3.1. Description of the Chinese Lithium-Ion Battery Market
Driven by the confluence of two goals of Chinese policy makers, i.e., environmental pollution reduction and conversion of low-cost, low-tech products to others that require higher levels of skills and technology to produce (the ‘Made in China 2025′ strategic national plan), LiB growth is largely dependent on the development of NEVs. Based on statistics provided by the Global Lithium-Ion Battery Supply Chain Ranking report [
55], the Chinese LiB market is immense, and this is due primarily to large domestic battery demands, control of over 80% of the global raw material refining, 77% of cell capacity and 60% component manufacturing. The Battery—Global Market Trajectory & Analytics report [
56] suggests that the China compound annual growth rate of the LiBs will be at 16.6% by 2027 and is estimated to be worth US
$61.1 billion. The same report goes on the classify Ningde Times Energy Technology Co., Ltd. (CATL), BYD Co. Ltd., Tianjin Lishen Battery Co., Ltd., Beijing Hezhong Pufang New Energy Technology Co., Ltd., the Wanxiang Group, China Aviation Lithium Battery (CALB), Gotion High-tech Co., Ltd. (formerly Hefei Guoxuan Hi-Tech Power Energy Co., Ltd.), OptimumNano, Coslight, and Microvast Power Systems as the leading power battery manufacturers in the country. Among these Beijing Hezhong Pufang New Energy Technology, Tianjin Lishen Battery Co., Ltd., Gotion High-tech Co., Ltd., CALB, OptimumNano, Coslight, and Microvast are private companies. BYD is also a private company but additionally relies on foreign investments. These and other companies have provided a third of global battery production capacity in 2019 and this trend is expected to remain up to 2025 [
57], allowing Jin et al. to suggest that because of China’s decade-long pilot and subsidy programs that were designed to drive EV technology advancement, the average battery capacity for pure electric cars increased from 35 to 44 kWh (approximately 23%) over the past five years, though this remains lower than in the United States and Europe in 2019 [
58]. The dataset of companies used in this paper includes a wide array of private, private and foreign invested, in addition to state-owned enterprises.
Based on research conducted by Zhang et al. [
14], following the return of Hong Kong to the P.R.C. in 1997 and application of national policies concerning NEVs, EV battery technology patents grew exponentially, with this trend increasing annually. The authors suggest two policies, namely ‘The New Energy Infrastructure Project Management Interim Provisions’, and the ‘Decision on Speed Up the Cultivating and Developing Strategic Emerging Industries’, show that technological R&D is closely linked to national policy. Indeed, in a new report on China’s electric vehicle development, Jin et al. identify that China’s top-down management styles, i.e., its Five-Year-Plans, and resultant policies such as the ‘Ten Cities, Thousand Vehicles’ pilot projects, subsidy programs, tax breaks, and technical standards are largely responsible for the “strategic and policy continuity in China’s new energy vehicle development” [
58]. The Chinese State Council Information Office on 2 November 2020 released the New Energy Vehicle Industry Development Plan (2021–2035) that proposed that by 2025, the sales volume of NEVs and intelligent connected vehicles should reach 25% and 30%, respectively [
59]. Although the literature documents a wide array of Chinese government policies and subsidies meant to spur R&D into improving LiBs, which often encourages automakers to sell EVs below manufacturing costs [
60,
61], case studies on their effects on specific companies are scarce. Nonetheless, in 2012, China BAK Battery, a manufacturer of LiBs received a US
$1.9 million subsidy from the Chinese National Development and Reform Commission and MIIT [
62] for its battery module project. As reported, government funds were slated to be used to enhance battery module efficiency for EVs and e-bikes. More recently, Scott and Ireland document that BYD and CATL are increasing their investments into battery production [
63]. Masiero et al. in an examination of government subsidies on the largest Chinese EV manufacturer BYD, found that government subsidies (
Table 1), combined with BYD-implemented strategies, could explain the successful expansion of the emerging industry in China [
64]. Generally, China provides an average subsidy of
$10,000 per vehicle and with 770,000 EVs sold in 2017, China’s central and local governments spent a total of
$7.7 billion on EV subsidies that allowed the country to capture control of and dominate LiB manufacturing [
65].
3.2. Regression Analysis, Fixed and Random Effects Models and the Hausman Test
Regression analysis fixed and random effects models are widely used in the study of panel data to investigate the impact of government subsidies and firm R&D investments [
66,
67]. Consider the following cross-sectional multiple regression with explanatory variables
X1 and
X2:
where
X1 is the covariate of
X2, and vice versa. Covariates act as governing factors for a given variable. When control variables are present,
are partial regression coefficients and thus
represents the marginal effects of
X1 on
Y, keeping all other variables (e.g.,
X2) constant. In keeping
X2 constant, the marginal effect of
X1 on
Y is obtained after eliminating the linear effect of
X2 from both
X1 and
Y.
is explained identically. Consequently, multiple regression facilitates pure marginal effects to be obtained by including all relevant covariates and thus their heterogeneity can be controlled. If we further consider the multiple regression of the following time series with the same two explanatory variables,
X1 and
X2:
We have the same explanation for the marginal effects but can now track the system’s evolution over time. In order to account for time heterogeneity, we can combine Equations (1) and (2) to arrive at a pooled dataset, which forms panel data with the following panel regression:
Using a two-way error component assumption for perturbations, cross section and time heterogeneity can be controlled for:
where
is the unobserved individual (cross section) heterogeneity,
is the unobserved time heterogeneity, and
is the remaining random error term. The
and
are the within components, and the
is the panel or between components. Based on the assumptions made concerning these error components, i.e., whether they are fixed, or random, fixed, or random effect models are formed. Specifically, if
and
are fixed parameters to be estimated and
is independently distributed with zero mean and constant variance, a two-way fixed effects model is built. If, however,
and
are random, identical to the error term, and are all independent of each other and of explanatory variables, a two-way random effects model is built. If only one component is considered at a time, one-way fixed or random effects models can be built by replacing
(Equation (3)) and depending on either the fixed or random assumption of
and
, becomes:
In this paper, innovation in LiB is measured by the level of research and development (RD) investments and by patent output (PO). RD investments, as the name suggests, is expressed as the funds invested by LiB companies into technological advancements and reflects the willingness of these companies to innovate. Patent output (PO), by contrast, is expressed in terms of the number of patents applied for due to technological breakthroughs, rather than the actual number of patents received as there is a lag between patent application and authorization phases [
42,
68]. Government subsidies (GS) was chosen as this study’s main explanatory variable. If the growth of corporate RD investments is higher than the growth of GS, this implies that GS has a stimulating effect on RD investments. Correspondingly, if RD investments are lower than GS growth, this suggests that GS are not having their intended effect and subsides should be redesigned. All variables are listed in
Table 2.
Among the control variables, the RD investment of lithium battery companies is mainly affected by the size and operating conditions of the company. Main business income (MBI) reflects the sales and market share of LiB companies and can be used as a representative variable reflecting the scale of the company. If the business is not operating well, it will lead to a debt crisis, and the company must prioritize repayment of principal and interest, thereby reducing RD investment. The asset-liability ratio (LEV) is used to reflect the business status of the enterprise, and it is expressed as the proportion of total liabilities to total assets. As an emerging industry, the innovation of lithium battery companies is also affected by the level of organizational knowledge. The patent output of lithium companies is mainly affected by the company’s RD intensity. The RD investment ratio (DS) is used here to express the importance of LiB companies on RD. Generally, the greater the RD intensity of the company, more emphasis is placed on RD activities, the better the output of high-quality patents. The largest shareholders ratio (LSR) was chosen as an index, representing willingness of shareholders with the largest share in the company to invest in innovations, focusing on maximization of their income from selling high-quality inventions. Net Profit (NP) is used to reflect the amount of money that companies can spend directly on innovations and expresses the size of profit. Top 10 shareholders ratio (or major shareholders) (TTSR) as the largest shareholders ratio was used to represent the possible shareholders’ contribution to RD, limited to within 10 main distributors [
69].
To investigate the effects of GS on LiB innovation, we use a well-balanced panel dataset of 95 LiB-manufacturing companies that ranges from the year 2015 to 2018. To consider both fixed and random effects, the Hausman specification test is employed [
70]:
where RD is RD investments and PO is the patent output;
i = 1, 2, …,
n, represents different enterprises;
t = 1, 2, …,
n, represents time, and
represents random effects. The data excludes the impact of price factors. Patent number was collected from the patent information service platform network [
71], and the rest of the data comes from the China Stock Market & Accounting Research Database [
72]. The data consists of the company name, company number, PO, total assets, total liabilities, etc. To guarantee model robustness, although there is data for 178 companies, data curation was performed to exclude companies that had large tracts of missing data, resulting in a total dataset of 95 companies.
According to
Table 3, there is a large difference between the maximum value and the minimum value of RD investment, and the median is much smaller than the average value, indicating that the overall RD investment of LiB companies is relatively small, and the RD investment of a few lithium-ion battery companies has increased. The number of invention patents and RD investment shows a similar distribution, and the median is much lower than the average, indicating that the overall output of invention patents of LiB companies is relatively small and unevenly distributed. With the aid of the econometric model, the influence of government subsidies on the RD investment of LiB companies can be further tested. Considering the large discrepancies in values between variables, before all further analyses were conducted, the data was normalized to a 0 to 1 range.
To determine the most appropriate model amongst the fixed and random effects models, the Hausman or Durbin-Wu-Hausman (DWH) test is performed on panel data and determines the presence of endogeneity (predictor variables) in the panel model. If the p-value is greater than 0.05 (>0.05) then the random effects model is chosen. Alternatively, if the p-value is less than 0.05 (<0.05), then null hypothesis is rejected, and the fixed effect model is chosen.
3.3. Information Flow
To ascertain the bidirectionality of causality between GS and other variables, information flow (IF) is applied and studied. IF is a real physical notion developed and rigorized Liang [
43], to quantitively assess causality between two time series. Causality is measured by the rate of information transfer from a given variable’s time series, to another. Given two time series
X1 and
X2, the maximum likelihood estimator of IF from
X2 to
X1 is:
where
is the sample covariance between
and
, and
is the covariance between
and
. Information flow in the opposite direction, i.e.,
, through switching the indices 1 and 2. The units are in natural units of information (nats). IF is used primarily to quantitively assess causality between government subsidies and RD investment. For completeness, information flow is also calculated for all other variables and comparisons made. At this juncture, it is prudent to note that a standard procedure for testing the relative importance of a detected causality has been made available [
73]:
where the phase expansion in the
X1 direction is
and
is the random effect (note: this is unrelated to the random effects model). The larger the value, the more significant the causal relationship between
X2 and
X1. When the significance level is 0.1,
> 0.1 indicates that the causal relationship is significant. The relative importance of detected causality will also be measured.