total debts to total assets; SOE = a dummy variable that equals 1 if the firm is a state-owned enterprise, and 0 otherwise; Size = natural logarithm of total assets; GI1 i,t = natural logarithm of (1 + the number of green patent applications); GI2i,t = natural logarithm of (1 + the number of green utility patent applications).

lists the descriptive statistics of main variables analyzed in our study. On average, R&D expenditures of the sample firms increased by about 20.3 percent which account for 4.2 percent of the company’s sales revenue. We divide R&D costs by the number of employees (R&D/Employee) to estimate the average R&D costs used by each employee and find that the average R&D costs per employee is CNY 41,231. Statistics on the sales decrease dummy, Dec, indicate that the percentage of sales decrease for firm-years in our sample is 25.7 percent. The consecutive decrease dummy, Sdec, indicates that the percentage of consecutive sales decrease is 11. The percentage of SEOs (stated-owned firms) in our sample is 26.6.

Table 2. Pearson Correlations.

	ΔRD	ΔSales	Dec	Sub	Sdec	AI	EI	GDP	Lev
ΔRD	1
ΔSales	0.357 ***	1
Dec	−0.241 ***	−0.657 ***	1
Sub	0.048 ***	0.028 *	−0.02	1
Sdec	−0.145 ***	−0.417 ***	0.598 ***	−0.006	1
AI	−0.062 ***	−0.152 ***	0.183 ***	0.034 **	0.151 ***	1
EI	0.012	−0.098 ***	0.051 ***	0.123 ***	0.056 ***	0.258 ***	1
GDP	0.110 ***	0.160 ***	−0.130 ***	−0.014	−0.070 ***	−0.095 ***	0.085 ***	1
Lev	−0.088 ***	−0.014	0.075 ***	−0.041 **	0.032 **	−0.144 ***	−0.144 ***	0.027 *	1

t statistics in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

presents the Pearson correlations matrix for the main and control variables in the sample mentioned above. We find that the change in R&D costs is significantly and positively correlated with changes in sales revenues (0.357, p-value < 0.001), suggesting that an increase in sales revenue is associated with an increase in R&D costs. This implies that resource adjustments relevant to those cost changes are decided based on sales activities. As expected, there are also significant correlations between ΔSales and ΔRD as well as ΔRD and Sub. We also perform a heteroskedasticity test between the main variables with variance inflation factors (VIFs) below 10 (as shown in

Table 3. VIF values.

Variable	VIF	1/VIF
Dec	2.31	0.433
ΔSales	1.8	0.555
Sdec	1.57	0.639
AI	1.14	0.879
EI	1.13	0.888
GDP	1.05	0.951
Lev	1.05	0.953
Sub	1.02	0.981
Mean VIF	1.38

). The significant correlations for variables are all below 0.8. We conclude that there is no problem with multicollinearity and heteroskedasticity.

4. Empirical Results

We first test the positive and signaling effects of government subsidies on R&D cost stickiness using the ABJ (2003) model [1]. We then use the Banker et al. (2014) [5] model to confirm that, even in the face of pessimistic market expectations, subsidies can still motivate managers to make decisions to maintain resources.

4.1. Regression Results

First, to test Hypothesis 1, we use Model A to investigate the effect of government subsidies on R&D cost behavior. We follow Chen et al. (2012) [9] and estimate R&D cost stickiness with firm-clustered standard errors. We include firm, industry, and year dummy variables in the regression models.

In the first column of

Table 4. The Effect of Government Subsidies on R&D Cost Behavior.

ΔRD	Pred.		Model A
		(1)	(2)	(3)	(4)
ΔSales (λ0)	+	0.524 ***	0.885 ***	0.789 ***	0.789 ***
		(11.394)	(6.162)	(5.211)	(5.189)
Dec×ΔSales (α0)	−	−0.193 **	−0.654 **	−0.493 *	−0.493 *
		(−2.133)	(−2.431)	(−1.720)	(−1.712)
ΔSales×Sub (λ1)	+		0.048	0.039	0.039
			(1.421)	(0.94)	(0.936)
Dec×ΔSales×Sub (α1)	−		−0.222 ***	−0.229 ***	−0.229 ***
			(−3.059)	(−2.732)	(−2.720)
Dec×ΔSales×AI			0.267 **	0.092	0.092
			(2.541)	(0.723)	(0.719)
Dec×ΔSales×EI			−1.393 **	−0.526	−0.526
			(−2.518)	(−0.802)	(−0.798)
Dec×ΔSales×Sdec			0.088	0.122	0.122
			(0.886)	(1.175)	(1.17)
Dec×ΔSales×GDP			−0.918	−0.722	−0.722
			(−0.477)	(−0.379)	(−0.377)
Dec×ΔSales×Lev			1.035 ***	0.652 **	0.652 **
			(4.101)	(2.096)	(2.087)
ΔSales×AI			−0.221 ***	−0.159 **	−0.159 **
			(−4.086)	(−2.478)	(−2.467)
ΔSales×EI			1.187 ***	0.797 *	0.797 *
			(3.61)	(1.947)	(1.938)
ΔSales×GDP			−0.024 **	−0.025 **	−0.025 **
			(−2.557)	(−2.356)	(−2.345)
ΔSales×Lev			−0.287 *	−0.119	−0.119
			(−1.680)	(−0.622)	(−0.620)
Intercept		0.161 ***	−0.186 ***	0.081	0.112 ***
		(3.424)	(−3.139)	(1.64)	(11.564)
N		3825	3825	3829	3761
Adjusted R2		0.162	0.186	0.158	0.273
Industry		No	Yes	No	Yes
Year		Yes	Yes	Yes	Yes
Firm		Yes	No	Yes	Yes

t statistics in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

, we find that the coefficient on λ₀ (ΔSales) is positive and that on α₀ (Dec×Sales) is negative, which validates that R&D costs are sticky in the NEV industry. Dec is a dummy variable equal to 1 when current-period sales decline. A negative coefficient on the interaction term (Dec×ΔSales) indicates the presence of cost stickiness. Therefore, a significantly negative coefficient on the triple interaction term (Dec×ΔSales×Sub) implies that subsidies are positively associated with cost stickiness. Conversely, a positive or statistically insignificant coefficient suggests that subsidies are not significantly related to cost stickiness. In the second column, we can see R&D costs get stickier as subsidies increase. To be specific, the coefficient on α₁ (Dec×ΔSales×Sub) is significantly negative at a₁ percent level, implying that an increase in government subsidies increases R&D cost stickiness. In the third and fourth columns, we perform other two fixed effects regressions, controlling for firm and industry fixed effects to avoid a spurious relationship between R&D cost stickiness and subsidies. The presented results in the third and fourth columns continue to support the positive relationship between R&D cost stickiness and government subsidies. The negative and significant coefficient of α₁ (Dec×ΔSales×Sub) indicates that an increase in government subsidies increases the R&D cost stickiness of NEV firms. Overall, our results are qualitatively similar across different models that government subsidies have a robust and positive effect on R&D cost stickiness. This is consistent with our first hypothesis.

Then we divide the data into subsidized and unsubsidized firms to test the signaling effect of subsidies on asymmetric R&D costs.

Table 5. Subsamples.

	(1)		(2)		(3)
ΔRD	Sub = 0	Sub = 1	Sub = 0	Sub = 1	Sub = 0	Sub = 1
ΔSales (λ0)	0.463 *	1.053 ***	0.618 ***	0.989 ***	0.470 *	0.989 ***
	(1.96)	(5.977)	(3.364)	(5.035)	(1.899)	(5.008)
Dec×ΔSales (α0)	−0.047	−0.940 **	0.049	−0.866 *	0.226	−0.866 *
	(−0.116)	(−2.330)	(0.129)	(−1.927)	(0.539)	(−1.917)
Intercept	−0.132	−0.199 ***	0.082	0.114 **	0.079 ***	0.116 ***
	(−1.294)	(−2.646)	(0.643)	(2.018)	(3.911)	(9.878)
Controls and their interaction terms				Included
N	717	3108	718	3111	617	3047
Adjusted R2	0.331	0.168	0.262	0.135	0.531	0.275
Industry	Yes		No		Yes
Year	Yes		Yes		Yes
Firm	No		Yes		Yes

t statistics in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

shows that R&D cost stickiness is significant for the subsample with subsidies in all regressions. In the first column of Table 4 we can see that the coefficient on α₀ is significantly negative at the 5% level for the group with subsidies (Sub = 1) and insignificant for the group without subsidies (Sub = 0). In both the second and third columns, the coefficients on α0 are significantly negative, but coefficients on λ₀ are positive, implying that unsubsidized firms’ R&D costs show anti-stickiness during economic downturns compared to subsidized firms. This is consistent with our second hypothesis.

In

Table 6. The Signaling Effect of Subsidies on R&D Cost Stickiness.

ΔRD	Pred.	Model B
I×ΔSales (α1)		0.464 ***
		(3.266)
I×ΔSales×Des (α2)	−	−0.448 **
		(−1.994)
I×ΔSales×Sub (α3)		−0.001
		(−0.028)
I×ΔSales×Sub×Des (α4)	−	−0.273 *
		(−1.954)
D×ΔSales (λ1)		0.495 ***
		(2.708)
D×Des×ΔSales (λ2)	+	0.006
		(0.022)
D×ΔSales×Sub (λ3)		0.008
		(0.112)
D×Des×ΔSales×Sub (λ4)	?	−0.215 **
		(−2.292)
		−4.338
Intercept		0.049
		(1.118)
Controls and their interaction terms		Included
Adjusted R2		0.154
Year		Yes

t statistics in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

we see the positive signaling effect of subsidies on R&D cost stickiness regardless of whether managers have optimistic future sales expectations (a prior sales increase) or pessimistic future sales expectations (a prior sales decrease). Following prior literature, we acknowledge that the coefficient on α₂ (I×ΔSales×Des) is negative conditional on a prior sales increase and λ₂ (D×Des×ΔSales) is positive conditional on a prior sales decrease. We are surprised to find a significantly negative coefficient on λ₄ (D×Des×ΔSales×Sub), implying that managers are inclined to rely on subsidies as additional signals of sales expectations, i.e., they choose to perform asymmetric R&D cost behavior even in the face of a continuous decline in sales.

To examine the moderating role of green innovation outcomes in the relationship between subsidies and R&D cost stickiness, we divide the sample into two groups based on the average number of green patent applications: High and low green innovation levels. Given that the median value of green patent applications is zero, we use the mean rather than the median as the cutoff point to better capture variation in innovation performance. As shown in

Table 7. The Moderating Role of Green Innovation.

	Green Innovation
ΔRD	High	Low	High	Low
ΔSales (λ0)	0.821 ***	0.922 ***	0.881 ***	0.912 ***
	(3.858)	(5.264)	(3.908)	(5.264)
Dec×ΔSales (α0)	−1.258 **	−0.461	−1.058 **	−0.531 *
	(−2.352)	(−1.418)	(−2.050)	(−1.671)
ΔSales×Sub (λ1)	0.064	0.029	0.125 ***	−0.020
	(1.315)	(0.687)	(4.141)	(−0.473)
Dec×ΔSales×Sub (α1)	−0.441 ***	−0.150 *	−0.524 ***	−0.089
	(−2.770)	(−1.882)	(−3.535)	(−1.205)
Intercept	−0.294	−0.177 ***	−0.458 *	−0.185 ***
	(−1.220)	(−2.872)	(−1.731)	(−2.898)
Controls and their interaction terms		Included		Included
Ho: b0 (Dec×ΔSales×Sub) = b1 (Dec×ΔSales×Sub) Empirical p-value = 0.015			Empirical p-value = 0.010
N	1403	2422	1531	2294
Adjusted R2	0.193	0.16	0.240	0.190
Industry	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes

t statistics in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

, the positive effect of subsidies on R&D cost stickiness is more pronounced among firms with high green innovation levels. In the first column, where the number of green patent applications (GI1) is used as the proxy, the coefficient of the triple interaction term α₁ is −0.441 and statistically significant at the 1% level for the high innovation group, while the corresponding coefficient for the low innovation group is −0.150 and not significant. Fisher’s permutation test confirms that the difference in α₁ between the two groups is statistically meaningful. We further verify this finding using green utility model patent applications (GI2) as an alternative measure, which again shows that the positive relationship between subsidies and R&D cost stickiness is stronger in firms with higher levels of green innovation.

4.2. Additional Analyses

4.2.1. Firm Size

There has long been debate regarding the effect of firm size on R&D activities. Following Schumpeter’s perspective, some scholars argue that larger firms benefit from economies of scale, making their R&D output more efficient. Empirical studies support that large enterprises are often more capable of leveraging policy incentives to enhance innovation activities and improve performance through R&D [60,61]. Conversely, other scholars contend that small firms, constrained by limited resources, are more motivated to engage in R&D by capitalizing on spillover effects [22,62]. In this view, government subsidies play a vital role in alleviating financing constraints and stimulating R&D investments among smaller firms. In the context of the new energy industry, investigating the heterogeneous impact of firm size on the relationship between subsidies and R&D cost stickiness is both theoretically and practically significant.

We classify firms into large and small groups based on the median value of total assets among samples and estimate Model A to examine the effect of subsidies on R&D cost stickiness. As shown in

Table 8. Firm Size.

ΔR&D	Large	Small
ΔSales (λ0)	0.691 ***	0.919 ***
	(3.073)	(4.535)
Dec×ΔSales (α0)	−0.624	−0.485
	(−1.435)	(−1.374)
ΔSales×Sub (λ1)	0.032	0.023
	(0.631)	(0.32)
DecΔSales×Sub (α1)	−0.350 **	−0.097
	(−2.405)	(−0.889)
Intercept	−0.000	0.234 ***
	(−0.001)	(3.588)
Controls and their interaction terms	Included
N	1917	1912
Adjusted R2	0.211	0.137
Firm	Yes	Yes
Year	Yes	Yes

t statistics in parentheses. ** p < 0.05, *** p < 0.01.

, the coefficient of the triple interaction term α₁ (Dec×ΔSales×Sub) is significantly negative in large firms (coefficient = −0.350, t = −2.285), whereas it is negative but insignificant in small firms (coefficient = −0.097, t = −0.889). This suggests that subsidies play a more substantial role in enhancing R&D cost stickiness in large enterprises.

4.2.2. Ownership Structure (SOEs vs. Non-SOEs)

The literature presents mixed evidence on the differential effects of subsidies between state-owned enterprises (SOEs) and private firms. Some studies suggest that SOEs are more adept at capturing policy benefits, while others find that private firms respond more effectively to subsidy incentives. Yu et al. (2016) [63], for example, argue that, in the new energy sector, subsidies significantly encourage R&D investment by SOEs and crowd out private sector innovation. In contrast, Jin (2018) [64] finds that subsidies have a more pronounced incentive effect on private enterprises. Sun et al. (2020) [65] further demonstrate that reducing state ownership enhances the positive effect of subsidies on firms’ R&D spending.

Table 9. Ownership Structure.

ΔR&D	SOE = 1	SOE = 0
ΔSales (λ0)	0.587 *	0.873 ***
	(1.928)	(4.594)
Dec×ΔSales (α0)	−0.247	−0.503
	(−0.392)	(−1.492)
ΔSales×Sub (λ1)	0.023	0.048
	(0.323)	(0.988)
Dec×ΔSales×Sub (α1)	−0.147	−0.290 **
	(−1.312)	(−2.580)
Intercept	−0.025	0.189 **
	(−0.414)	(2.222)
Controls and their interaction terms	Included
N	1019	2806
Adjusted R2	0.126	0.18
firm	Yes	Yes
year	Yes	Yes

t statistics in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

indicates that, for state-owned enterprises, the coefficient α₁ is −0.147 and not statistically significant, while for non-state-owned enterprises, the coefficient is −0.290 and significant at the 5% level. These results imply that subsidies are more effective in supporting the persistence of R&D efforts among private firms.

4.2.3. Managerial Ownership

The divorce of ownership and control thesis suggests that, while shareholders tend to support innovation activities, managers prefer diversification. Risk-averse managers may avoid uncertain R&D investments due to concerns over short-term financial performance [66]. However, our findings suggest that managers, when incentivized by government subsidies, are not short-sighted; they tend not to significantly cut R&D expenditures even amid consecutive sales declines. As discussed earlier, the signaling effect of subsidies may boost both external investor and managerial confidence in future sales. We argue that the positive relationship between subsidies and R&D cost stickiness becomes more pronounced when managerial interests are closely aligned with those of shareholders. To test this, we examine the moderating role of equity incentives and ownership structure, drawing on Brüggen and Zehnder (2014) [42], who use high equity incentives as a proxy for interest alignment and reduced agency problems.

Finally,

Table 10. Managerial Ownership.

	Management Stockholding
ΔRD	HIGH	LOW
ΔSales (λ0)	0.753 ***	0.800 ***
	−3.458	−4.084
Dec×ΔSales (α0)	−0.073	−0.494
	(−0.136)	(−1.320)
ΔSales×Sub (λ1)	−0.017	0.049
	(−0.322)	(0.968)
Dec×ΔSales×Sub (α1)	−0.348 **	−0.199 **
	(−2.285)	(−2.325)
Intercept	0.09	0.044
	−0.666	−0.887
Controls and their interaction terms		Included
Ho: b0 (Dec×ΔSales×Sub) = b1 (Dec×ΔSales×Sub) Empirical p-value = 0.077
N	1933	1896
Adjusted R2	0.193	0.16
Firm	Yes	Yes
Year	Yes	Yes

t statistics in parentheses. ** p < 0.05, *** p < 0.01.

shows that the coefficient of α₁ (D×ΔSales×Sub) is significantly negative in both groups, but the coefficient for the group with high management stockholdings (coefficient = −0.348, t = −2.285) is larger than that of the group with low management stockholdings (coefficient = −0.199, t = −2.325). The difference between the two subsamples is significant (p-value = 0.077), consistent with our inference.

4.3. Robustness Test

4.3.1. Alternative R&D Costs and Subsidies Components

The results are robust to alternative specifications of the independent and dependent variables. First, we rerun the regression replacing R&D costs with R&D expenditures per employee, following Baysinger et al. (1991) [59], as this measure is more stable and less sensitive to business cycle effects. As shown in

Table 11. Regression Results.

Δln (R&D/Employee)
ΔSales (λ0)	0.551 **	0.383 *	0.452 **
	(2.07)	(1.709)	(2.001)
Dec×ΔSales (α0)	−0.594	−0.324	−0.399
	(−0.923)	(−0.611)	(−0.737)
ΔSales×Sub (λ1)	0.017	−0.011	−0.012
	(0.209)	(−0.153)	(−0.172)
Dec×ΔSales×Sub (α1)	−0.543 ***	−0.453 ***	−0.459 ***
	(−2.881)	(−2.958)	(−3.013)
Intercept	−0.049	0.039	0.887 ***
	(−0.213)	(0.234)	(5.548)
Controls and their interaction terms		Included
N	2915	2915	2915
R2	0.046	0.038	0.052
Industry	No	Yes	Yes
Firm	Yes	Yes	Yes
Year	Yes	No	Yes

t statistics in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

, the results remain robust. Next, we redefine government subsidies by using the log of the subsidy indicator and lagging the subsidy by one period. Again, as shown in

Table 12. Regression Results.

ΔR&D	(1)	(2)	(3)
ΔSales	0.808 ***	0.846 ***	0.720 ***
	(5.136)	(5.777)	(4.598)
Dec×ΔSales	−0.496 *	−0.556 *	−0.363
	(−1.683)	(−1.957)	(−1.144)
ΔSales×lnSub	0.008 *	0.007 *	0.010 **
	(1.925)	(1.8)	(2.055)
Dec×ΔSales×lnSub	−0.017 **	−0.015 *	−0.018 **
	(−2.189)	(−1.895)	(−1.988)
Intercept	0.022	−0.198 ***	0.082 *
	−0.462	(−3.400)	(1.676)
Controls and their interaction terms		Included
N	3825	3825	3825
Adjusted R2	0.171	0.185	0.157
Industry	No	Yes	Yes
Firm	Yes	No	Yes
Year	Yes	Yes	Yes

Where lnSub = ln(subsidies). t statistics in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

, our findings are consistent, demonstrating the robustness of our results across different variable specifications.

4.3.2. Accounting for Endogeneity

We use a propensity score matching approach to address endogeneity concerns. Research has documented that ownership nature affects the subsidies, with state-owned enterprises being more likely to receive subsidies than private enterprises. Also, the firm size affects access to subsidies, with some scholars confirming that substantial subsidies are granted to small and medium-sized enterprises (SMEs) because of the funding gap in R&D. Based on this rationale it is valid to additionally use the average change in sales, asset intensity, and SG&A costs to weaken the effect of some characteristics of the firm itself on subsidies, which makes our model more robust. Specifically, we use a logit model to estimate propensity scores based on the average change in the selected variables. A 1-to-1 nearest-neighbor matching with a caliper of 0.05 is applied, and the average treatment effect (ATE) on ΔRD is estimated.

We employ propensity score matching (PSM) to address sample selection bias. After matching, standardized differences for most covariates—such as Size and AI—drop by over 70%, indicating improved balance (as shown in

Table 13. Covariate Balance Before and After Matching.

Variable	Match Status	Treated Mean	Control Mean	% Bias	t	p-Value
ΔSales	Unmatched	0.1374	0.1420	−1.6	−0.40	0.686
	Matched	0.1364	0.1300	2.2	0.87	0.382
Size	Unmatched	22.101	22.201	−7.9	−1.98	0.048
	Matched	22.099	22.074	2.0	0.81	0.417
ΔSGA	Unmatched	0.1446	0.1409	1.0	0.27	0.785
	Matched	0.1410	0.1457	−1.3	−0.56	0.574
Top10	Unmatched	57.607	57.041	3.9	0.93	0.355
	Matched	57.587	58.558	−6.7	−2.65	0.008
AI	Unmatched	0.5395	0.6040	−11.7	−2.90	0.004
	Matched	0.5406	0.5446	−0.7	−0.29	0.771
Soe	Unmatched	0.2770	0.2204	13.1	3.10	0.002
	Matched	0.2763	0.2554	4.9	1.87	0.062

). Although some differences remain (e.g., Top10), overall comparability between treated and control groups is substantially enhanced. Balance diagnostics further support the match quality: The pseudo R² decreases from 0.007 to 0.002, and the p-value of the LR test rises from 0.000 to 0.003. The average standardized bias falls from 6.5% to 3.0%, Rubin’s B drops from 21.6 to 11.2, and Rubin’s R remains within the acceptable range (0.75), confirming satisfactory covariate balance. Subsidized and unsubsidized firms are matched so that the absolute value of the difference between the two propensity scores is less than or equal to 0.01 as shown in Figure 1. There are 438 firms in the sample and 1265 firm-year observations and our results remain robust in

Table 14. PSM and Regression Results.

ΔRD	Model A
ΔSales (λ0)	0.779 ***
	(3.01)
Dec×ΔSales (α0)	−0.382
	(−0.86)
ΔSales×Sub (λ1)	0.089
	(1.31)
Dec×ΔSales×Sub (α1)	−0.237 **
	(−2.47)
Dec×ΔSales×AI	0.134
	(0.77)
Dec×ΔSales×EI	−0.466
	(−0.40)
Dec×ΔSales×Sdec	0.261 *
	(1.84)
Dec×ΔSales×GDP	1.375
	(0.43)
Dec×ΔSales×Lev	−0.186
	(−0.40)
ΔSales×AI	−0.256 ***
	(−3.05)
ΔSales×EI	0.837
	(0.91)
ΔSales×GDP	−0.041 **
	(−2.01)
ΔSales×Lev	0.349
	(1.08)
	−0.066
Intercept	0.013
	(1.83)
Observations	1265
Number of id	438
R-squared	0.233
Year	Yes
Firm	Yes

t statistics in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

. We can see that the coefficient of α₁ (Dec×ΔSales×Sub) is significantly negative, indicating that the subsidy positively affects the increase in R&D cost stickiness.

5. Conclusions

This study examines the positive effect of subsidies on R&D cost stickiness. We focus on how public funding in an economic downturn facilitates managers’ retention of resources with high adjustment costs and dynamic trade-offs between slack resources and weak market demand.

We select 573 NEV firms in China’s A-share market in the CSMAR database from 2007 to 2022 and use OLS regression analysis to verify the impact of subsidies on asymmetric R&D costs. We demonstrate that R&D cost stickiness is associated with government subsidies by selected financial data of Chinese NEV firms, i.e., the more government subsidies, the higher R&D cost stickiness. In addition, firms without subsidies are either insignificantly sticky or anti-sticky, while firms with subsidies show significant stickiness. We use Banker’s model (2014) to exclude the effect of prior sales further, suggesting that the existence of R&D cost stickiness is not a mechanical cost adjustment by managers but rather a dynamic discretion influenced by the government funding, serving as an optimistic signal.

Furthermore, we use the number of green patents to measure the level of green innovation in NEV firms. The results show that the positive effect of government subsidies on R&D cost stickiness is more pronounced in firms with higher levels of green innovation. This suggests that subsidies not only support R&D persistence but also reinforce firms’ commitment to environmentally oriented innovation strategies.

Finally, our heterogeneity analyses reveal that the relationship between government subsidies and R&D cost stickiness is stronger in firms where managerial and shareholder interests are more closely aligned. We also find that larger NEV firms tend to exhibit higher R&D cost stickiness compared to smaller firms, possibly because they are better positioned to invest in innovation and realize economies of scale. Moreover, non-state-owned NEV firms display more pronounced R&D cost asymmetry than their state-owned counterparts—this may be partly due to the relatively small proportion of SOEs in our sample and partly indicative of the growth potential of private firms in the sector.

Overall, our study provides empirical evidence that government support plays a critical role in helping firms in emerging industries retain R&D resources. During economic downturns, subsidies encourage managers to maintain stable and smooth R&D investment, enhancing long-term innovation capacity. Our study has several limitations. First, the empirical model faces complexity due to multiple interaction terms, which makes it challenging to isolate the moderating effects clearly. Second, the identification of causal relationships remains limited despite our methodological efforts. Third, while we propose a signaling mechanism, we do not directly observe managerial perceptions. We aim to adopt more specific and targeted methods in future research to address these limitations and further validate our findings.