Industries' heterogeneous reactions during the COVID‐19 outbreak: Evidence from Chinese stock markets

Abstract

This study examines the heterogeneous effects of the COVID‐19 outbreak on stock prices in China. We confirm what is already known, that the pandemic has had a significant negative impact on stock market returns. Additionally, we find, this effect is heterogeneous across industries. Second, fear sentiment can directly cause stock prices to fall and panic exacerbates the negative impact of the pandemic on stock returns. Third, and most importantly, we demonstrate the underlying mechanisms of four firm characteristics and find that those with high asset intensity, low labor intensity, high inventory‐to‐revenue ratio, and small market value are more negatively affected than others. For labor‐intensive state‐owned firms, in particular, stock performance worsened because of higher idle labor costs. Finally, we created an index to measure the relative position of an industry in the supply chain, which shows that downstream companies were more vulnerable to the effects of the pandemic.

1. INTRODUCTION

The impact of the COVID‐19 pandemic on the world economy exceeded our worst expectations (Baker et al., 2020). The shock was quickly reflected in the stock markets, which dropped sharply in the early stages of the pandemic (Gormsen & Koijen, 2020; Ramelli & Wagner, 2020). The stock markets were adversely affected not only in China, which was exposed to the pandemic the earliest but also in other countries. Existing studies capture the effects of the pandemic on the stock markets from various perspectives. In addition to the focus on stock prices (or returns), much of the literature is concerned about the impacts on stock volatility, asset portfolios, financial contagion, and so on. While it was predictable that the pandemic would hurt stock performance, the potential heterogeneity effects, driving factors, and underlying mechanisms deserve further investigation.

This study examines the effects of the COVID‐19 outbreak on stock returns in China and its hidden mechanisms from distinct perspectives. We not only consider the stock price reactions but also observe the different response patterns in various industries. Although stock prices dropped as the pandemic took root, some industries were hit harder than the rest; these include accommodation and catering, leasing and business services, transportation, warehousing, and postal services. In contrast, some industries were much less affected—stock prices of pharmaceutical manufacturing and chemical industries rose against the general trend.

Complementing the existing literature on the relationship between the COVID‐19 pandemic and stock market performance, we study the underlying mechanism of this relationship from different perspectives. We first consider the role of fear sentiment and find that the negative effect on stock returns is partly due to fear sentiment. We also find that public panic exacerbated the pandemic's negative impact.

We further study the impacts of the COVID‐19 outbreak using four firm‐level influencing factors: asset intensity, labor intensity, inventory‐to‐sales ratio, and firm size. As expected, the higher the asset intensity of a company, the greater the negative impact of the pandemic on the performance of its stock. Unexpectedly but interestingly, we find that the higher the labor intensity, the smaller the adverse effect. We argue that this is because of labor‐intensive enterprises' lower firing costs; during the pandemic, most of these enterprises paid very little or no salary to their workers who did not come to work. We regard state‐owned enterprises as companies with higher firing costs to verify our hypothesis, and the empirical evidence supports this. Regarding the role of the inventory‐to‐sales ratio, we find that the stock prices of companies with a high inventory ratio were more negatively affected by the pandemic. This is easy to understand: companies with high inventory occupancy rates need to pay more when encountering the inventory backlog caused by the pandemic. Regarding the firm size effect, we find that the stock prices of small companies dropped more than those of larger firms.

Finally, by creating a novel supplier–customer index, we analyze the effect of the COVID‐19 outbreak from the industry‐level supply chain perspective and find that companies located downstream were more negatively affected. We argue that this is because the impacts on demand and supply are asymmetric, including the direct impacts from the outbreak and indirect impacts from the government's differentiated policies.

This paper adds to the existing literature in several ways. First, we used the search volume of COVID‐19‐related words to construct a fear sentiment index for the pandemic. Because the impact of panic on the stock market may be ahead of the impact of the pandemic, this index can help us more accurately understand the potential impact mechanism.

Second, we add new insights to our knowledge of how capital‐ and labor‐intensive firms responded differently to the pandemic. We compare the diametrically opposite effects of the pandemic on capital‐ and labor‐intensive companies. We find that stock prices of capital‐intensive companies fell more sharply than those of labor‐intensive firms. We provide a novel explanation from the perspective of dismissal costs and find substantial evidence supporting our hypothesis by distinguishing between state‐owned and nonstate‐owned enterprises.

Third, we consider the impact of a possible inventory backlog caused by the decline in market demand during the pandemic. Although existing literature notes how the pandemic affects inventory costs (Kargar et al., 2020), it has not explored how stock prices react to this effect. We highlight the role of low inventory costs in mitigating the pandemic's negative impact on stock prices.

Fourth, we find that the firm‐size effect exists in the impact of the pandemic on stock prices. Specifically, the stocks of small companies performed relatively worse during the pandemic, perhaps due to the companies' insufficient ability to deal with the impact of the pandemic, including financing difficulties, imperfect management systems, and insufficient market competitiveness.

Finally, we try to explain the heterogeneous influences of the pandemic in different industries from the perspective of the supply chain. Based on the World Input–Output Database (WIOD), we create a new index called the supplier–customer index to measure the relative position of an industry in the supply chain. We find that downstream industries were more vulnerable to the COVID‐19 shock.

The remainder of this paper is organized as follows. Section 2 provides a brief literature review. Section 3 describes the data and variables, including descriptive statistics. Section 4 presents the main results of the impact of COVID‐19 on stock returns, and Section 5 provides further discussion from multi‐perspectives. Section 6 provides the robustness tests, and Section 7 concludes the paper.

2. LITERATURE REVIEW

With several severe public health crises in human history, such as the “Spanish flu” pandemic of 1918, research has focused on the impact of dangerous infectious diseases (such as SARS, H1N1, and Ebola) on the financial markets (Barro et al., 2020). M. P. Chen et al. (2018) examined the effect of the SARS epidemic on the long‐term integration of Asian stock markets and found that the epidemic had weakened the relationship between China's stock market and four other Asian stock markets. Gong et al. (2020) empirically analyzed the impact of the 2009 H1N1 pandemic on bank lending to find that the pandemic had increased the cost of bank loans and reduced the average loan size.

Studying the impact of Black‐Swan (or extreme) events on the stock market is important for financial risk prevention (Boubaker et al., 2015, 2016; L. Chen, Wen, et al., 2022; Wen, Zhang, et al., 2022). There is ample evidence to show that an epidemic generally hurts stock market returns. M. H. Chen et al. (2007) used an event‐study approach to examine the effect of the SARS epidemic on Taiwanese hotel stock performance and found a significantly harmful effect. Ichev and Marinč (2018) found the 2014–2016 Ebola outbreak events had resulted in negative returns in the stock market. However, the impact of the pandemic on the stock market is not always negative and can be heterogeneous. For example, Nippani* and Washer (2004) examined the impact of SARS on the affected countries' (regions') stock markets to find no negative impact, except in China and Vietnam. Loh (2006) examined the impact of SARS on airline stocks and found that SARS did not affect returns, but increased market volatility. Donadelli et al. (2017) found that dangerous infectious diseases, viewed by investors as profitable investment opportunities, had a positive effect on pharmaceutical stock returns. Chun‐Da chen et al. (2009) found that the effects of the SARS outbreak on different sectors were asymmetrical. Their empirical results confirmed the outbreak's negative impacts on tourism and the wholesale and retail sectors, while indicating a positive impact on the biotechnology sector.

Being a unique global public health crisis, the COVID‐19 pandemic's impact on financial markets has received unprecedented attention from researchers. Several studies have examined its effects from different perspectives, such as stock returns (Baker et al., 2020; L. Chen, Min, et al., 2022; Xu, 2021), financial volatility (risk) (Akhtaruzzaman et al., 2022; Wen et al., 2021; Zhao & Wen, 2022), crash risk (Dai et al., 2021; Z. Liu et al., 2021), financial contagion (Akhtaruzzaman et al., 2020; Zhao et al., 2021), asset allocation (Akhtaruzzaman, Boubaker, Lucey, et al., 2021; Ji et al., 2020; Wen, Tong, et al., 2022), corporate activities (Akhtaruzzaman et al., 2021; Boubaker et al., 2022), credit spread (Cicchiello et al., 2022), and oil prices (Akhtaruzzaman, Boubaker, Chiah, et al., 2021).

Some other studies have focused on the stock market reaction to the COVID‐19 pandemic, with empirical evidence. Ramelli and Wagner (2020) presented an analysis of how stock prices responded to the outbreak from the perspective of international trade and global value chains. Al‐Awadhi et al. (2020) conducted a panel regression to analyze the effect of the COVID‐19 outbreak on the Chinese stock market; the results show that both daily growth in total confirmed cases and total deaths significantly reduced stock returns. Ashraf (2020) and H. Liu et al. (2020) found that the negative effect held true for more markets. Gormsen and Koijen (2020) showed that although the response of the US and European Union stock markets to the outbreak in China was not evident, but stock prices sharply declined when Italy, South Korea, and Iran were affected.

However, heterogeneous impacts may exist in different sectors or with various firm characteristics. Yilmazkuday (2021) and Nguyen (2022) found different sectors performing differently during the COVID‐19 pandemic. H. Liu et al. (2020) analyzed the different industry index responses and found that the pharmaceutical manufacturing, software, and IT services sectors benefited from the pandemic, while the transportation, lodging, and catering sectors suffered a significant negative impact. As the tourism sector has been one of the worst affected by the pandemic due to lockdowns or travel restrictions, its stock prices have also been battered (Carter et al., 2022; Sikiru & Salisu, 2021; Wu et al., 2021). Harjoto et al. (2020) found that the negative shock of the COVID‐19 pandemic has been more significant for small firms. Alfaro et al. (2020) confirmed the universally negative changes in market value driven by the pandemic but found that market performance varied widely both within and across industries and that relatively more capital‐intensive, debt‐laden, and less profitable firms performed worse than others. Fahlenbrach et al. (2020) found that firms with high financial flexibility experienced a more significant stock price decline than those with low financial flexibility. Albuquerque et al. (2020) showed stocks with high environmental and social ratings significantly outperformed others.

It is precisely the uniqueness of the COVID‐19 pandemic that makes it necessary to uncover interesting problems that may still be unaddressed. The impact of COVID‐19‐induced fear sentiment is one of the biggest concerns here. Duan et al. (2020) investigated the impact of COVID‐19‐induced sentiment on the Chinese stock market and quantitatively measured it by using 6.3 million textual data from both official news media and Sina Weibo. C. Chen et al. (2020) examined the role of fear sentiment in the bitcoin market during the pandemic. They developed a novel proxy for fear sentiment by aggregating the hourly Google search volume of terms related to coronavirus. Our study is in line with these studies. We construct a COVID‐19‐induced fear sentiment model using data collected through the Baidu search engine in China, and examine the relationship between fear sentiment and stock performance. However, what we have further research based on these studies is the role of fear sentiment in the impact of COVID‐19 cases and deaths on stock returns, that is, the interaction effect between fear sentiment and daily cases (or deaths).

Notwithstanding the findings in the literature, it is still promising to study the effects of the pandemic on various firm characteristics to uncover more hidden mechanisms. We consider four firm characteristics: asset intensity, labor intensity, inventory‐to‐revenue ratio, and company size. Some of our findings confirm the conclusions of previous studies. For example, we find that stock returns for firms with high capital intensity were relatively more negatively affected, which is consistent with Alfaro et al. (2020). Second, our finding that the pandemic's negative effect was stronger for small firms confirms that of Harjoto et al. (2020). However, when we discuss these two types of firms' characteristics, we add more evidence and explanation. First, we compare the effects of capital intensity and labor intensity to find new mechanisms. Moreover, for the size effect, we directly estimate the differences in the effects of pandemic cases and deaths on companies of different sizes by using panel data regression methods, not just through the event study method. Both the abovementioned studies focused on the US stock market but lacked attention to emerging markets, while our study sheds light on similar problems in the Chinese stock market. More importantly, to the best of our knowledge, we are the first to consider firm characteristics, such as labor intensity and inventory ratio, in assessing the impact of the pandemic on stock market performance. In addition, we propose a novel industry‐level supply chain index to study the heterogeneous influence of the pandemic on upstream and downstream enterprises.

Overall, our study attempts to address and explain the heterogeneous impacts of the pandemic on stock market returns in view of the different firm and industry characteristics, and reveal the underlying mechanisms. As the current research from the perspective of company characteristics is still not comprehensive enough, and there is a scarcity of research from the perspective of industry characteristics, we believe our study adds new insights to existing knowledge.

3. DATA AND DESCRIPTIVE STATISTICS

We use the growth rate of daily confirmed cases (rCases) as the main proxy indicator to measure the severity of the pandemic, and the growth rate of daily deaths (rDths) as an alternative proxy to conduct a robustness analysis.

In the early days of the pandemic, it was expected that people would panic. Motivated by Da et al. (2011, 2015), Y. Chen et al. (2021), and X. Chen et al. (2022), we used the Baidu search index to construct a fear sentiment index for COVID‐19. The idea behind this index was that the higher the search volume (SVI) of COVID‐19‐related keywords, ¹ the more it reflected the panic among the people. We constructed a dummy variable fearsent_dummy for fear sentiment ² ; when the search volume is higher than the average, we define the value of this dummy variable as 1, which means that it is in a panic period; otherwise, the value is 0.

To further examine the underlying mechanism, we constructed the following variables to capture the various characteristics of firms and industries. The primary industry classification is based on the China Security Regulation Commission (CSRC) data, and the secondary industry is defined according to WIOD. We selected all A‐share companies in the Shanghai and Shenzhen stock markets as samples. We exclude firms with missing values and observations in the finance sector.

Return on Stock/industry: We define the return of a firm or industry as the natural log difference of closing prices, $R_{it}=\log (clos{e}_{i,t})-\log (clos{e}_{i,t-1})$, where represents the closing price of stock/industry i on day t.

Capital intensity (Cintensity): We define capital intensity as log(TAsset)/log(Orevenue), where TAsset represents the total assets at the end of the last fiscal year and Orevenue is the operating revenue. A large value means high capital intensity.

Labor intensity (Lintensity): Labor intensity is defined as log(Nstaff)/log(Orevenue), where Nstaff represents the number of employees at the end of the previous fiscal year, and Orevenue is the operating revenue. The larger the value, the more labor‐intensive the company.

Relative capital‐labor intensity (Rintensity): We construct this variable to measure the relative capital–labor intensity, defined as log(TAsset)/log(Nstaff). A large value indicates that a company is (relatively) more capital‐intensive or (relatively) less labor‐intensive.

State‐owned dummy (State‐owned): This dummy variable takes the value of 1 when a company is state‐owned and 0 otherwise.

Inventory‐to‐revenue (InvenRatio): This variable is defined as the ratio of inventory to operating revenue.

Firm size (lnSize and d_Size): We construct two variables for firm size, lnSize and d_Size, where lnSize is defined as the natural logarithm of a firm's market value. d_Size is a dummy variable, with a value of 1 when the market value is higher than the median of the market value, and 0 otherwise.

Supplier–customer index (SCIndex): We construct an industry‐level SCIndex to measure the relative position of an industry in the supply chain. The standard industry code is from the CSRC, and we match the industry codes with the WIOD database, according to Li et al. (2020). From the input–output (IO) tables, we can learn information about goods and service flows between sectors. We define the supplier–customer index for industry k as $(Ou{t}_k-I{n}_k)/(Ou{t}_k+I{n}_k)$, where $Ou{t}_k$ represents the amount of goods and services flowing from industry k to other industries, and $I{n}_k$ represents the amount of goods and services flowing from other industries to industry k. A large index value indicates that the industry is relatively upstream in the supply chain.

The search data we used to construct the COVID‐19 fear sentiment index are from the Baidu Index, and the Fama‐French three‐factor data were calculated by the China Asset Management Research Center, Central University of Finance and Economics. All other data were obtained from the CSMAR database. The sample period for our main empirical analysis was from January 1 to April 3, 2020. To calculate the abnormal returns (ARs), we use the sample for the entire year of 2019 to estimate the parameters. Table 1 reports the descriptive statistics for the main variables and Table 2 reports Pearson's correlations. ³

Table 1.

Descriptive statistics

Variables	Mean	Standard deviation	Fifth percentile	First quartile	Median	Third quartile	95th percentile	Observations
r	−0.0009	0.0343	−0.0579	−0.0178	0.0000	0.0162	0.0539	160,358
R Mt	−0.0006	0.0196	−0.0370	−0.0081	0.0020	0.0105	0.0243	160,358
SMB	0.0008	0.0067	−0.0262	−0.0026	0.0006	0.0043	0.0102	160,358
HML	−0.0014	0.0073	−0.0104	−0.0056	−0.0034	0.0021	0.0110	160,358
rCases	0.0429	0.8989	−1.0907	−0.2800	0.0000	0.1335	1.6094	160,358
rDths	0.0231	0.5168	−0.6931	−0.1978	0.0000	0.1144	1.3863	160,358
fearsent_dummy	0.3282	0.4696	0.0000	0.0000	0.0000	1.0000	1.0000	160,358
Cintensity	1.0337	0.0388	0.9813	1.0107	1.0284	1.0495	1.1026	148,305
Lintensity	0.3864	0.0669	0.3033	0.3475	0.3787	0.4120	0.5071	144,532
Rintensity	2.7474	0.4629	2.0436	2.4923	2.7204	2.9707	3.4600	144,532
InvenRatio	0.4037	1.1210	0.0103	0.0920	0.1719	0.3103	1.4080	156,622

Note: Descriptive statistics for the main variables used in our study, including the means, standard deviations, quantiles (5% quantile, the first quartile, median, the third quartile, 95% quantile), and the number of observations are reported.

Table 2.

Correlation matrix

	r	mkt_rf	SMB	HML	rCases	rDths	fearsent_dummy	Cintensity	Lintensity	Rintensity	InvenRatio
r	1.0000	0.5934***	0.2812***	−0.0551***	−0.1652***	−0.1381***	0.0151***	−0.0076***	0.0002	−0.0024	−0.0042*
R Mt	0.5934***	1.0000	0.2579***	−0.2125***	−0.2548***	−0.1801***	0.0651***	−0.0001	−0.0001	0.0002	0.0000
SMB	0.2812***	0.2579***	1.0000	0.11615***	−0.1081***	−0.1491***	−0.0055**	0.0000	0.0001	−0.0002	−0.0001
HML	−0.0551***	−0.2125***	0.11615***	1.0000	−0.1908***	−0.3500***	−0.2612***	0.0001	0.0000	−0.0001	−0.0000
rCases	−0.1652***	−0.2548***	−0.1081***	−0.1908***	1.0000	0.5900***	0.1203***	0.0009	0.0011	−0.0006	0.0004
rDths	−0.1381***	−0.1801***	−0.1491***	−0.3500***	0.5900***	1.0000	0.2374***	0.0008	0.0010	−0.0005	0.0004
fearsent_dummy	0.0151***	0.0651***	−0.0055**	−0.2612***	0.1203***	0.2374***	1.0000	0.0000	−0.0001	0.0005	0.0003
Cintensity	−0.0076***	−0.0001	0.0000	0.0001	0.0009	0.0008	0.0000	1.0000	0.0083***	0.2661***	0.3905***
Lintensity	0.0002	−0.0001	0.0001	0.0000	0.0011	0.001	−0.0001	0.0083***	1.0000	−0.9061***	−0.0367***
Rintensity	−0.0024	0.0002	−0.0002	−0.0001	−0.0006	−0.0005	0.0005	0.2661***	−0.9061***	1.0000	0.1599***
InvenRatio	−0.0042*	0.0000	−0.0001	−0.0000	0.0004	0.0004	0.0003	0.3905***	−0.0367***	0.1596***	1.0000

Note: Pearson's correlations between the main variables used in our study are reported. The numbers with ∗∗∗, ∗∗, and ∗ are significant at 1%, 5%, and 10% levels, respectively.

4. RESULTS

4.1. COVID‐19 and the cumulative ARs

We started with an event study of the COVID‐19 outbreak (Pandey & Kumari, 2021). To estimate ARs, the Fama‐French three‐factor model was used, ⁴ estimated as

$$R_{it}=\alpha +{\beta }_1(R_{mt}-R_{ft})+{\beta }_{2}SM{B}_t+{\beta }_{3}HM{L}_t+{\epsilon }_{it},$$ (1)

where $R_{it}$ is the return for individual firm i on day t, $R_{mt}-R_{ft}$ is the excess market return (market return minus risk‐free return), $SMB$ is the size factor, and $HML$ is the book‐to‐market factor. The error term, ${\epsilon }_{it}$, is the AR. We estimated the model using 2019 data to determine the parameters. Then we can calculate the abnormal returns (AR) and the cumulative abnormal returns (CARs) as:

$$A{R}_{it}=R_{it}-E\left(R_{it}\right)=R_{it}-(\hat{\alpha }+{\hat{\beta }}_1(R_{mt}-R_{ft})+{\hat{\beta }}_{2}SM{B}_t+{\hat{\beta }}_{3}HM{L}_t),$$ (2)

$$CA{R}_i(t_0,t_1)=\sum _{t=t_0}^{t_1}A{R}_{it.}$$ (3)

Figure 1 presents firms' average CARs (left axis) and the daily confirmed case line (right axis). The CARs show a declining trend in the early stage of the pandemic, starting from the confirmation of human‐to‐human transmission of COVID‐19 by Zhong Nanshan, especially after the Wuhan lockdown, and stock market returns show a steep downward trend. It was not until the government gradually began to ease the lockdown that stock market returns rebounded. Intuitively, the CAR trend reflects a trend opposite to that of daily cases, implying a direct negative correlation between the COVID‐19 outbreak and stock returns.

Figure 1.

Daily confirmed cases and cumulative abnormal returns. This figure presents firms' average cumulative abnormal returns (left axis), and the daily confirmed case line (right axis). We have added five key moments in the timeline of the COVID‐19 outbreak in China.

Figure 2 shows firms' average CARs and the daily death trend, and they show a similar reverse relationship. When the death toll increased, CARs experienced a sustained decline.

Figure 2.

Daily deaths and cumulative abnormal returns. This figure presents firms' average cumulative abnormal returns (left axis), and the daily death line (right axis). We added five key moments to the timeline of the COVID‐19 outbreak in China.

Overall, from the results of the event study, we can see that both the number of daily confirmed cases and the number of daily deaths caused a short‐term decline in stock market returns. However, stock prices rebounded as the pandemic gradually came under control.

4.2. The panel regression approach

We conduct a panel regression analysis to further test the relationship between COVID‐19 and stock returns. The dependent variable is the daily return on individual stocks and the independent variable is the growth rate of the daily confirmed cases. In the estimation model, the Fama‐French factors are added as control variables, and both industry and province fixed effects are included. Specifically, we employ the following panel model with fixed effects.

$$R_{it}=\alpha \prime +{\beta }_1^{\prime }(R_{mt}-R_{ft})+{\beta }_2^{\prime }SM{B}_t+{\beta }_3^{\prime }HM{L}_t+\lambda rCase{s}_{t-1}+e_{pro}+u_{ind}+{\epsilon \prime }_{it},$$ (4)

where $R_{it}$ is the return on stock i on day t, and $R_{mt}-R_{ft}$, $SMB$, and $HML$ are the excess market return, size factor, and book‐to‐market factor, respectively. $rCase{s}_{t-1}$ is the growth rate of daily confirmed cases on day t–1. $e_{pro}$ and $u_{ind}$ represent province‐ and industry‐fixed effects, respectively.

The estimation results are shown in Table 3. The coefficient of $rCase{s}_{t-1}$ is significantly negative, indicating that the outbreak of COVID‐19 had a significantly negative effect on stock market returns. In addition, the coefficients of the other lagged terms, including $rCase{s}_{t-2}$ and $rCase{s}_{t-3}$, are negative and significant at the 1% level, showing the persistent influence of daily cases on stock prices and their predictive power for changes in stock returns. This finding is consistent with our expectations. As the number of confirmed cases increased, and measures such as lockdowns, quarantines, and travel restrictions were taken in response to curb the spread of the disease, it brought about a negative shock to economic activities, with companies shutdown. This led to a decline in firm values, prompting savvy investor action based on expectations, finally causing a fall in stock prices.

Table 3.

Estimation results from daily confirmed cases

	(1)	(2)	(3)	(4)	(5)	(6)
Intercept	−0.002636** (−2.19)	0.002519* (−2.09)	−0.002417* (−2.01)	−0.002424** (−2.52)	−0.002405** (−2.50)	−0.002402** (−2.49)
R Mt				0.997431*** (268.31)	0.997699*** (268.31)	0.997296*** (267.19)
SMB				0.646491*** (58.67)	0.638710*** (56.33)	0.638410*** (56.29)
HML				0.237848*** (24.67)	0.236733*** (24.54)	0.234261*** (23.78)
rCases (t−1)	−0.005038*** (−52.34)	−0.005031*** (−52.34)	−0.004955*** (−51.56)	−0.000408*** (−5.08)	−0.000423*** (−5.26)	−0.000422*** (−5.25)
rCases (t−2)		−0.001976*** (−20.55)	−0.001971*** (−20.52)		−0.000231*** (−2.91)	−0.000233*** (−2.93)
rCases (t−3)			−0.001798*** (−18.70)			−0.000098*** (−1.24)
Industry‐fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Observations	160,297	160,297	160,297	160,297	160,297	160,297
R 2	0.017262	0.019846	0.021980	0.372731	0.372764	0.372770

Note: The panel regression results to test how COVID‐19 daily confirmed cases affected stock returns are reported. The dependent variable is the daily return on individual stocks and the independent variable is the growth rate of the daily confirmed cases (three lag terms are considered). In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

Table 4 presents the results of the daily deaths, which are consistent with those from daily confirmed cases: an increase in the number of deaths leads to a fall in stock prices, indicating the negative impact of the pandemic. Nevertheless, the results show some slight differences: the coefficient of $rDt{h}_{t-3}$ is not significant. This result shows that the impact of the death toll on stock prices is less persistent.

Table 4.

Estimation results from daily deaths

	(1)	(2)	(3)	(4)	(5)	(6)
Intercept	−0.002640** (−2.20)	−0.002741** (−2.29)	−0.002734** (−2.28)	−0.002419** (−2.51)	−0.002393** (−2.48)	−0.002391** (−2.48)
R Mt				0.996688*** (267.68)	0.999444*** (267.27)	0.999459*** (267.26)
SMB				0.638889*** (56.89)	0.635854*** (56.60)	0.635403*** (56.32)
HML				0.237731*** (24.67)	0.226782*** (23.30)	0.225697*** (22.45)
rDths (t−1)	−0.010683*** (−64.61)	−0.010367*** (−62.38)	−0.010351*** (−62.05)	−0.000873*** (−6.15)	−0.000994*** (−6.97)	−0.000992*** (−6.95)
rDths (t−2)		0.002878*** (17.24)	0.002858*** (17.03)		−0.001068*** (−7.81)	−0.001078*** (−7.79)
rDths (t−3)			−0.000196 (−1.17)			−0.000061 (−0.43)
Industry‐fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Observations	160,297	160,297	160,297	160,297	160,297	160,297
R 2	0.025841	0.027645	0.027653	0.372778	0.373017	0.373017

Note: The panel regression results for testing how COVID‐19 daily deaths affected stock returns are reported. The dependent variable is the daily return on individual stocks and the independent variable is the growth rate of daily deaths (three lag terms are considered). In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

These findings further strengthen the conclusions drawn by recent studies such as Al‐Awadhi et al. (2020), Baker et al. (2020), and Ramelli and Wagner (2020).

4.3. The role of fear sentiment

We constructed a fear sentiment index using the Baidu search index, which includes a series of keywords about COVID‐19. The red line in Figure 3 shows the trend in fear sentiments. Zhong Nanshan, an academician of the Chinese Academy of Engineering who gained a high reputation during the fight against SARS, confirmed on TV on January 20 that there was evidence of human‐to‐human transmission of COVID‐19 in Wuhan. Subsequently, the public's fear sentiment began to increase sharply. On January 23, the central government announced that it would seal Wuhan, not allowing anyone in or out, which further exacerbated public panic. This situation continued until major leadership changes in Hubei and the highest number of confirmed cases were reported. ⁵

Figure 3.

Fear sentiment and cumulative abnormal returns. This figure presents firms' average cumulative abnormal returns (left axis) and the fear sentiment (right axis). We used the search volume (SVI) of COVID‐19‐related keywords from the Baidu search index to construct a fear sentiment index for COVID‐19.

Thus, we focus on how fear sentiment affects stock market returns. We add the dummy variable fearsent_dummy to the previous panel model (Model 4) and construct an interaction term between fearsent_dummy and rCases. The results are presented in Table 5.

Table 5.

Fear sentiment and stock returns

	(1)	(2)	(3)
Intercept	−0.002261** (−2.34)	−0.002278** (−2.36)	−0.002249** (−2.33)
rCases (t−1)		−0.000354*** (−4.32)	0.000018 (0.09)
fearsent_dummy	−0.000626*** (−4.18)	−0.000492*** (−3.22)	−0.000501*** (−3.27)
rCases (t−1) × fearsent_dummy			−0.000463** (−2.11)
Fama‐French three factors	Yes	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes
Observations	160,297	160,297	160,297
R 2	0.372698	0.372771	0.372789

Note: The panel regression results for further testing whether and how fear sentiments during the COVID‐19 pandemic affected stock returns are reported. The dependent variable is the daily return on individual stocks. We add the growth rate of daily confirmed cases (a one‐period lag term is considered), a dummy variable of fear sentiment, and their interaction term into the model. In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

First, the coefficient of fear sentiment is negative and statistically significant. This is in line with our expectations that a panic sentiment will directly cause a significant decline in stock market returns in the short term. In addition, the coefficient of the interaction term is significantly negative, indicating that fear sentiment exacerbated the negative impact of COVID‐19 on stock market returns.

Our findings are consistent with Duan et al. (2020); they used 6.3 million textual messages from social media to measure the COVID‐19 sentiment index in China and found that it could positively predict stock returns. Sun et al. (2021) also reached a similar conclusion in the Chinese stock market. The novelty of our study is that it not only examines the impact of COVID‐19‐induced sentiment on stock returns but also shows how fear sentiment reinforces this negative impact. We have revealed a new mechanism by which the pandemic caused stock prices to fall; when people panic about a worsening situation, a confirmation (increase in positive cases) will more severely affect stock prices.

4.4. Effects in different industries

We redo our panel data analysis by industry (primary industry according to CSRC and secondary industry according to WIOD ⁶ ), and the results are shown in Tables 6 and 7, respectively.

Table 6.

Results by CSRC primary industry

Industry name	Intercept	rCases(t−1)	Observations	R 2
Agriculture, forestry, animal husbandry, and fishery	0.003273 (1.26)	−0.001316 (−1.46)	1890	0.291991
Mining industry	−0.001215 (−0.87)	0.000263 (0.70)	4199	0.451836
Manufacturing	−0.001212 (−1.53)	−0.000094 (−0.90)	102,045	0.359208
Manufacturing (exclude for Pharmaceutical manufacturing)	−0.001017 (−1.28)	−0.000390*** (−3.65)	92,295	0.393693
Manufacturing (exclude for Pharmaceutical manufacturing and chemical industry)	−0.001015 (−1.28)	−0.000425*** (−3.95)	91,014	0.395850
Electricity, heat, gas, and water production and supply	−0.000912 (−1.02)	−0.000877*** (−3.33)	6271	0.532459
Construction industry	0.000582 (0.28)	−0.001661*** (−4.37)	5178	0.487365
Wholesale and retail industry	−0.000612 (−0.36)	−0.000329 (−0.98)	9318	0.312867
Transportation, warehousing, and postal services	−0.003022 (−1.64)	−0.001989*** (−6.30)	5849	0.468575
Accommodation and catering	0.001178 (0.33)	−0.004954*** (−3.15)	488	0.390676
Information transmission, software, and information technology services	−0.006023 (−1.62)	−0.000561 (−1.37)	7381	0.473819
Real estate	0.001959 (0.66)	−0.001578*** (−4.63)	7007	0.442148
Leasing and business services	−0.003819 (−1.28)	−0.002323*** (3.86)	2316	0.503670
Scientific research and technical services	−0.000473 (−0.15)	−0.000403 (−0.60)	1889	0.440717
Water conservancy, environment, and public facilities management	−0.002071 (−0.99)	−0.001103* (−1.75)	2074	0.471501
Education	0.003718 (0.88)	−0.000154 (−0.06)	244	0.199886
Health and social work	−0.002177 (−0.57)	−0.002711 (−1.53)	427	0.289530
Culture, sports, and entertainment	−0.003955 (−1.21)	−0.001117* (−1.83)	2562	0.486380
Comprehensive industry	−0.005837* (−1.66)	0.0002173 (0.24)	1159	0.354856

Note: The panel regression results for testing how COVID‐19 daily confirmed cases affected stock returns in primary industries are reported. The primary industry classification is based on the CSRC. The dependent variable is the daily return on individual stocks and the independent variable is the growth rate of the daily confirmed cases (the one‐period lag term is considered). In the estimation model, Fama‐French three factors were added as control variables, and province fixed effects were included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

Abbreviation: CSRC, China Security Regulation Commission.

Table 7.

Results by WIOD secondary industry

Industry	rCases (t−1)	N	R 2	Industry	rCases (t−1)	Observations	R 2
A01	−0.000900 (−0.78)	1280	0.274043	D35	−0.001181*** (−4.28)	5356	0.546256
A02	−0.002333 (−1.00)	183	0.357746	E36	0.000973 (1.22)	915	0.471823
A03	−0.002506 (−1.45)	366	0.368397	E37‐E39	0.000660 (0.28)	244	0.337193
B	0.000263 (0.70)	4199	0.451836	F	−0.001661*** (−4.37)	5178	0.487365
C10–C12	−0.001055*** (−2.93)	7130	0.367861	G46	−0.000009 (−0.02)	4317	0.305738
C13–C15	−0.000620 (−1.17)	4931	0.247343	G47	−0.000607 (1.35)	5001	0.324754
C16	−0.002730** (−2.07)	488	0.467585	H49	−0.001780*** (−4.30)	2374	0.542101
C17	−0.000734 (−1.18)	1708	0.459798	H50	−0.003068*** (−5.30)	1708	0.454780
C18	0.000825 (0.75)	732	0.466829	H51	−0.002524*** (−2.76)	732	0.508964
C19	0.000633 (0.70)	915	0.386291	H52	−0.000032 (−0.03)	730	0.448630
C20	0.000586** (2.14)	11994	0.395893	H53	−0.000451 (0.27)	305	0.368912
C21	0.002718*** (7.51)	9750	0.138381	I	−0.004954*** (−3.15)	488	0.390676
C22	0.000374 (0.60)	2988	0.327696	J61	−0.002552** (−2.35)	793	0.515353
C23	−0.000354 (−0.71)	3721	0.419535	J62‐J63	−0.000340 (−0.78)	6588	0.474882
C24	−0.000468 (−1.24)	5419	0.450693	L68	−0.001578*** (−4.63)	7007	0.442148
C25	−0.000500 (−1.02)	3294	0.446984	M69‐M70	−0.002323*** (−3.86)	2316	0.503670
C26	−0.000948*** (−3.00)	13146	0.474084	M72	0.001351 (0.64)	244	0.349638
C27	−0.000889*** (−2.83)	11034	0.411018	M74‐M75	−0.000707 (−1.01)	1645	0.471004
C28	−0.000424 (−1.50)	12917	0.381440	O84	−0.001103* (−1.75)	2074	0.471501
C29	0.000410 (1.03)	6647	0.442086	P85	−0.000154 (−0.06)	244	0.199886
C30	−0.001401** (−2.22)	2242	0.472727	Q	−0.003247*** (−2.75)	793	0.383242
C31–C32	−0.000365 (−0.58)	2745	0.394761	R‐S	0.000217 (0.24)	1159	0.354856

Note: The panel regression results for testing how COVID‐19 daily confirmed cases affected stock returns by secondary industry are reported. The secondary industry classification is based on WIOD, and only the industry code is given here (see the appendix for a detailed explanation of the abbreviations). The dependent variable is the daily return on individual stocks and the independent variable is the growth rate of the daily confirmed cases (the one‐period lag term is considered). In the estimation model, Fama‐French three factors were added as control variables, and province fixed effects were included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

Table 6 shows that the effects of the pandemic on stock market returns in different primary industries are heterogeneous. While most industries are still under the negative impact of the pandemic, 2 years on, in some industries, this impact was not very significant. The three industries most affected by the pandemic were accommodation and catering, leasing and business services, and transportation, warehousing, and postal services. However, agriculture, mining, and manufacturing industries, for example, did not seem significantly affected.

The other results from the secondary industries in Table 7 show more detail. Even in the pharmaceutical and chemical industries, the effect of COVID‐19 on stock market returns is significantly positive. When we exclude the pharmaceutical manufacturing industry and reanalyze the impact of the pandemic on the manufacturing industry, we find that the coefficient becomes significantly negative (−0.0004 with t‐statistic = −3.65; Table 6). The positive impact in the pharmaceutical industry is obvious; the pandemic increased people's demand for pharmaceutical products and their hopes for future discoveries. Why, then, was there a significant positive impact on the chemical industry? This may be because the chemical industry is the upstream industry for pharmaceuticals. For example, because of the very high demand for masks, melt‐blown nonwoven fabric, the original and core material used in masks, is in short supply, and so is its primary raw material, nonwoven polypropylene.

5. FURTHER STUDIES BASED ON FIRM AND INDUSTRY CHARACTERISTICS

To further reveal more hidden mechanisms, we add some firm‐ or industry‐specific variables to Model (4), including capital and labor intensity, firm size, inventory‐to‐revenue ratio, and supplier–customer index. Specifically, we use the following model as the basic empirical model in this section:

$$R_{it}=\alpha \prime +{\beta }_1^{\prime }(R_{mt}-R_{ft})+{\beta }_2^{\prime }SM{B}_t+{\beta }_3^{\prime }HM{L}_t+\lambda rCase{s}_{t-1}+\theta X_i+\rho \cdot rCase{s}_{t-1}\cdot X_i+e_{pro}+u_{ind}+{\epsilon \prime }_{it},$$ (5)

where $X_i$ represents a firm‐ or industry‐level variable for stock i.

5.1. Heterogeneous effects in asset/labor‐intensive companies

In this section, we study how companies with different capital/labor intensities are affected by the pandemic. From the results in Table 8, the coefficient of the interaction between the growth rate of daily confirmed cases and asset intensity was significantly negative. This shows that the pandemic had a greater negative impact on the stock performance of firms with higher capital intensity. This is because during the pandemic, it was difficult for firms with high capital intensity to reduce the cost of capital occupation in a short period.

Table 8.

The role of capital intensity

	(1)	(2)	(3)	(4)
Intercept	0.0011 (0.35)	0.0004 (0.12)	0.0013 (0.53)	0.0008 (0.30)
rCases (t−1)	−0.0050*** (−49.59)	0.0080*** (−2.96)	−0.0004*** (−4.29)	0.0099** (4.63)
Cintensity	−0.0034 (−1.31)	−0.0027 (−1.03)	−0.0034 (−1.63)	−0.0028 (−1.36)
rCases (t−1) × Cintensity		−0.0125*** (−4.81)		−0.0099*** (−4.80)
Fama‐French three factors	No	No	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes
Observations	148244	148244	148244	148244
R 2	0.0168	0.0169	0.3758	0.3759
F‐statistic	52.71	52.11	1749.46	1716.51

Note: The panel regression results to test whether and how capital intensity affected the relationship between the COVID‐19 pandemic and stock returns are reported. The dependent variable is the daily return on individual stocks. We add the growth rate of daily confirmed cases (a one‐period lagged term is considered), capital intensity, and their interaction term into the model. We define capital intensity as log(TAsset)/log(Orevenue), where TAsset represents the total assets at the end of the last fiscal year and Orevenue is the operating revenue. In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

By contrast, as can be seen from the results in Table 9, the coefficient of the interaction term of labor intensity is significantly positive, which means that firms with higher labor intensity are less affected. The results of the relative capital–labor intensity in Table 10 further confirm the heterogeneous effects in capital‐ or labor‐intensive firms. This seems contra intuitive: a higher labor intensity will have a greater negative impact on such firms because of the lockdown, as they rely more on the workers to meet production. Notably, labor‐intensive companies, most of which belong to COVID‐19‐exposed sectors, employ the lowest‐paid workers. As recent studies have mentioned (Bell & Blanchflower, 2020), such employees are easily affected by low wages or income losses, and the lowest‐income groups were the hardest hit in the pandemic. We argue that these workers are often temporarily hired, and their dismissal costs are meager. As a result, during the economic shutdown, the firms only needed to pay the worker a lower salary or not pay it at all. Therefore, although these firms faced a decline in operating income, labor costs also decreased, warding off further decline in stock prices.

Table 9.

The role of labor intensity

	(1)	(2)	(3)	(4)
Intercept	−0.002960** (−2.15)	−0.002817** (−2.04)	−0.002734** (−2.49)	−0.002570** (−2.34)
rCases (t−1)	−0.004980*** (−49.18)	−0.007485*** (−12.36)	−0.000372*** (−4.42)	−0.003264*** (−6.76)
Lintensity	0.000633 (0.46)	0.000262 (0.19)	0.000649 (0.59)	0.000221 (0.20)
rCases (t−1) × Lintensity		0.006504*** (4.20)		0.007508*** (6.08)
Fama‐French three factors	No	No	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes
Observations	144,471	144,471	144,471	144,471
R 2	0.016919	0.017039	0.376159	0.376319

Note: The panel regression results for testing whether and how labor intensity affected the relationship between the COVID‐19 pandemic and stock returns are reported. The dependent variable is the daily return on individual stocks. We add the growth rate of daily confirmed cases (a one‐period lag term is considered), labor intensity, and their interaction term into the model. Labor intensity is defined as log(Nstaff)/log(Orevenue), where Nstaff represents the number of employees at the end of the previous fiscal year, and Orevenue is the operating revenue. In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. Numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and t‐statistics are in parentheses.

Table 10.

The role of relative capital‐labor intensity

	(1)	(2)	(3)	(4)
Intercept	−0.002265* (−1.68)	−0.002455* (−1.82)	−0.002007* (−1.87)	−0.002208** (−2.06)
rCases (t−1)	−0.004980*** (−49.18)	−0.001662*** (−2.69)	−0.000372*** (−4.42)	0.003151*** (6.39)
Rintensity	−0.000147 (−0.74)	−0.000078 (−0.39)	−0.000156 (−0.99)	−0.000082 (−0.52)
rCases (t−1) × Rintensity		−0.001206*** (−5.44)		−0.001280*** (−7.25)
Fama‐French three factors	No	No	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes
Observations	144,471	144,471	144,471	144,471
R 2	0.016922	0.017123	0.376162	0.376389

Note: The panel regression results for testing whether and how relative capital‐labor intensity affected the relationship between the COVID‐19 pandemic and stock returns are reported. The dependent variable is the daily return on individual stocks. We add the growth rate of daily confirmed cases (one‐period lag term is considered), relative capital‐labor intensity, and their interaction term into the model. We define relative capital‐labor intensity as log(TAsset)/log(Nstaff), where TAsset represents the total assets at the end of the last fiscal year, and Nstaff represents the number of employees at the end of the last fiscal year. In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

If our hypothesis from the perspective of layoff costs is correct, when labor intensity remains unchanged, firms with higher layoff costs would be more affected by the pandemic. While the pandemic caused workers to suspend work, a firm had to pay some wages to them even if they did not work, as long as they could not be arbitrarily dismissed. In other words, the firm had to pay higher idling costs to its employees. Consequently, for firms with high dismissal costs, high labor intensity also means high idle labor costs. We did not find a variable that could accurately measure the dismissal costs of an individual firm, so we use the dummy variable of state‐owned enterprises as the proxy variable (State_owned). In China, for institutional and historical reasons, state‐owned enterprises are usually less flexible in dismissing employees. Therefore, they have higher dismissal costs than non‐state‐owned enterprises do. We add the interaction terms of rCases, Lintensity and State_owned to the benchmark model to verify our hypothesis. We expected the coefficient to be negative.

The results are presented in Table 11. The coefficient of the interaction term between rCases and Lintensity is still significantly positive. In addition, the coefficient of the interaction term between rCases and State_owned is significantly positive, indicating that state‐owned enterprises are relatively less affected by the pandemic. This may be because such enterprises can respond to the impact of the pandemic more effectively through various measures, such as obtaining loans to ease the financial pressure.

Table 11.

Test results for the dismissal cost hypothesis

	(1)	(2)	(3)	(4)
Intercept	−0.002488** (−2.26)	−0.002487** (−2.26)	−0.002990** (−2.53)	−0.002929 (−2.47)
rCases (t−1)	−0.00326*** (−6.76)	−0.003271*** (−6.77)	−0.003271*** (−6.77)	−0.004328*** (−6.85)
Lintensity	0.000502 (0.45)	0.000496 (0.45)	0.001612 (1.09)	0.001452 (0.98)
State_owned	−0.000320* (−1.95)	−0.000314* (−1.91)	0.000681 (0.77)	0.000535 (0.60)
rCases (t−1) × Lintensity	0.007508*** (6.08)	0.007617*** (6.09)	0.007616*** (6.09)	0.010422*** (6.31)
rCases (t−1) × State_owned		−0.000092 (−0.54)	−0.000092 (−0.54)	0.002462** (2.47)
Lintensity × State_owned			−0.002546 (−1.14)	−0.002173 (−0.97)
rCases (t−1) × Lintensity × State_owned				−0.006558*** (−2.60)
Fama‐French three factors	Yes	Yes	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes
Observations	144,471	144,471	144,471	144,471
R 2	0.376336	0.376337	0.376342	0.376372

Note: The panel regression results of testing the dismissal‐cost hypothesis are reported. The dependent variable is the daily return on individual stocks. We add the growth rate of daily confirmed cases (one‐period lag term is considered), labor intensity, a dummy of state‐owned, and all their interaction terms into the model. Labor intensity is defined as log(Nstaff)/log(Orevenue), where Nstaff represents the number of employees at the end of the previous fiscal year, and Orevenue is the operating revenue. The dummy variable State_owned equals one when the company is state‐owned. In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

However, we are most concerned about the coefficient estimation result of the interaction term of rCases, Lintensity, and State_owned. The coefficient is negative and significant (−0.0066 with t‐statistic = −2.60), and consistent with our expectations. This means that when the cost of dismissal is high (Stated‐owned = 1), firms with higher labor intensity are more negatively affected by the pandemic than those with low dismissal costs (Stated‐owned = 0). Alternatively, when the labor intensity is constant (>0), firms with higher dismissal costs will experience a more significant decline in stock prices in the pandemic.

Our findings confirm the conclusions in Alfaro et al. (2020), which found that COVID‐19‐induced losses in market value at the company level increase with capital intensity. However, we have added additional information to this knowledge. We also discuss these effects when considering labor intensity. Interestingly, contrary to the impact of capital intensity, the pandemic had a greater impact on companies with low labor intensity. When we jointly consider capital and labor intensity, their reverse effects still hold. We provide a novel and reasonable explanation for the impact of labor intensity from the perspective of dismissal costs.

5.2. What is the impact of inventory?

Another problem that a shutdown brings is the inventory backlog, which is a key reason for the increase in the short‐term costs of a firm. Specifically, the COVID‐19 outbreak caused poor circulation of goods in the short term, resulting in a backlog of commodities and increased inventory costs. Fransoo and Udenio (2020) pointed out that due to the Covid‐19 lockdown, inventory build‐up was to last for a long time until demand starts to recover. He et al. (2020) constructed an industry‐level inventory index using a synthetic index compilation method. This shows an upward trend under the impact of the pandemic, reflecting an increasing inventory backlog. However, the extent of the increase in inventory costs may differ. Generally, a firm with high inventory may face severe backlogs, leading to a further increase in corresponding inventory costs, and in turn affecting the firms. Therefore, we propose the hypothesis that if an enterprise's ratio of inventory to operating revenue was relatively high when the COVID‐19 pandemic arrived, the negative impact of the inventory backlog on its stock performance would be worse.

We tested this hypothesis, and the results are presented in Table 12. As expected, the coefficient of the interaction term is significantly negative, meaning that firms with a higher inventory‐to‐revenue ratio show a greater negative impact on their stock returns.

Table 12.

The role of inventory

	(1)	(2)	(3)	(4)
Intercept	−0.002543** (−2.08)	−0.002547** (−2.09)	−0.002323** (−2.39)	−0.002328** (−2.39)
rCases (t−1)	−0.005002*** (−51.42)	−0.004921*** (−47.53)	−0.000386*** (−4.76)	−0.000307*** (−3.58)
InvenRatio	−0.000062 (−0.65)	−0.000050 (−0.53)	−0.000064 (−0.85)	−0.000053 (−0.70)
rCases (t−1) × InvenRatio		−0.000204** (−2.29)		−0.000198** (−2.77)
Fama‐French three factors	No	No	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes
Observations	156,561	156,561	156,561	156,561
R 2	0.017078	0.017111	0.374525	0.374556

Note: The panel regression results to test whether and how inventory affected the relationship between the COVID‐19 pandemic and stock returns are reported. The dependent variable is the daily return on individual stocks. We add the growth rate of daily confirmed cases (a one‐period lag term is considered), inventory‐to‐revenue ratio, and their interaction term into the model. InvenRatio is defined as the ratio of inventory to operating revenue. In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

Our findings are an extension of those of Fransoo and Udenio (2020) and He et al. (2020) that the pandemic increase inventory costs. Our extended inference is that increased inventory costs reduce a firm's value, thereby causing the stock price to fall. Our empirical evidence supports this inference.

5.3. Does size matter?

The impacts of the COVID‐19 pandemic on small businesses have received extensive attention (Bartik et al., 2020). Generally, small and medium‐sized enterprises (SMEs) face relatively more developmental difficulties, such as financing challenges, imperfect management systems, and insufficient market competitiveness, which could lead to weaker SMEs. As the existing literature shows, a reasonable expectation is that large companies' stock prices will be relatively less affected by the pandemic, and small companies' stock prices could fall under the shock.

The results in Table 13 provide evidence supporting our expectations. The coefficient of rCase is significantly negative, indicating that the stock prices of small companies were significantly negatively affected by the pandemic. By contrast, the coefficient of the interaction term is significantly positive, which shows that the stock prices of large companies are comparatively less affected. In other words, when small companies are hit by the pandemic, their stock performance worsens.

Table 13.

The role of market value

	(1) Size = lnSize	(2) Size = lnSize	(3) Size = lnSize	(4) Size = lnSize	(5) Size = d_Size
Intercept	−0.012434*** (−7.28)	−0.011376*** (−6.66)	−0.009964*** (−7.31)	−0.008968*** (−6.57)	−0.002867*** (−2.97)
rCases (t−1)	−0.005030*** (−52.26)	−0.023170*** (−17.15)	−0.000404*** (−5.04)	−0.017509*** (−16.21)	−0.001415*** (−12.84)
Size	0.000642*** (8.11)	0.001175*** (13.46)	0.000494*** (7.81)	0.000431*** (6.80)	0.000919*** (6.60)
rCases (t−1) × Size		0.001175*** (13.46)		0.001108*** (15.88)	0.002054*** (13.36)
Fama‐French three factors	No	No	Yes	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes	Yes
Observations	160,297	160,297	160,297	16,0297	160,297
R 2	0.017665	0.018774	0.372969	0.373954	0.373645

Note: The panel regression results to test whether and how firm size affected the relationship between the pandemic and stock returns are reported. The dependent variable is the daily return on individual stocks. We add the growth rate of daily confirmed cases (the one‐period lag term is considered), firm size, and their interaction terms into the model. We construct two variables for firm size, lnSize and d_Size, where lnSize is defined as the natural logarithm of the company's market value and d_Size is a dummy variable. In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

The size effect was also documented by Harjoto et al. (2020), who used the event study method, treating the WHO announcement on March 11, 2020, as the event. We confirm this conclusion using panel data regression methods, to provide a more accurate estimate of the size effect. More importantly, while Harjoto et al. (2020) focused their attention on the US market, our main concern is the size effect in the Chinese stock market, which is a valuable supplement to our understanding of the impact of COVID‐19 on emerging markets.

5.4. An industry‐level supply chain perspective

COVID‐19 causes supply chain disruption, posing a huge challenge to supply chain management (Guan et al., 2020; Ivanov, 2020; Nikolopoulos et al., 2021). However, only a few studies have addressed the impact of the COVID‐19‐induced supply chain crisis on stock returns. In this section, we f study whether the impact of the pandemic differs for the firms in various positions in the supply chain. Based on the WIOD IO table, we construct an industry‐level supplier–customer index to describe the relative position of an industry in the industry‐level supply chain. A positive value indicates that supplies from this industry to other industries are higher than vice versa. The larger the value, the more upstream the industry is in the supply chain.

The empirical results are presented in Table 14. The coefficient of the interaction term is significantly positive, meaning that firms located downstream in the supply chain are more negatively affected.

Table 14.

The effects from an industry‐level supply chain perspective

	(1)	(2)	(3)	(4)	(5)
Intercept	−0.002410** (−2.50)	−0.002414** (−2.50)	−0.002273** (−2.35)	−0.002279** (−2.36)	−0.002251** (−2.33)
rCases (t−1)	−0.000407*** (−5.04)	−0.000322*** (−3.94)	−0.000357*** (−4.32)	−0.000272*** (−3.27)	0.000089 (0.45)
SCIndex	−0.000079 (−0.19)	−0.000178 (−0.42)	−0.000079 (−0.19)	−0.000178 (−0.42)	−0.000179 (−0.42)
rCases (t−1) × SCIndex		0.001752*** (6.99)		0.001750*** (6.99)	0.001751*** (6.99)
fearsent_dummy			−0.000460*** (−2.99)	−0.000457*** (−2.97)	−0.000466*** (−3.02)
SCIndex × fearsent_dummy					−0.000449* (−2.04)
Fama‐French three factors	Yes	Yes	Yes	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes	Yes
Observations	1,508,040	1,508,040	1,508,040	1,508,040	1,508,040
R 2	0.371424	0.371618	0.371459	0.371653	0.371670

Note: The panel regression results for examining the different impacts of the COVID‐19 pandemic on upstream and downstream companies in the supply chain are reported. The dependent variable is the daily return on individual stocks. We add the growth rate of daily confirmed cases (a one‐period lag term is considered), supplier–customer index, and their interaction term into the model. We define the supplier–customer index using data from the WIOD input–output tables. A large index value indicates the industry is relatively upstream in the supply chain. In the estimation model, Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

We argue that this is partly because, on the one hand, the impacts of the pandemic on demand and supply are asymmetric; the impact on demand is far greater than the disturbance to supply. On the other hand, China's total demand nearly did not recover until the gradual unblocking, but the government had begun to help companies resume work and production. Moreover, the government's stimulation efforts on the supply side were greater than those on the demand side. Consequently, in the early stages of the pandemic, the substantial negative impact on the demand side meant downstream companies in the supply chain were affected first. As the government gained control of the pandemic, policy support on the supply side enabled companies upstream of the supply chain to recover relatively quickly. The combination of these factors caused upstream companies to be relatively less affected.

6. ROBUSTNESS TESTS

In this section, we first perform robustness tests using the growth rate of daily deaths (rDths) as an alternative proxy for measuring the severity of the pandemic. The corresponding results are shown in Table 15. As we can see, this is consistent with the results from the daily confirmed cases.

Table 15.

Robustness results for daily deaths

	Vb = fearsent_dummy	Vb = Cintensity	Vb = Lintensity	Vb = Rintensity	Vb = InvenRatio	Vb = lnSize	Vb = SCIndex
Intercept	−0.0023** (−2.33)	0.0010 (0.41)	−0.0026** (−2.34)	−0.0022** (−2.04)	−0.0023** (−2.38)	−0.0090*** (−6.57)	−0.0024** (−2.50)
rDths	−0.0002 (−1.09)	0.0110*** (2.98)	−0.0068*** (−8.33)	0.0062*** (7.35)	−0.0007*** (−4.87)	−0.0352*** (−19.00)	−0.0008*** (−5.22)
Vb	−0.0004** (−2.26)	−0.0031 (−1.49)	0.0002 (0.22)	−0.0001 (−0.56)	−0.0001 (−0.76)	0.0004*** (6.80)	−0.0002 (−0.37)
rDths × Vb	−0.0010*** (−3.24)	−0.0114*** (−3.22)	0.0155*** (7.42)	−0.0026*** (−8.49)	−0.0002** (−2.04)	0.0022*** (18.59)	0.0029*** (6.76)
Fama‐French three factors	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	160,297	148,244	144,471	144,471	156,561	160,297	158,040
R 2	0.3728	0.3759	0.3765	0.3765	0.3746	0.3744	0.3717
F‐statistic	1832.03	1716.58	1676.72	1677.26	1802.68	1843.96	1797.10

Note: The dependent variable is the daily return on individual stocks and the independent variable is the growth rate of daily deaths (the one‐period lag term is considered). In each column, we run a panel regression model by considering the roles of fear sentiment, capital intensity, labor intensity, relative capital‐labor intensity, inventory ratio, firm size, and the firm's industry's position in the supply chain. In the estimation models, Fama‐French three factors were added as control variables, and both industry and province fixed effects were included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

We also run a regression analysis by group for robustness testing, and the results are shown in Table 16. On the one hand, the stock prices of firms with high capital intensity, low labor intensity, high inventory ratio, and small size are more susceptible to the negative impact of the pandemic, which is consistent with our previous conclusions. On the other hand, from the perspective of the industry‐level supply chain, no matter what company characteristics are used as the grouping variable, the coefficients of the interaction terms between rCase and the supplier–customer index are still significantly positive. This shows that the conclusion that downstream companies are more severely affected by the pandemic is sound, and firm‐level characteristics cannot explain this result from the supplier–customer index.

Table 16.

Regression results grouped by firm characteristics

	Capital intensity		Labor intensity		Capital‐labor intensity		Inventory‐to‐revenue ratio		State‐owned
	Low	High	Low	High	Low	High	Low	High	No	Yes
Intercept	−0.0035 (−1.58)	−0.0024** (−2.10)	−0.0042*** (−2.69)	−0.0012 (−0.90)	−0.0012 (−0.84)	−0.0038*** (−2.72)	−0.0035** (−2.32)	−0.0015 (−1.13)	−0.0024 (−1.59)	−0.0023* (−1.86)
rCases(t−1)	−0.0001 (−1.13)	−0.0004*** (−3.30)	−0.0004*** (−2.95)	−0.0002 (−1.59)	−0.0001 (−0.91)	−0.0004*** (−3.68)	−0.0002** (−2.21)	−0.0003*** (−2.67)	−0.0001 (−1.26)	−0.0006*** (−5.10)
SCIndex	−0.0011* (−1.79)	0.0001 (0.15)	−0.0004 (−0.55)	−0.0005 (−0.79)	−0.0004 (−0.70)	−0.0005 (−0.74)	−0.0007 (−1.14)	0.0000 (0.05)	0.0002 (0.35)	−0.0008 (−1.33)
rCases(t−1) × SCIndex	0.0021*** (5.19)	0.0016*** (4.42)	0.0022*** (5.79)	0.0014*** (3.74)	0.0015*** (3.78)	0.0022*** (5.80)	0.0010*** (2.77)	0.0022*** (5.96)	0.0019*** (5.43)	0.0016*** (4.46)
Fama‐French three factors	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Industry‐fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Province‐fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	73,336	72,895	71,380	71,078	70,939	71,519	77,186	77,179	99,982	58,058
R 2	0.3690	0.3816	0.3826	0.3698	0.3660	0.3869	0.3815	0.3680	0.3614	0.4005
F‐statistic	824.11	864.35	866.75	801.47	787.09	884.23	909.73	880.61	1087.41	775.01

Note: The robustness results from the group regression are reported. The dependent variable is the daily return of individual stocks and the independent variable is the growth rate of the daily confirmed cases (the one‐period lag term is considered). We run the panel regression models grouped by capital intensity, labor intensity, relative capital‐labor intensity, inventory ratio, and state‐owned dummy variables. In the estimation models, the Fama‐French three factors were added as control variables, and both industry and province fixed effects were also included. The numbers with ∗∗∗, ∗∗, and ∗ are significant at the 1%, 5%, and 10% levels, respectively, and the t‐statistics are in parentheses.

7. CONCLUSIONS

This study attempted to uncover hidden mechanisms by examining the heterogeneous effects of the COVID‐19 outbreak on stock returns in China from multiple perspectives.

In general, we confirm a very intuitive result that stock prices suffered a significant decline during the pandemic. However, we find the impact to be highly heterogeneous in various industries—industries such as accommodation and catering, leasing and business services, transportation, warehousing, and postal services, were hit harder than the rest. We find that stock prices rose against the trend in the pharmaceutical manufacturing and chemical industries. Finally, fear sentiments surrounding COVID‐19 also resulted in a direct drop in stock prices. This exacerbated the negative impact on stock prices.

We discuss four firm‐level driving factors of this effect to expand our research: capital intensity, labor intensity, inventory‐to‐sales ratio, and firm size. The results show that the higher the capital intensity of a firm, the greater the negative impact of the pandemic on stock prices. If a firm with high capital intensity faced too high capital costs during the pandemic, its stock prices fell. In contrast, the higher the labor intensity, the smaller the adverse effect. We argue that labor‐intensive firms would be less impacted by the pandemic because of lower dismissal costs. We set a dummy variable for state‐owned enterprises to test our hypothesis, and empirical evidence supports this. Considering a possible increase in inventory costs due to the pandemic, we examine the role of the inventory‐to‐sales ratio and find that firms with a higher inventory‐to‐sales ratio were relatively more affected, perhaps because they needed to pay more to manage the inventory backlog. Moreover, we provide evidence of the firm size effect in how the pandemic influenced stock market returns and find that small firm stocks gave lower returns as the pandemic hit them.

Finally, we created a novel industry‐level supplier–customer index to analyze the effect of the pandemic from the supply chain perspective. Our main finding is that companies located downstream in the supply chain are more susceptible to the pandemic than are those located upstream.

APPENDIX A. DEFINITION OF VARIABLES

This appendix lists the main variables used in the empirical analysis.

Variable	Abbreviation	Definition
Stock/industry return	r	The natural log difference of closing prices of a firm or industry
Capital intensity	Cintensity	We define capital intensity as log(TAsset)/log(Orevenue), where TAsset represents the total asset at the end of the last fiscal year, and Orevenue is the operating revenue.
Labor intensity	Lintensity	We define labor intensity as log(Nstaff)/log(Orevenue), where Nstaff represents the number of employees at the end of the last fiscal year, and Orevenue is the operating revenue.
Relative capital‐labor intensity	Rintensity	We construct another variable, relative capital‐labor intensity, to measure the relative capital/labor intensity, which is defined as log(TAsset)/log(Nstaff).
State‐owned dummy	State‐owned	This variable is a dummy variable. When the company is state‐owned, its value is 1; otherwise, 0.
Inventory‐to‐revenue	InvenRatio	The variable is defined as the ratio of inventory to operating revenue.
Firm size	lnSize	The variable is defined as the natural logarithm of the company's market value
Supplier–customer index	SCIndex	A variable to measure the relative position of an industry in the supply chain.

APPENDIX B. SECONDARY INDUSTRY LABELS (WIOD)

The standard industry code is from the China Security Regulation Commission (CSRC), and we match the industry codes from the WIOD database, according to Li et al. (2020).

Code	WIOD industry
H53	Postal and courier activities
L68	Real estate activities
B	Mining and quarrying
J61	Telecommunications
A01	Crop and animal production, hunting, and related service activities
A02	Forestry and logging
G47	Retail trade, except of motor vehicles and motorcycles
G46	Wholesale trade, except of motor vehicles and motorcycles
H49	Land transport and transport via pipelines
H52	Warehousing and support activities for transportation
M72	Scientific research and development
M74_M75	Other professional, scientific, and technical activities; veterinary activities
A03	Fishing and aquaculture
M69_M70	Legal and accounting activities; activities of head offices; management consultancy activities
C18	Printing and reproduction of recorded media
C23	Manufacture of other nonmetallic mineral products
C19	Manufacture of coke and refined petroleum products
C16	Manufacture of wood and of products of wood and cork, except furniture; manufacture of articles of straw and plaiting materials
D35	Electricity, gas, steam, and air conditioning supply
C17	Manufacture of paper and paper products
C24	Manufacture of basic metals
H50	Water transport
E37–E39	Sewerage; waste collection, treatment, and disposal activities; materials recovery; remediation activities and other waste management services
C20	Manufacture of chemicals and chemical products
C21	Manufacture of basic pharmaceutical products and pharmaceutical preparations
C22	Manufacture of rubber and plastic products
R_S	Other service activities
E36	Water collection, treatment, and supply
H51	Air transport
C25	Manufacture of fabricated metal products, except machinery and equipment
I	Accommodation and food service activities
C26	Manufacture of computer, electronic and optical products
C27	Manufacture of electrical equipment
C13–C15	Manufacture of textiles, wearing apparel and leather products
C10–C12	Manufacture of food products, beverages, and tobacco products
C28	Manufacture of machinery and equipment n.e.c.
C29	Manufacture of motor vehicles, trailers, and semitrailers
C31_C32	Manufacture of furniture; other manufacturing
C30	Manufacture of other transport equipment
J62_J63	Computer programming, consultancy, and related activities; information service activities
P85	Education
O84	Public administration and defense; compulsory social security
F	Construction
Q	Human health and social work activities

Liu, Z. , Dai, P.‐F. , Huynh, T. L. D. , Zhang, T. , & Zhang, G. (2022). Industries' heterogeneous reactions during the COVID‐19 outbreak: Evidence from Chinese stock markets. Journal of International Financial Management & Accounting, 1–36. 10.1111/jifm.12166