How Does Firm-Level Economic Policy Uncertainty Affect Corporate Innovation? Evidence from China

Abstract

Innovation is the main driving force of the sustainable development of enterprises. Economic policy uncertainty has increased dramatically in recent years due to events such as COVID-19, which will alter the business environment of enterprises and ultimately affect their innovation behavior. How economic policy uncertainty will affect corporate innovation has become a crucial topic, but empirical studies have not reached consistent conclusions, and few have noted the heterogeneity of different firms’ perceptions of uncertainty. This study used a textual analysis approach to create firm-level economic policy uncertainty indicators from the texts of annual reports of Chinese A-share listed firms. Based on the effectiveness of our measure of economic policy uncertainty, we further examined its impact on firm innovation. We find that our uncertainty measure has negative effects on enterprise innovation activity, and this negative impact is more significant among non-state-owned enterprises, and firms with higher financial constraints and lower government subsidies. We extend the measurement of economic policy uncertainty from the micro level and provide some suggestions for policymakers at the macro level. In the period of increasing uncertainty in the external environment, the government should try to maintain the stability and transparency of economic policies, and provide more targeted policy support to enterprises, such as by broadening their financing channels and providing innovation subsidies.

1. Introduction

The concept of economic policy uncertainty (EPU) refers to the inability of economic agents to predict with certainty whether, when, and how governments will alter their current economic policies [1]. In order to avoid sharp economic downturns after the 2008 financial crisis, governments have modified their economic policies regularly, resulting in a considerable increase in global economic policy uncertainty. Innovation is often considered to be the driving force behind a country’s long-term economic growth [2], and the ability to innovate is essential to the sustainable development of enterprises. As economic policy changes frequently, the business environment will become turbulent and complicated, ultimately affecting corporate innovation. The existing research on the link between EPU and firm innovation, however, has reached no consistent conclusion, and few have noted the heterogeneity of firms’ perceptions of uncertainty. This paper will investigate the impact of economic policy uncertainty on firms’ innovation behavior by constructing a firm-level uncertainty perception index so as to address this gap.

A growing body of recent work investigates the effects of uncertainty on macroeconomic activities such as recession, trade, unemployment, etc. [3,4,5,6,7,8,9], as well as how it influences micro-firms’ decisions on investment and export [10,11,12,13,14]. The COVID-19 pandemic in 2020, in particular, has been the largest contemporary threat, causing unprecedented levels of global economic policy uncertainty [15]. One of the most crucial operational environments for enterprises is the macroeconomic environment, which has an impact on both strategic choices and management activities. Corporation innovation is characterized by lengthy investment cycles, substantial investment, and high irreversibility. When facing high uncertainty, firms may choose to postpone such expenditures, which has an impact on innovation output.

There is now a relatively rich pool of literature examining the impact of economic policy uncertainty on firm innovation. The real option theory states that innovation investment is highly irreversible, and uncertainty increases the value of waiting options. Therefore, when companies are confronted with uncertainty, they opt to postpone investments in innovation and lower the number of patent applications [16,17,18,19,20,21,22,23,24]. Another strand of the literature, based on the strategic growth options theory, contends that policy uncertainty creates both opportunities and risks for enterprises. If we take enterprise competition into account, when one company chooses to defer investments, other firms may seize the opportunity to invest more in innovation and consolidate their market position. This means that increased uncertainty can stimulate innovative behavior [25,26,27,28,29,30,31,32,33]. Existing research does not reach a clear conclusion about the relationship between economic policy uncertainty and enterprise innovation, mostly due to the inconsistent selection of economic policy uncertainty measures [3]. In addition to this, the present research ignores how businesses themselves feel about policy uncertainty. As economic policy uncertainty is not directly observable, it remains challenging to measure.

The EPU indicators have undergone a process of ongoing development. The first category of the literature employs economic or financial indicators that represent volatility to evaluate economic policy uncertainty indirectly, such as market volatility and economic volatility [34,35]. However, fluctuations in these indicators could also be due to changes in non-economic policy. National elections or the turnover of local officials have both been employed in prior research as proxies for uncertainty [16,36]. While these events are strictly exogenous, they can only capture one component of the source of uncertainty, and do not reflect policy uncertainty in non-election years. Additionally, there is a body of literature that uses indices of economic policy uncertainty constructed from news media texts. The economic policy uncertainty index developed by Baker et al. [37] has been widely used due to its better continuity and time variability. The authors measure economic policy uncertainty by counting the frequency of articles per month that include the terms “uncertainty”, “economy”, and “policy”. Further, Huang and Luk followed this method to construct a more suitable EPU for China using information from many top Chinese city newspapers [38]. However, these studies only evaluated EPU at the macro level in order to study how it influences company choices at the micro level, ignoring the reality that economic policies differ among industries and regions. More importantly, Bloom [34] argued that uncertainty caused by economic policy changes is a subjective perception of firms. Even with precisely the same macroeconomic policies, it is unlikely that various enterprises would perceive uncertainty in the same way. Therefore, it is crucial to measure uncertainty at the firm level.

Referring to Caldara et al. [39], we used a textual analysis approach to create a firm-level economic policy uncertainty index that varied with company and time. Based on the “Management Discussion and Analysis” (MD&A) part of Chinese A-share listed companies’ annual reports, we calculated the frequency of economic-policy-uncertainty-related terms to measure the firm-level economic policy uncertainty perception (FEPU). This approach mirrors the basic idea behind Baker’s construction of EPU, except that it is applied at the company level.

The method used in this research has precedent in the literature, where it is common practice to extract management information from public corporate documents, such as annual reports and earnings reports [40,41]. Hassan et al. [42] constructed the political risk faced by US companies using the share of time spent discussing political risk during earnings calls. Benguria et al. [43] used annual reports of Chinese listed companies to construct a firm-level trade policy uncertainty index. The MD&A consists of a summary of the company’s previous operating conditions and a consideration of potential risks associated with its future development. It is the most informative section of the annual report and plays a critical function in providing information [44]. Finally, the Securities and Futures Commission requires listed companies to disclose in their MD&A their external business environment, as well as risk factors that might adversely impact future growth. There are additional strict criteria for the annual report’s truthfulness, regularity, and accuracy.

Notably, we show that our FEPU has good properties in a variety of ways; for example, it closely tracks the evolution of the Chinese EPU index created by Baker et al. [37]. Furthermore, we investigate the effects of FEPU on Chinese enterprises’ innovation activities using data from Chinese listed companies from 2007 to 2020. Our findings suggest that enterprises’ perceptions of uncertainty have a negative effect on their innovation efforts. In other words, increased economic policy uncertainty at the firm level reduces the number of innovation patents that are filed and granted. Our results echo the real options theory. Such negative effects are robust to a series of tests. Moreover, we find that firms’ responses to FEPU are heterogeneous. In particular, the negative impact of FEPU on firm innovation is more obvious among non-state enterprises, firms with strong financing constraints, and firms with low government subsidies.

China provides an ideal environment for studying this issue. Firstly, compared with developed countries, China, as an emerging economy, is in a stage of transformation. The Chinese government tends to intervene and regulate resource allocation, and therefore has a strong influence on national economic activities [45,46]. At the same time, trade conflicts with the United States have increased economic uncertainty significantly. The Chinese government is frequently adjusting its economic policies to deal with the complicated economic environment and to meet the requirements of high-quality development. According to the index constructed by Baker et al. [37], the average value of China’s EPU Index for 2021 was 788.96. Secondly, China’s economy is transitioning from high-speed growth to high-quality development, and innovation is the key to high-quality development. Therefore, the Chinese government has implemented a series of policies to encourage enterprise innovation. It is essential for academic research in China as well as policymakers to have an awareness of how EPU influences corporate innovation.

This paper contributes to the literature in two ways. First, we construct an index of perceived economic policy uncertainty at the firm level using a textual analysis approach, contributing to the literature on measuring uncertainty. The existing research mainly constructs indicators evaluating economic policy uncertainty at the macro level, such as market volatility or economic volatility [34,35], and the changeover of local officials [16,36]. In addition, there is a large body of literature that uses news media text to construct an uncertainty index [22,23,47,48,49], most of which was developed by Baker et al. [37] and Huang and Luk [38]. Unlike these indices, our research demonstrates that firms have heterogeneous perceptions of EPU. The extent to which firms in different industries and regions perceive economic policy uncertainty varies considerably over time. Therefore, it is necessary to construct a firm-level economic policy uncertainty perception index. In addition, we extend the work of Yu et al. [50]. Their firm-level EPU was obtained by dividing the number of uncertainty-related words by the total number of words, while the FEPU constructed in this paper was obtained by dividing the number of uncertainty-related words by the number of words related only to economic policy. Only considering uncertainty and ignoring economic policy may lead to uncertainty for reasons other than economic policy. Moreover, these authors did not test the indicators’ rationality. We believe that it is critical to build indicators, but it is even more critical to test whether the indicators are valuable and can actually measure economic uncertainty at the firm level.

Secondly, we extend the literature on the impact of uncertainty on corporate innovation activities at the micro level. Although there is a growing body of literature examining the impact of economic policy uncertainty on firms’ innovative behavior, it does not reach consistent conclusions. Some of the literature, based on the real options theory, argues that economic policy uncertainty discourages firms from investing in innovation [16,17,18,19,20,21,22,23,24]. In contrast, another section of the literature argues that uncertainty in the external environment is actually what stimulates the competition of enterprises, and thus innovation [25,26,27,28,29,30,31,32,33]. Although previous research provides substantial evidence that economic policy uncertainty has important effects on firms’ innovative behavior, the majority of these studies use an indicator of macroeconomic policy uncertainty, which implies that all enterprises perceive the same uncertainty. Few studies have considered the heterogeneity of firms’ perceptions of economic policy. Our study addresses this gap by constructing FEPUs using the textual content of firms’ annual reports. We find that the heterogeneity of firms’ perceptions of uncertainty is important to consider, and, on average, FEPU has a negative correlation with business innovation. This means that firms will limit their filing and granting of invention patents when they perceive a higher level of uncertainty. Moreover, we find that perceived uncertainty has a selection effect on innovation, and our findings offer investors and policymakers more evidence that the impact of FEPU on innovation behavior may be affected by firm characteristics. State-owned companies may alleviate the negative effects of FEPU on the innovation output due to their superior endowment of resources. Moreover, we show that enterprises with greater government subsidies and fewer financing restrictions are less impacted by FEPU.

The rest of this paper is structured as follows: Section 2 reports on the related studies and hypothesis development. Section 3 introduces the data and methodology. Section 4 presents the baseline and robustness analysis results. Section 5 provides a further discussion, focusing on the selection effect. Section 6 gives the conclusions and policy recommendations.

2. Hypothesis Development

Economic policy uncertainty can affect corporate innovation in a variety of ways. One of the most commonly recognized models for explaining business investment decisions in an uncertain environment is the real options theory [1]. According to the theory, corporations regard investment choices as a series of options, and since uncertainty enhances the value of deferral options, it discourages present investment activity. Myers [51] proposed that the value provided by a strategic investment includes both the value created directly by the project and the value created by having the option to invest in future opportunities. Bernanke [52], following the theory of real options, converts the investment behavior of enterprises into a stochastic dynamic optimization issue. Investors have to make a choice between spot options and deferral options based on the assumptions of investment irreversibility and uncertainty. When the external environment becomes unclear and there is an irreversible investment, corporations often postpone investment. Firms’ current investment behavior will be hampered by rising economic policy uncertainty [1]. Given the highly irreversible nature of innovation investment and the complexities of the patent application procedure, increased economic policy uncertainty may lead corporations to reduce or postpone R&D spending, resulting in reduced innovation output [16,24]. Using cross-country industry data, Mbanyele and Wang [22] found that economic policy uncertainty hinders innovation, and this negative impact is smaller in countries with a better legal infrastructure.

Second, in times of high economic policy uncertainty, it may be difficult for enterprises to access financial resources. On the one hand, economic policy swings will make financing debt more difficult [53,54,55]. In the face of increased uncertainty, business risks for enterprises will grow, causing commercial banks to tighten credit availability and slow down their approval of loans. Simultaneously, commercial banks will raise lending rates to balance risk and return, exacerbating the financing constraints faced by businesses. On the other hand, economic policy volatility can make external investors more risk-averse, and therefore also make equity financing more difficult. As a result, increased economic policy uncertainty can adversely affect a company’s financial constraints [48], making it more difficult for them to access funding for innovation investment, which reduces the motivation for enterprise innovation and inhibits the innovation output [18,20,21]. Zhong et al. [56] argued that when the government frequently adjusts economic policies, creating high levels of economic policy uncertainty, banks and external investors will demand stringent financing conditions from firms, which significantly increases the financing costs for firms and thus leads to a decrease in their innovation ability.

Third, the managerial risk aversion hypothesis analyzes uncertainty from the standpoint of the manager. The strategic choice of the company depends on the risk aversion of the manager [57]. Since corporate managers possess substantial quantities of company shares or other securities, they are highly cautious when making long-term investment choices in firms when the external economic situation is unclear. As a result, when firm-level economic policy uncertainty rises, managers are more careful about making choices regarding innovation in order to reduce the risk of innovation failure. Panousi and Papanikolaou [58] found that if managers are risk-averse, they may underinvest as uncertainty increases, especially if they control a greater proportion of the company. Wen et al. [59] argued that when external uncertainty increases, management becomes more pessimistic about the future, and investment decisions are more conservative, meaning that EPU significantly reduces firms’ risk-taking. Lou et al. [23] found that higher economic policy uncertainty leads to a decrease in firms’ innovation output, and this negative correlation is mainly found in firms with lower appetites for risk among executives.

According to the analysis above, we argue that companies’ perceptions of uncertainty delay firms’ investments in innovation, thus reducing innovation output. We suggest the following hypothesis:

H1a. 

Firm-level economic policy uncertainty has a negative impact on corporate innovation.

However, the strategic option theory holds the opposite view, suggesting that uncertainty promotes firm innovation [28,30,32], which analyzes enterprise investment under the assumption of imperfect competition. The real option theory holds that when the uncertainty increases, enterprises can hold deferred options by delaying investment. However, the strategic growth theory shows that enterprises can hold the growth option by executing investment, because uncertainty is not only a risk but also an opportunity for enterprises. Kulatilaka and Perotti [60] found that if enterprises choose to postpone investment in the face of increased uncertainty, this is equivalent to giving investment opportunities to competitors, thus allowing them to be more profitable and competitive in the future. Therefore, when the external environment is more uncertain, enterprises with a spirit of risk and innovation may take greater risks and seize opportunities [47]. Vo and Le [25] argued that competition is a very important force in the positive relationship between uncertainty and investment. R&D investment is significantly and positively associated with firm competitiveness, so firms choose to increase R&D investment to offset the negative effects of higher economic policy uncertainty. This means that the earlier companies invest when economic policy uncertainty is high, especially in innovation, the better their chances of gaining a competitive advantage in the future [33].

Bloom [61] considers the impact of uncertainty on innovation in terms of adjustment costs. He argues that despite the negative impact of EPU on investment, employment, and productivity, it is important to take into account the differences in adjustment cost characteristics. Compared with other investments, innovation investment will respond differently to uncertainty due to different adjustment costs. R&D investment may show a relatively delayed response to changes in environmental uncertainty, so uncertainty does not necessarily hinder investment in innovation, but rather provides an incentive for firms to increase R&D investment [62]. Stein and Stone [63] found that uncertainty depresses capital investment, hiring, and advertising, but encourages R&D spending. Atanassov et al. [64] showed that uncertainty in government policy stimulates R&D at the firm level by using the timing of US gubernatorial elections as a source of exogenous variation in uncertainty. A recent study by He et al. [27] found that, overall, EPU is positively associated with firm innovation. In terms of time periods, EPU induced more innovative activity during the low-EPU period before 2008, but reduced firm innovation during the high-EPU period after 2008. Dividing investment into innovation investment and maintenance investment, Liu et al. [31] found that innovation and maintenance investment are mutually exclusive, and that firms are more willing to invest in innovation when faced with higher economic policy uncertainty.

According to the analysis above, we argue that companies’ perceptions of uncertainty promote firms’ investments in innovation, and thus increase innovation output. We suggest the following hypothesis:

H1b. 

Firm-level economic policy uncertainty has a positive impact on corporate innovation.

3. Data and Methods

3.1. Data Sources

Samples of China’s A-share listed firms from 2007 to 2020 have been chosen for this study. The Ministry of Finance mandates that listed firms follow the new accounting rules beginning in 2007; hence, we use 2007 as the beginning year of our sample. The patent data and basic company information were derived from CSMAR and Wind. The FEPU index was derived from the annual reports of all listed A-share companies from 2007 to 2020, which were obtained from the Juchao Information Websit (http://www.cninfo.com.cn/new/index (accessed on 20 March 2022)).

3.2. Measurement of Firm-Level Economic Policy Uncertainty

With the development of computer technology, it is common to introduce unstructured data, such as text, into corporate finance research. We searched the annual reports of all A-share listed companies in China from 2007 to 2020 via the Juchao Information Website. Through both procedural and manual verification, we removed the English versions of the annual reports and the versions that contained only the abstract. Some enterprises have updated their annual reports since their release, and only the latest versions are collected here. Based on a total of 36,867 annual reports, we constructed a time-varying firm-level economic policy uncertainty index. The construction method is similar to that of Caldara et al. [39], and consists of three steps, as shown in Figure 1.

3.2.1. Text Preprocessing

This paper uses Java to convert the annual report text in PDF (Portable Document Format) format to TXT format, as the TXT format allows for full editing. Compared with other studies that chose Python for conversion, we found that the pdfbox library of Java can convert PDF files more efficiently and accurately. Due to the complexity of the annual reports, the MD&A sections are taken for the evaluation and analysis of the current operation’s performance and future development trends outlined by the enterprise management. Therefore, this paper focuses on the MD&A sections of the annual reports.

The name of the MD&A section changed several times during the sample interval of 2007 to 2020. In 2007–2014, it was referred to as “Board Report”, in 2015 as “Management Discussion and Analysis”, and in 2016–2020 as “Discussion and Analysis of Operations”. The section after the MD&A is usually “Material Matters” or “Report of the Board of Supervisors”. Therefore, these words were selected to design regular expressions for the extraction of the MD&A. After extraction, the 109 annual reports that failed to yield MD&A were manually verified and the regular expressions were modified based on the reasons for the failure to extract MD&A content. After several rounds of amendments and supplements, the MD&A sections of all the annual reports were successfully extracted. Further, we checked for two conditions, in which the length of the MD&A accounted for too large or too small a proportion of the full text, respectively, which would mean that the program may extract content that should not be extracted, or omit some content. Of the 36,867 annual reports, the authors manually verified a total of 508 texts with MD&A text characters accounting for more than 25% of the total, and 252 texts with MD&A representing less than 5% of the full text. The regular expressions were also corrected again based on the cause of incorrect extraction. Finally, the authors randomly selected 5% of the annual reports from the uncorrected samples by year, resulting in a total of 1791 annual reports. All samples were checked manually and only four were found to be incorrect, indicating that the MD&A extraction procedure was already very accurate.

3.2.2. Construction of the Index

Referring to Baker et al. [37], this paper uses the “word list method” to analyze MD&A text and construct the firm-level economic policy uncertainty index. If economic policy-related and uncertainty-related words simultaneously appear in a sentence, it is assumed that this sentence is a statement by the company’s management that the company is facing economic policy uncertainty. Before constructing the index, we preprocessed the MD&A text—we removed the tables, headers, and footers that are not related to the analysis; we also eliminated all numbers, English letters, and all punctuation marks and special symbols other than full stops. This paper first divides the MD&A text into sentences with a full stop as a sentence break, and primarily considers the language habits of Chinese. Further, each sentence is segmented using the jieba module in Python. In order to reduce the ambiguity caused by word segmentation, we have set out a series of word lists, which include the full name and abbreviation of the listed company, the accounting subjects, and the word lists related to “economic policy” and “uncertainty”. After word segmentation, each sentence is traversed. If an economic-policy-related word appears in a sentence, it is considered to indicate economic policy (E), and the total number of words in this sentence is recorded as Ns. If an economic-policy-related word and an uncertainty-related word appear in a sentence at the same time, it is considered to indicate economic policy uncertainty (U). The total number of words indicating “uncertainty” in this sentence is ns. Finally, two approaches are adopted to measure the economic policy uncertainty perceived by enterprises; the first uses the proportion of the number of economic-policy-uncertainty-related sentences of the total number of economic policy sentences, whereby FEPUs is calculated as follows:

$${\mathrm{F}\mathrm{E}\mathrm{P}\mathrm{U}\mathrm{s}}_{\mathrm{i},\mathrm{t}}=\frac{{\mathrm{t}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l} \mathrm{n}\mathrm{u}\mathrm{m}\mathrm{b}\mathrm{e}\mathrm{r} \mathrm{o}\mathrm{f} \mathrm{U}}_{\mathrm{i},\mathrm{t}}}{{\mathrm{t}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l} \mathrm{n}\mathrm{u}\mathrm{m}\mathrm{b}\mathrm{e}\mathrm{r} \mathrm{o}\mathrm{f} \mathrm{F}}_{\mathrm{i},\mathrm{t}}}$$

(1)

In the robustness check, we scale the proportion of the total number of uncertain words in economic-policy-uncertainty-related sentences to the total number of words in economic policy sentences. FEPUw is calculated as follows:

$${\mathrm{F}\mathrm{E}\mathrm{P}\mathrm{U}\mathrm{w}}_{\mathrm{i},\mathrm{t}}=\frac{{\mathrm{n}}_{\mathrm{s},\mathrm{i},\mathrm{t}}}{{\mathrm{N}}_{\mathrm{s},\mathrm{i},\mathrm{t}}}$$

(2)

3.2.3. Properties of the Index

Next, we test the effectiveness of the firm-level economic policy uncertainty indicator.

First, we compare our new indices to those developed by Baker et al. [37], which are the most commonly used indicators of economic policy uncertainty. We separately calculate the annual average of our FEPU index and the monthly EPU index constructed by Baker et al. [37]. In Figure 2a, we see that our FEPU closely tracks the EPU constructed by Baker et al. [37]. The correlation coefficient between the two is 0.882. However, there are also notable differences: the EPU index jumps significantly in 2016, while the jump in FEPUs occurs in 2015; the EPU index makes a small recovery in 2014, but the FEPUs continue to drop. This means that FEPUs provide a relatively objective description of macroeconomic policy uncertainty, but the economic policy uncertainty perceived by firms is slightly different from the actual economic policy uncertainty. At the same time, Jurado et al. [65] and Scotti [66] have shown that economic policy uncertainty is highly counter-cyclical. Figure 2b shows that the FEPU and GDP growth rates have opposite trends, and the correlation coefficient is −0.628.

Secondly, Baginski et al. [67] and Diether et al. [68] concluded that uncertainty is positively related to the degree of divergence of individuals’ economic expectations. We regress FEPUs on year dummy variables, industry dummy variables, and the interaction term of the two. After removing the influence of industry characteristics, we analyze the correlation between FEPU and its dispersion. As shown in Figure 3, the mean values of FEPUs and the standard deviations of their residuals change in similar ways, with the correlation coefficient between the two as high as 0.886. This shows that FEPUs are able to effectively portray differences in economic policy uncertainty at the firm level.

Finally, this paper further regresses FEPUs on the year × industry dummy variables and year × province dummy variables, so as to eliminate the influence of industry and regional factors that change over time on FEPU. Figure 4 shows the probability density function of the FEPU residuals, which differs significantly from the normal distribution. Furthermore, we also use Shapiro–Wilk and Shapiro–Francia statistics to test whether FEPU residuals conform to a normal distribution, and both results reject the null hypothesis of obeying a normal distribution. The findings reveal that even firms in the same region and industry perceive varying levels of economic policy uncertainty, which may be influenced by individual company features (such as the ability to obtain and process information).

3.3. Dependent Variables

Some studies use R&D spending as a measure of innovation [69,70,71]; however, this measure only explains the intensity or input of innovation, and does not show its success or failure. In line with the innovation literature [72,73,74,75], we use patent data (i.e., outputs of R&D) rather than R&D expenditures (i.e., inputs to R&D) to capture firms’ actual innovation activities in China. Nevertheless, we continue to utilize R&D expenditure and innovation efficiency as proxies for innovation activity in robustness tests.

There are three types of patents in China: innovation patents, utility model patents, and appearance design patents. Appearance design and utility model patents imply a modification to an existing product’s look or functionality. Invention patents often require higher R&D expenses, technical complexity, and filing difficulties, and they are thus a better indication of listed firms’ innovation efforts [26,76]. Subsidized patent filing fees may encourage the over-filing of low-quality patents. To ease this problem, we focus on patent applications for inventions, while also taking into account those that are awarded.

Considering the right-skewed distribution of patent application data, we add 1 to the total number of invention patent applications, and take its natural logarithm as the enterprise innovation, following previous research [77,78,79].

3.4. The Empirical Model

To measure the impact of FEPU on corporate innovation, we construct the following regression model:

$${\mathrm{l}\mathrm{n}\mathrm{n}\mathrm{o}\mathrm{v}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_{\mathrm{i},\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{F}\mathrm{E}\mathrm{P}\mathrm{U}}_{\mathrm{i},\mathrm{t}-1}+\sum _{\mathrm{k}}{\mathsf{\beta }}_{\mathrm{k}}{\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{\mathrm{k},\mathrm{i},\mathrm{t}-1}+{\mathsf{\lambda }}_{\mathrm{f}}+{\mathsf{\lambda }}_{\mathrm{y}}+{\epsilon }_{\mathrm{i},\mathrm{t}}$$

(3)

where i represents firms, and t represents years. ${\mathrm{l}\mathrm{n}\mathrm{n}\mathrm{o}\mathrm{v}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_{\mathrm{i},\mathrm{t}}$ denotes the innovation outputs, which is measured by the number of invention patent applications and the number of those granted. Considering the lag of innovation activities, and in order to reduce the endogenous problem, all explanatory and control variables are lagged by one period. We control the firm and year fixed effects, and robust standard errors are clustered at the firm level.

Referring to the previous literature [71,80,81,82,83], we control for the following variables in the model that may affect innovation: firm scale, which is denoted as Size; return on assets, which is denoted as ROA; asset–liability ratio, which is denoted as Lev; firm age, which is denoted as Age; Tobin Q value, which is denoted as TobinQ; and fixed assets ratio, which is denoted as FA. Variable names and definitions are shown in

Table 1. Measurements of variables.

Variable Name	Symbol	Measurement
FEPUs	FEPUs	The proportion of the number of economic policy uncertain sentences to the number of economic policy sentences
FEPUw	FEPUw	The proportion of the total number of uncertain words in economic policy uncertain sentences to the total number of words of economic policy sentences
Patent_application	Patent_application	Natural logarithm of (number of invention patent applications + 1)
Patent_award	Patent_award	Natural logarithm of (number of invention patent granted + 1)
R&D expenditure	R&D	Research and development expenditure/operating income
Innovation Efficiency	InnoEff	The percentage of invention patent applications in R&D expenditure
Firm scale	Size	Natural logarithm of total assets
Return on Assets	ROA	Net profits/average balance of total assets
Asset–liability ratio	Lev	Year-end total liabilities/year-end total assets
Firm age	Age	Sample year − establishment year + 1
Tobin Q value	TobinQ	(Market value of tradable shares + number of non − tradable shares × net assets per share + book value of liabilities)/total assets
Fixed assets ratio	FA	Fixed assets/total assets
Firm ownership	SOE	State-owned enterprises, SOE = 1; if not, SOE = 0
Financing constraints	SA	Absolute value of ($-0.737\times \mathrm{S}\mathrm{i}\mathrm{z}\mathrm{e}+0.043\times {\mathrm{S}\mathrm{i}\mathrm{z}\mathrm{e}}^2-0.04\times \mathrm{A}\mathrm{g}\mathrm{e}$)
Government subsidy	Subsidy	Government subsidy/year-end total assets

.

In addition, we eliminate (1) financial services firms, (2) firms receiving special treatment (ST), (3) samples lacking control variables, and (4) companies established in the year of observation. All the continuous variables are winsorized at 1%. The total sample consists of 28,524 observations for firm-years.

4. Empirical Results

4.1. Descriptive Statistics of the Main Variables

Table 2. Descriptive statistics.

	(1)	(2)	(3)	(4)	(5)	(6)
Variables	N	Mean	sd	Min	p50	Max
FEPUs	28,524	0.265	0.151	0	0.250	0.812
FEPUw	28,524	0.012	0.009	0	0.010	0.059
Patent_application	28,524	2.344	1.750	0	2.485	7.007
Patent_award	28,524	2.170	1.675	0	2.197	6.843
R&D	22,674	0.043	0.044	0	0.034	0.281
InnoEff	22,674	0.154	0.084	0	0.164	0.352
Size	28,524	22.176	1.290	19.390	21.999	26.511
ROA	28,524	0.039	0.066	−0.415	0.038	0.245
Lev	28,524	0.436	0.205	0.027	0.432	0.924
Age	28,524	17.984	5.913	2.000	18.000	63.000
TobinQ	28,524	2.035	1.390	0.792	1.603	17.653
FA	28,524	0.221	0.164	0.001	0.187	0.769
SOE	28,524	0.387	0.487	0	0	1
SA	28,524	3.802	0.252	2.988	3.800	4.555
Subsidy	28,524	0.437	0.502	0	0.283	3.745

reports the descriptive statistics, including the mean, standard deviation, minimum, median, and maximum values of the main variables. As shown in Table 2, the mean value and the standard deviation of the number of patent applications are 2.344 and 1.750, respectively. The same is true of the number of patents granted, which shows that companies differ greatly in terms of innovation. It is worth noting that the maximum and minimum values of FEPUs are 0.812 and 0, respectively, while the average value is 0.265, which indicates that FEPUs fluctuate a lot among the sample companies. This means that the perceived sensitivity to economic policy uncertainty varies across firms. This result attests to the necessity of constructing firm-level indicators of perceived economic policy uncertainty. For other variables, the summary statistics are also within a reasonable range.

4.2. Baseline Results

Table 3. Basic regression results.

	Patent_Application		Patent_Award
	(1)	(2)	(3)	(4)
FEPUs	−0.169 ***	−0.141 **	−0.113 **	−0.102 *
	(0.06)	(0.06)	(0.06)	(0.05)
Size		0.416 ***		0.406 ***
		(0.03)		(0.03)
ROA		0.923 ***		0.295 *
		(0.15)		(0.15)
Lev		−0.235 **		−0.083
		(0.10)		(0.10)
Age		−0.790 ***		0.262 ***
		(0.02)		(0.02)
TobinQ		0.021**		0.018 **
		(0.01)		(0.01)
FA		0.271**		0.410 ***
		(0.12)		(0.12)
Constant	2.295 ***	10.869 ***	2.930 ***	−12.325 ***
	(0.03)	(0.71)	(0.02)	(0.69)
Firm Fe	Yes	Yes	Yes	Yes
Year Fe	Yes	Yes	Yes	Yes
Observations	28,524	28,524	28,524	28,524
R2	0.211	0.240	0.245	0.275

Note: Robust standard errors are shown in parentheses, * represents p < 0.10, ** represents p < 0.05, and *** represents p < 0.01.

reports the main findings of the baseline regression regarding the effects of firms’ perceived economic policy uncertainty on their innovation output. In columns (1) and (2), the dependent variable is set as the number of invention patent applications. Controlling for firm and year fixed effects, the coefficient of FEPUs is negative and significant at the 1% level in column (1), indicating that when the economic policy uncertainty perceived by enterprises increases, the output of innovation activities will decrease. After controlling for other firm-level variations in column (2), the regression coefficient of FEPUs is −0.141 and is significant at the 5% level. On average, a one-standard-deviation increase in FEPUs reduces the number of invention patent applications by 0.9% in the following year. In columns (3) and (4), the dependent variable is the number of invention patents granted. We find that the coefficient of FEPU is still negative and significant at the 5% and 10% levels, respectively. The above results show that, although the degree of perceived policy uncertainty varies across firms, on average, firms tend to delay innovation activities when faced with higher levels of policy uncertainty. Our results are consistent with the claims of the real options theory, which means that firms reduce investment in high-risk and irreversible projects when there is a high degree of uncertainty. Our findings are consistent with recent studies that argue that economic policy uncertainty hinders firm innovation [18,20,22,23,24].

With regard to the control variables, the Size, ROA, Tobin Q, and FA coefficients are all significantly positive, indicating that firms with larger sizes, higher profitability, and higher investments in fixed assets are more innovative. This finding is consistent with the existing literature [78,79,83]. Further, the asset–liability ratio is positively related to innovation, indicating that when enterprise liabilities increase, enterprise innovation activities are more inclined to reduce.

4.3. Robustness Tests

To further verify the reliability of our results, we conduct a series of robustness tests.

4.3.1. Time × Province Fixed Effects and Time × Industry Fixed Effects

Although we control for firm fixed effects and year fixed effects in the benchmark regressions, unobservable factors at the industry and region level over time may also affect the results of the benchmark regressions. Following Cui et al. [49] and Zhang et al. [84], to make our regression results more credible, we further add year × province fixed effects and year × industry fixed effects to the benchmark regression model.

Table 4. Robustness result with year × province fixed effects and year × industry fixed effects.

	Patent_Application		Patent_Award
	(1)	(2)	(3)	(4)
FEPUs	−0.160 ***	−0.131 **	−0.126 **	−0.112 **
	(0.06)	(0.06)	(0.05)	(0.05)
Size		0.432 ***		0.424 ***
		(0.03)		(0.03)
ROA		0.971 ***		0.279 *
		(0.15)		(0.15)
Lev		−0.189 **		−0.075
		(0.09)		(0.09)
Age		−0.117 **		−0.166 ***
		(0.05)		(0.04)
TobinQ		0.021 **		0.018 **
		(0.01)		(0.01)
FA		0.144		0.221 **
		(0.11)		(0.11)
Constant	1.780 ***	−6.252 ***	1.672 ***	−5.690 ***
Firm Fe Year × Province Fe Year × Industry Fe	Yes Yes Yes	Yes Yes Yes	Yes Yes Yes	Yes Yes Yes
Observations	28,524	28,524	28,524	28,524
R2	0.263	0.292	0.303	0.332

Note: Robust standard errors are in parentheses, * represents p < 0.10, ** represents p < 0.05, and *** represents p < 0.01.

shows that the main conclusions from the basic regressions do not change. When the effect of unobservable factors is addressed, firm-level economic policy uncertainty has a significant negative impact on the number of invention patent applications and authorizations, which is significant at the 5% level. This means that, although uncertainty is felt differently among firms, higher uncertainty on average discourages innovative behavior. The results also provide further evidence that differences in the perceptions of economic policy uncertainty between firms cannot be fully explained by industry differences and regional differences, and that it is necessary to construct indicators of perceived economic policy uncertainty at the firm level.

4.3.2. Instrumental Variable (IV) Method

There may be a reverse causal relationship between the firm-level economic policy uncertainty and the innovation output. That is, enterprises with low innovation output may deliberately exaggerate the uncertainty they face in their annual reports, which brings about endogenous problems. This paper uses a first-order lag in the number of patent applications in all the regressions, which can mitigate the endogeneity problem to some extent. To further alleviate the endogenous problem, we also adopt the instrumental variable method. This paper uses the average firm-level uncertainty of other firms in the same industry, city, and year as an instrumental variable [20]. This variable meets the criteria of relevance and exogeneity. The average perception of these companies is related to the perceived level of the enterprise, and will not directly affect the innovation output of the enterprise. We use 2SLS regression to re-estimate the relationship between firm-level economic policy uncertainty and innovation output.

Table 5. Instrumental variable (IV) method.

	Patent_Application		Patent_Award
	(1)	(2)	(3)	(4)
FEPUs	−0.292 ***	−0.257 ***	−0.187 **	−0.174 **
	(0.09)	(0.09)	(0.08)	(0.08)
Size		0.416 ***		0.405 ***
		(0.02)		(0.01)
ROA		0.908 ***		0.286 **
		(0.13)		(0.12)
Lev		−0.233 ***		−0.081
		(0.06)		(0.06)
Age		−0.788 ***		0.263 ***
		(0.02)		(0.02)
TobinQ		0.021 ***		0.018 ***
		(0.01)		(0.01)
FA		0.270 ***		0.409 ***
		(0.07)		(0.07)
Year Fe Firm Fe	Yes Yes	Yes Yes	Yes Yes	Yes Yes
Observations	28,524	28,524	28,524	28,524
R2	0.211	0.240	0.245	0.275
Cragg-Donald Wald F	9015.909	8982.609	9015.909	8982.609

Note: Robust standard errors are in parentheses, ** represents p < 0.05, and *** represents p < 0.01.

shows the results of the second-stage regression. It can be seen that FEPUs still adversely impact the innovation output of enterprises after mitigating the possible endogenous problems. This result is consistent with our previous hypothesis that firms will be more cautious about investing in innovation in an uncertain environment, thus delaying their innovative behavior.

4.3.3. Alternative Measure of FEPU

Following Yu et al. [50], for the robustness check, we adopt another measure of economic policy uncertainty perceived by enterprises in

Table 6. Estimation results of alternative measure of FEPU.

	Patent_Application		Patent_Award
	(1)	(2)	(3)	(4)
FEPUw	−2.733 ***	−2.241 **	−2.364 **	−2.026 **
	(1.06)	(1.01)	(0.98)	(0.93)
Size		0.416 ***		0.405 ***
		(0.03)		(0.03)
ROA		0.933 ***		0.300 **
		(0.15)		(0.15)
Lev		−0.236 **		−0.083
		(0.10)		(0.10)
Age		−0.790 ***		0.262 ***
		(0.02)		(0.02)
TobinQ		0.021**		0.018 **
		(0.01)		(0.01)
FA		0.271**		0.410 ***
		(0.12)		(0.12)
Constant	2.283 ***	10.868 ***	2.930 ***	−12.325 ***
	(0.02)	(0.71)	(0.02)	(0.69)
Year Fe Firm Fe	Yes Yes	Yes Yes	Yes Yes	Yes Yes
Observations	28,524	28,524	28,524	28,524
R2	0.211	0.240	0.245	0.275

Note: Robust standard errors are in parentheses, ** represents p < 0.05, and *** represents p < 0.01.

—the proportion of the total number of uncertain words in economic policy uncertain sentences out of the total number of economic policy sentences. Using different measurements of the FEPU index does not result in a significant change, demonstrating that FEPU has a strong restraining influence on corporate innovation output. This result is consistent with H1a; that is, firm-level economic policy uncertainty has a negative impact on corporate innovation.

4.3.4. Alternative Measure of Innovation

In addition to replacing the dependent variables, this paper adjusts the measurement for innovation activity. Compared with the value of innovation output, R&D investment is a measure of innovation input, which we also use as a proxy for corporate innovation activity [69,70,71]. In addition, innovation efficiency is also suitable for use in measuring the innovation of enterprises [85,86]. This paper uses the percentage of invention patent applications in R&D investments to measure innovation efficiency. In columns (1) to (2), the dependent variable is set as the ratio of R&D investment to total assets. In columns (3) to (4), the dependent variable is set as innovation efficiency.

Table 7. Estimation results of the alternative measure of innovation.

	R&D		InnoEff
	(1)	(2)	(3)	(4)
FEPUs	−0.003 *	−0.003 *	−0.012 ***	−0.010 ***
	(0.00)	(0.00)	(0.00)	(0.00)
Size		0.002 ***		0.020 ***
		(0.00)		(0.00)
ROA		−0.022 ***		0.047 ***
		(0.01)		(0.01)
Lev		−0.028 ***		−0.012 *
		(0.00)		(0.01)
Age		0.002 ***		−0.047 ***
		(0.00)		(0.00)
TobinQ		0.000		0.001 *
		(0.00)		(0.00)
FA		−0.006		0.011
		(0.00)		(0.01)
Constant	0.050 ***	−0.024	0.139 ***	0.724 ***
	(0.00)	(0.02)	(0.00)	(0.04)
Year Fe Firm Fe	Yes Yes	Yes Yes	Yes Yes	Yes Yes
Observations	22,674	22,674	22,674	22,674
R2	0.079	0.094	0.153	0.173

Note: Robust standard errors are in parentheses, * represents p < 0.10, and *** represents p < 0.01.

shows that the economic policy uncertainty perceived by enterprises will not only significantly reduce innovation investment, but also reduce innovation efficiency. The empirical finding demonstrates that, even when the measure of innovation is changed, the relationship between FEPU and corporate innovation is still robust. In addition to this, our findings are consistent with the prediction of the real options theory, which predicts that, when firms face higher uncertainty, they will delay innovation investment, which further affects innovation output.

4.3.5. Estimation Using a Tobit model

We mainly use the two-way fixed effect model to examine the relationship between FEPU and corporate innovation in the basic regression. Since the number of some patents is 0, we further use the Tobit model to retest. We control for year fixed effects, industry fixed effects, and province fixed effects in all regressions. The results in

Table 8. Estimation results using the Tobit model.

	Patent_Application		Patent_Award
	(1)	(2)	(3)	(4)
FEPUs	−0.652 ***	−0.711 ***	−0.522 ***	−0.593 ***
	(0.08)	(0.07)	(0.07)	(0.07)
Size		0.686 ***		0.644 ***
		(0.01)		(0.01)
ROA		1.676 ***		0.881 ***
		(0.19)		(0.18)
Lev		0.140 **		0.172 ***
		(0.07)		(0.06)
Age		−0.022 ***		−0.022 ***
		(0.00)		(0.00)
TobinQ		−0.002		−0.010
		(0.01)		(0.01)
FA		−1.859 ***		−1.810 ***
		(0.07)		(0.07)
Year Fe	Yes	Yes	Yes	Yes
Industry Fe	Yes	Yes	Yes	Yes
Province Fe	Yes	Yes	Yes	Yes
Observations	28,524	28,524	28,524	28,524
r2_p	0.103	0.159	0.114	0.170
chi2	11,442.540	17,612.409	12,352.962	18,365.564

Note: Robust standard errors are in parentheses, ** represents p < 0.05, and *** represents p < 0.01.

show that when firms perceive an increase in economic policy uncertainty, their innovation output will decrease, both in terms of the number of patent applications and the number of those granted. This means that our conclusions remain the same, that uncertainty does reduce firm innovation even if we change the estimation model.

4.3.6. Use of Full Sample

The sample in this paper excludes ST companies. Given the relatively weak resistance of these companies to risk, their exclusion may lead to a loss of generality due to the generally high risk-resistance of the sample. Therefore, these companies have been reintroduced into the sample for testing, and the results in

Table 9. Estimation results using full sample.

	Patent_Application		Patent_Award
	(1)	(2)	(3)	(4)
FEPUs	−0.163 ***	−0.139 **	−0.102 *	−0.092 *
	(0.06)	(0.06)	(0.05)	(0.05)
Size		0.382 ***		0.378 ***
		(0.03)		(0.03)
ROA		0.900 ***		0.373 ***
		(0.13)		(0.13)
Lev		−0.108		−0.002
		(0.08)		(0.08)
Age		−0.782 ***		0.254 ***
		(0.02)		(0.02)
TobinQ		0.024 ***		0.021 ***
		(0.01)		(0.01)
FA		0.267 **		0.377 ***
		(0.11)		(0.11)
Constant	2.215 ***	11.458 ***	2.848 ***	−11.650 ***
	(0.02)	(0.64)	(0.02)	(0.62)
Year Fe Firm Fe	Yes Yes	Yes Yes	Yes Yes	Yes Yes
Observations	29,923	29,923	29,923	29,923
R2	0.203	0.235	0.234	0.266

Note: Robust standard errors are in parentheses, * represents p < 0.10, ** represents p < 0.05, and *** represents p < 0.01.

show that the main findings remain unchanged. For the full sample of firms, perceived higher uncertainty will cause firms to reduce their innovative behavior.

5. Additional Analysis

It is important to understand the heterogeneity of firms’ responses to FEPU so that policymakers can be more effective in reducing the negative effects of uncertainty on firm innovation.

5.1. Ownership Control Effect

In China, the attributes of corporate ownership have been an important cause of corporate heterogeneity. State-owned and non-state-owned enterprises differ greatly in their resource endowments, including credit resources and information resources. Chinese firms have relatively limited sources of financing, with bank loans serving as the primary source of financial funding. Compared with non-state-owned enterprises, state-owned enterprises may not only have easier access to credit support from banks [87,88,89], but can also borrow at preferential interest rates [90]. Considering that SOEs are to some extent responsible for the policy tasks of national development, they generally receive more government subsidies. In addition to credit resources, SOEs are able to access more information resources than non-SOEs by virtue of their political connections. SOEs can generally grasp policy directions more accurately, and have greater access to information because of their close ties with the government [27], thus reducing the impact of uncertainty. In contrast, non-state firms are more sensitive to the market environment. As a result, these companies have to continually optimize their resource allocation in order to survive the serious competition on the market. Therefore, changes in FEPU may have varying degrees of impact on the innovation of state-owned enterprises and non-state-owned enterprises.

We add the interaction term between the FEPU and SOE dummy variables in the base regression, so as to examine the disparities in the influence of economic policy uncertainty on innovation among enterprises of different ownership. We are investigating the effect of heterogeneity on firm ownership; hence, we concentrate on the sign of the interaction term’s coefficient. The results in

Table 10. Selective effects: ownership.

	Patent_Application		Patent_Award
	(1)	(2)	(3)	(4)
FEPUs	−0.284 ***	−0.252 ***	−0.216 ***	−0.202 ***
	(0.07)	(0.07)	(0.07)	(0.07)
FEPUs × SOE	0.315 ***	0.305 ***	0.280 ***	0.275 ***
	(0.12)	(0.11)	(0.11)	(0.10)
Size		0.416 ***		0.406 ***
		(0.03)		(0.03)
ROA		0.918 ***		0.290 *
		(0.15)		(0.15)
Lev		−0.236 **		−0.084
		(0.10)		(0.10)
Age		−0.790 ***		0.261 ***
		(0.02)		(0.02)
TobinQ		0.022 **		0.018 **
		(0.01)		(0.01)
FA		0.272 **		0.411 ***
		(0.12)		(0.12)
Constant	2.294 ***	10.883 ***	2.930 ***	−12.313 ***
	(0.03)	(0.71)	(0.02)	(0.69)
Year Fe Firm Fe	Yes Yes	Yes Yes	Yes Yes	Yes Yes
Observations	28,524	28,524	28,524	28,524
R2	0.212	0.241	0.246	0.275

Note: Robust standard errors are in parentheses, * represents p < 0.10, ** represents p < 0.05, and *** represents p < 0.01.

show that the cross-term coefficient between the FEPU and SOE dummy variables is significantly positive at the 1% level, which indicates that the negative impact of economic policy uncertainty perceived by enterprises on innovation output is more obvious in non-SOEs, consistent with the findings of Lou et al. [27] and Cui et al. [20]. SOEs can use their resource and information endowments—for example, their easier access to loans—to mitigate the negative effects of economic policy uncertainty on innovative behavior. Therefore, in a period of high uncertainty, the government should pay more attention to non-state-owned enterprises. By providing them with targeted preferential tax policies or innovation subsidies, the government can help non-state-owned enterprises in their ability to carry out innovation activities.

5.2. Financial Constraints Effect

Enterprises generally need a lot of capital to innovate and obtain patents. Strict financial constraints will have a negative impact on firms’ innovative activities [91,92,93]. Firms with low financial constraints have sufficient sources of funding to promote innovation. Information asymmetry will be exacerbated as economic policy uncertainty increases. Credit institutions are more willing to provide funds to financially steady companies, making it more difficult for companies with large financing constraints to obtain external funding. These enterprises are more inclined to hold cash and financial assets, and are motivated by precautionary saving, resulting in a reduction in innovation investment and output.

We use the SA index to measure the financial constraints of enterprises. Similarly, we add the interaction terms of FEPU and SA to the basic regression. In columns (1)–(4) of

Table 11. Selective effects: financial constraints.

	Patent_Application		Patent_Award
	(1)	(2)	(3)	(4)
FEPUs	3.058 ***	3.048 ***	3.228 ***	3.076 ***
	(0.91)	(0.88)	(0.84)	(0.82)
FEPUs × SA	−0.851 ***	−0.841 ***	−0.881 ***	−0.838 ***
	(0.24)	(0.23)	(0.22)	(0.22)
Size		0.415 ***		0.404 ***
		(0.03)		(0.03)
ROA		0.937 ***		0.309 **
		(0.15)		(0.15)
Lev		−0.236 **		−0.084
		(0.10)		(0.10)
Age		−0.778 ***		0.273 ***
		(0.02)		(0.02)
TobinQ		0.019 **		0.016 **
		(0.01)		(0.01)
FA		0.259 **		0.398 ***
		(0.12)		(0.12)
Constant	2.335 ***	10.686 ***	2.972 ***	−12.508 ***
	(0.03)	(0.72)	(0.03)	(0.69)
Year Fe Firm Fe	Yes Yes	Yes Yes	Yes Yes	Yes Yes
Observations	28,524	28,524	28,524	28,524
R2	0.212	0.241	0.246	0.276

Note: Robust standard errors are in parentheses, ** represents p < 0.05, and *** represents p < 0.01.

, we see that the coefficients of the interaction terms are significantly negative at the level of 1%. That is, the greater the financing constraints of enterprises, the greater the negative impact of FEPU on their innovation output. This finding is consistent with the existing literature [18]. Enterprises with greater financing constraints are less prone to innovative activities, and when they face a high degree of economic policy uncertainty, their innovative activities are more vulnerable to uncertainty compared to those with fewer financing constraints. Therefore, the government should accelerate the development of the financial market and the system of rational allocation of credit resources, so as to broaden the financing channels of enterprises.

5.3. Government Subsidies Effects

As a complementary mechanism to the market, government subsidies are one of the most important policy tools used to stimulate enterprise innovation [16,94]. On the one hand, government subsidies can directly compensate for the losses incurred in the process of innovation; on the other hand, they can create more financing channels for enterprises by giving assurances of their quality. Therefore, the negative impact of FEPU on innovation will be lower for companies with more government subsidies.

It can be seen from

Table 12. Selective effects: government subsidies.

	Patent_Application		Patent_Award
	(1)	(2)	(3)	(4)
FEPUs	−0.186 ***	−0.174 ***	−0.170 ***	−0.175 ***
	(0.07)	(0.06)	(0.06)	(0.06)
FEPUs × Subsidy	0.043	0.082	0.141 **	0.182 ***
	(0.06)	(0.06)	(0.06)	(0.05)
Size		0.417 ***		0.408 ***
		(0.03)		(0.03)
ROA		0.916 ***		0.280 *
		(0.15)		(0.15)
Lev		−0.236 **		−0.084
		(0.10)		(0.10)
Age		−0.790 ***		0.260 ***
		(0.02)		(0.02)
TobinQ		0.021 **		0.018 **
		(0.01)		(0.01)
FA		0.264 **		0.396 ***
		(0.12)		(0.12)
Constant	2.295 ***	10.868 ***	2.929 ***	−12.328 ***
	(0.03)	(0.71)	(0.02)	(0.69)
Year Fe Firm Fe	Yes Yes	Yes Yes	Yes Yes	Yes Yes
Observations	28,524	28,524	28,524	28,524
R2	0.211	0.240	0.246	0.275

Note: Robust standard errors are in parentheses, * represents p < 0.10, ** represents p < 0.05, and *** represents p < 0.01.

that the coefficient of the interaction term between FEPU and government subsidies is positive, which shows that government subsidies can alleviate the negative impact of FEPU on enterprise innovation. The coefficients on the interaction terms in columns (3) to (4) are significant at the 5% level, showing that the negative impact of FEPU on the number of invention patents granted is relatively lower for enterprises that receive more government subsidies compared to those that receive fewer government subsidies. Although the coefficient on the interaction term in columns (1) and (2) is not significant, it still does not affect our conclusion, because its coefficient is positive. This suggests that government subsidies can encourage firms to channel their investments into innovation in an uncertain environment. Our findings are consistent with recent studies that argue that the negative influence of uncertainty is more obvious for firms receiving fewer government subsidies [16,20].

6. Conclusions

Policy adjustments will change the external environment in which an enterprise operates and have a significant impact on enterprise behavior. In recent years, as economic policy uncertainty has increased worldwide, its impact on firms’ innovative behavior has attracted widespread scholarly attention. These studies focus on the impact of macro-level economic policy uncertainty on enterprise innovation, and few studies pay attention to the heterogeneity of uncertainty perception at the firm level. Our goal is to fill this gap by constructing a firm-level index of perceived economic policy uncertainty.

This study examined how economic policy uncertainty at the firm level affects enterprise innovation output. We constructed a time-varying index of firms’ perceptions of economic policy uncertainty using the texts of annual reports of Chinese A-share listed companies from 2007 to 2020. We used textual analysis methods to construct our FEPU based on the proportion of economic-policy-uncertainty-related words in MD&A texts. Our indicators closely track the EPU index constructed by Baker et al. [37], and we also verified the rationality and necessity of the indicators through a series of approaches. Furthermore, we used our FEPU to investigate how firm-level economic policy uncertainty affects firms’ innovative behavior, and to study its heterogeneous effects. We found that enterprises will choose to reduce their innovation output—both the number of invention patent applications and the number of grants—when faced with increased economic policy uncertainty. Our findings are consistent with the real options theory, which suggests that firms will invest cautiously when they perceive an increase in external uncertainty. Our conclusions were robust when we alleviated endogenous problems, adopted alternative proxies for innovation and FEPU indices, applied an alternative estimation method, and used a full sample. Furthermore, we discovered that the negative impact of FEPU on enterprise innovation output is more significant in non-state-owned enterprises, and in enterprises with higher financial constraints and lower government subsidies.

Our study has important policy implications. First, as Gulen and Ion [1] argue, frequent policy adjustments by policymakers can be more costly than a bad decision. Our study finds that firms reduce their innovation behavior when they perceive increased economic policy uncertainty. Frequent changes in economic policy can create a great deal of uncertainty in the business environment for companies, and have a significant negative impact on business innovation, which is detrimental to the role of firms in innovation-driven development. Therefore, on the one hand, the government should appropriately improve the continuity and stability of economic policies and reduce the frequency of policy adjustments. On the other hand, it should enhance the transparency of procedures when formulating and adjusting policies, as well as enhance the credibility of the government, so as to provide stable expectations among enterprises. Only by operating in a stable environment will companies be able to innovate with more confidence, thus realizing an “innovation-driven development strategy” in the real sense. Second, given the current context of rising global economic policy uncertainty, the government could consider creating an enabling environment for business innovation through complementary measures. Our study finds that the negative impact of FEPU on firms’ innovation output is more pronounced among non-state firms, firms with higher financial constraints, and firms with lower government subsidies. Therefore, in a period of rising external uncertainty, policymakers should guide the stable development of enterprises more directly. For example, more targeted government subsidies could be considered for non-state enterprises, as these firms are more affected by economic policy uncertainty than state-owned enterprises. At the same time, the government should improve the credit allocation system, and provide more policy support and credit resources to enterprises with greater financing constraints, so as to provide them with low-cost innovation funds.

There are some limitations to our study. First, we did not classify the economic policies, especially those directly related to innovation. Therefore, future research should further distinguish between the types of economic policy uncertainty perceived by enterprises. Second, this study focuses on the micro level of enterprises, and does not provide further analysis at the industry level. Enterprises in different industries may have different degrees of sensitivity to uncertainty—for example, the real estate industry is vulnerable to policy. Moreover, the innovation behavior of companies varies widely by industry; for example, the manufacturing industry is more innovative. Therefore, future research might collect data at the industry level.