Supply chain disruption recovery in the evolving crisis—Evidence from the early COVID-19 outbreak in China

Abstract

The speed of recovery from supply chain disruption has been identified as the predominant factor in building a resilient supply chain. However, COVID-19 as an example of an evolving crisis may challenge this assumption. Infection risk concerns may influence production resumption decision-making because any incidents of infection may lead to further shutdowns of production lines and undermine firms’ long-term cash flows. Sampling 244 production resumption announcements by Chinese manufacturers in the early COVID-19 crisis (February–March 2020), our analysis shows that, generally, investors react positively to production resumptions. However, investors perceived the earlier production resumptions were higher risk (indicated by declined stock price). Such concerns were exacerbated by more locally confirmed cases of COVID-19 but were less salient for manufacturers with high debts (liquidity pressure). This study calls for a reassessment of the current disruption management mindset in response to new evolving crises (e.g., COVID-19) and provides theoretical, practical, and policy implications for building resilient supply chains.

1. Introduction

The dynamic and uncertain international business environment has introduced tremendous risks to the now lengthy global supply chain. Operations management (OM) scholars, as well as practitioners, have paid attention to the supply chain disruption brought about by these risks. The literature has investigated how supply chains have been disrupted by high-profile events, such as the Northridge Earthquake (Dahlhamer & Tierney, 1998), the Rana Plaza collapse (Jacobs & Singhal, 2017), the Great East Japan Earthquake (Hendricks et al., 2020, Todo et al., 2015), and U.S.-China trade war (Fan et al., 2022b). A fundamental assumption about recovery from disruption is that it must happen quickly to ensure supply chain continuity and avoid long-term impacts (Ivanov et al., 2017). This field of study aims to guide recovery efficiently in from the disruption to mitigate the liquidity pressure.

However, this assumption neglects contexts in which risks could be embedded in decisions about production resumption. This is exactly the case for the COVID-19 outbreak. Compared with past events causing supply chain disruptions, COVID-19 is unique in terms of its nature as an evolving crisis. Events such as factory collapses, warehouse explosions, and hurricanes commence and conclude swiftly, meaning that the recovery process can start soon after the event. However, because people infected with COVID-19 often display mild or no symptoms, current public health measures cannot guarantee a risk-free return to normal production. In other words, the quick resumption of production for recovery, without considering these risk factors, might lead to further infection clusters in those factories that resume operations, thus requiring additional shutdowns. The unique features of evolving crises, such as the early COVID-19 outbreak, have led us to reconsider the assumptions of the current supply chain disruption literature: should disruption recovery take place as soon as possible amidst an evolving crisis? The assumption of prompt disruption recovery is straightforward: it can mitigate the immediate cash flow pressures introduced by the disruption and avoid bankruptcy caused by debt defaults. However, quick recovery following an early COVID-19 outbreak may involve tremendous uncertainty because of limited knowledge about the pandemic; thus, factories find it difficult to detect and mitigate the risks of infection within the production venues. This risk was demonstrated by 12 reports on infection clusters caused by production resumptions right after the first wave of the COVID-19 outbreak in China, leading to a shutdown of the facilities involved and the quarantining of 514 workers (Sina, 2020). Thus, quick production resumption provides instant gain through short-term cash flow but carries the risk of significant future losses sustained through factory shutdowns caused by infections, thereby undermining the potential long-term cash flow.

The discussion above led us to investigate production resumption amidst an evolving crisis. First, this research aims at answering the following research question (RQ1): How did investors react to the Chinese manufacturers’ production resumption amidst the early COVID-19 outbreak? Using the investor reactions as a proxy of expected cash flow, answering RQ1 can capture the extent to which production resumption can revive cash flow for the Chinese manufacturers.

Second, as the timing of production resumption can implicate infection risks that may lead to future shutdowns, this research aims at answering the following research question (RQ2): Did investors react less positively to production resumption soon after the early COVID-19 outbreak? We examine whether the infection risks associated with the early production resumption undermine investors’ positive reaction to the resumption. The negative investor reactions will consequently lead to a decreased firm value. It is important to answer RQ2 in the Chinese context because of its “world factory” role in the manufacturing sector. The dysfunction of this sector can bring tremendous impact on both domestic and global supply chains.

Through sampling 244 production resumption announcements in Chinese manufacturing sectors from February 3–March 28, 2020, we found that investors generally reacted positively to production resumption after the early COVID-19 outbreak. However, we also found that the earlier the resumptions after the outbreak, the less positive the investors’ reactions, which illustrates the infection risks associated with the production resumptions. We further found that these negative reactions were amplified when there were more confirmed local COVID-19 cases and attenuated when firms had greater liquidity pressures.

These results first challenge the assumption concerning the necessity for quick recovery in the current disruption recovery literature and practices (Chen et al., 2019). We quantify the risks of quick production resumption and reveal the need for risk management in disruption recovery (as well as before disruption; Ivanov, 2020). Second, this study contributes to supply chain risk management by providing evidence of risk propagation in supply chains (Hendricks et al., 2020). Our further analysis results also illustrate how upstream manufacturers and downstream customers evaluate the supply chain risks differently, and provides implications for manufacturers and their customers who are recovering from disruption. These implications are also important for governments when making policies to reboot their economies after evolving crises such as COVID-19.

2. Literature review

COVID-19 is considered a salient risk event that has seriously affected the Chinese manufacturing sector (Lin et al., 2021, Ye et al., 2023). The production resumptions by Chinese manufacturers can be conceptualized as a recovery after disruption by a crisis. Therefore, our study is closely related to two streams of literature, namely, supply chain disruption and disruption recovery. Disruption recovery occurs after supply chain disruption. Decision-making about the recovery process is reactive to the disruptions (caused by risk events) and aims to resume firm operations and cash flow as soon as possible (Ivanov et al., 2017).

Supply chain disruption is caused by unplanned and unexpected events that interfere with the physical flows (e.g., materials and goods) in supply chains (Craighead et al., 2007, Ellis et al., 2010, Kleindorfer and Saad, 2005). Disruption events are induced by exposure to serious turbulence and uncertainty (Christopher & Peck, 2004), causing operations to deviate from the normal plan (Zhao et al., 2011). A glitch in a supply chain can undermine firm sales growth, cost efficiency, inventory performance (Hendricks & Singhal, 2005), and firm value (Hendricks & Singhal, 2003). The disruptions caused by regulatory, catastrophic, and infrastructural crises were found to raise serious concerns for firms (Zsidisin et al., 2016). The major disruptions caused by COVID-19 make disruption management even more important, given the serious impacts of supply chain disruptions. Literature has developed two approaches to managing disruption: proactive and reactive (Ivanov et al., 2017).

The proactive approach focuses on anticipating future disruption events and creating protections to cushion against negative impacts. The proactive approach is based on the risk management process in terms of risk identification, assessment, mitigation, and monitoring. Such proactive practices will be implemented before risk events occur. Risk is a type of uncertainty that can produce unexpected negative consequences;1 it is a multidimensional concept. In supply chains, risk events have been categorized as macro risks including natural and man-made ones, and micro risks including demand, manufacturing, supply, and infrastructural risks (Ho et al., 2015, Sodhi et al., 2012). Macro risks such as COVID-19, incorporated alongside the risk-taking propensity of the managers, can increase organizational risks (Li and Tang, 2010, Palmer and Wiseman, 1999). Organizational risk is reflected in the firm’s future income stream or cash flow uncertainty (Palmer & Wiseman, 1999). The cash flow uncertainty can trigger risk-averse behavior from investors in selling firm stocks, leading to a decreased firm value (Fan et al., 2020).

The reactive approach focuses on contingency decision-making in the face of unexpected disruption events. The reactive practices can facilitate disruption recovery processes; however, research on recovery management during disruption events is scarce (Ivanov et al., 2017). A few exceptions focus on recovery services during disruptions to public transportation (Li et al., 2015) and the use of mathematical tools to forecast recovery times (Ivanov et al., 2017). Recovery in post-disruption times should include technical, capacity, and business aspects, and it should be followed by a learning process for continuous improvement (Chen et al., 2019). The capability to recover from a disruption quickly has widely been viewed as a demonstration of supply chain resilience (Blackhurst et al., 2011). The speed at which to recover has been identified as the most predominant factor in disruption recovery (Chen et al., 2019). This is understandable because the disruption poses direct threats to firms’ short-term cash flow and affects firm survival.

The discussion above illustrates the current mindset of supply chain disruption management: managers should focus on risk identification, assessment, and mitigation before disruption events; in the meantime, recovery practices should aim at a quick recovery once disruption has occurred. However, COVID-19 has introduced the possibility that quick recovery decisions may produce extra risks that could affect firms’ long-term cash flow. If so, the predominant factor of speed in disruption recovery should be reconsidered, and risk management in the post-disruption and pre-disruption periods is called for (Ivanov, 2020).

3. Hypothesis development

3.1. Stock price reaction to production resumption

The timeline of the early COVID-19 outbreak in China is illustrated in Appendix Fig. A. The novel coronavirus was first observed in late December 2019 in the city of Wuhan, the capital of China’s Hubei province. It then spread swiftly worldwide. As of October 2022, over 623 million cases had been confirmed across the globe, resulting in 6.55 million deaths (World Health Organization [WHO], 2022). In China, the most affected country, in the first quarter of 2020 the outbreak of COVID-19 corresponded with the timing of Chinese New Year celebrations. During that time factories shut down for two to four weeks to allow workers to attend family gatherings, which often involves traveling. To control the spread of the virus, the Chinese government mandated extended factory shutdowns until February 9. The extended shutdown seriously disrupted the operations and undermined the cash flow of Chinese manufacturers. The shutdown of world factories also caused local and global disruption to supply chains (Ivanov and Das, 2020).

When the positive signs began emerging in this battle against COVID-19, the central government of China required local governments to take measures to promote the resumption of production while still ensuring strict virus control. Resuming production can revive cash flows for manufacturers, which in the case of COVID-19 had been stopped by the prolonged shutdown during the Chinese New Year.

According to the net present value (NPV) decision rule, investors will take the investment project that can offer the highest NPV, which is measured by the difference between the present value of a project’s benefits and the present value of its costs. Expected future cash flow is a critical parameter in computing NPV. Assuming that other factors remain unchanged, the higher the expected future cash inflow (i.e., positive cash flow), the higher the present value of the project’s benefits; conversely, the higher the expected future cash outflow (i.e., negative cash flow), the higher the present value of the project’s costs (Berk et al., 2019). Therefore, announcing production resumption can mitigate an investor’s concern regarding a firm’s cash flow and can consequently affect stock price. We thus hypothesized the following:

H1: Investors reacted positively to the production resumption announcement amidst the early COVID-19 outbreak.

3.2. Production resumption timing and stock price reaction

After the mandated shutdown in response to the outbreak, factories were faced with the decision of when to resume production. The evaluation of production resumption timing is a dilemma between the needs for instant cash flow and the necessity of mitigating infection risk within the factory. Despite the fact that we expect an overall positive reaction to production resumption, investors may be less optimistic about those who resumed early.

Following the production disruptions caused by the COVID-19 outbreak, firms may opt to resume production earlier. In such cases, early production resumption results in a shorter time span between the disruption and resumption. Intuitively, resuming production earlier enables firms to better meet the demand during the outbreak, which in turn benefits the firm's cash flow. As a result, investors are likely to view such early production resumption decisions favorably.

However, in the production resumptions during the early COVID-19 outbreak, migrant workers were required to travel from their hometowns to their factory locations. Given that the novel coronavirus is highly contagious and infected people can be asymptomatic, the risk of COVID-19 infection increased tremendously for these workers during travel and at the workplace. If any infections were found at a factory, all workers were quarantined, and the factory would be mandated to undergo a further shutdown.

The risks of early production resumption during the COVID-19 crisis were demonstrated by anecdotal cases. For example, there were cluster infections that broke out in PepsiCo’s Beijing factory; the factory was shut down for over a month after the first confirmed COVID-19-infected worker was found there. The risks were also demonstrated by epidemiological simulation studies. Wang et al. (2020a) estimated that the risk of a second COVID-19 outbreak was high when firms in Hubei resumed work on February 17 and February 24, 2020, while the risk was substantially smaller when the resumption day was on March 2, 2020.

Therefore, the resolution to resume production early involves a higher level of uncertainty. Managers had little information about the spread pattern of the virus because of its evolving nature. Second, the limited amount of time and knowledge made it difficult for factories to prepare for precautions, such as purchasing personal protection equipment and training the operational workers to take proper social distancing measures. Therefore, the uncertainty made it difficult for managers to mitigate infection risks after the resumption.

Infection risks can create uncertainty in future cash flow because any confirmed cases can lead to a future factory shutdown. The loss aversion nature of investors may make such a concern undermine the short-term cash flow gain from production resumption. These investors then would be more likely to sell the firm’s stock if they perceived the firm’s long-term cash flow to be at risk because of early production resumption amid the COVID-19 crisis. Therefore, we hypothesized the following:

H2: Manufacturers’ investors react less positively when production resumption is announced earlier amidst the early COVID-19 outbreak compared to those announced later.

3.3. The role of infection risk and liquidity stress in the effect of early production resumption

Future infection risks and short-term cash flow are two contradictory evaluation factors in the dilemma concerning COVID-19 production resumption. The variations in infection risks and liquidity pressures faced by the manufacturer should affect the risk evaluation results. First, the worsening pandemic situation may shift attention further to the end of infection risk. The number of newly confirmed cases was a concerning topic among the Chinese public during the early period of the crisis (Zhao et al., 2020). Because of the lack of knowledge about the disease, the increasing number of confirmed cases increased public anxiety about the infectiousness of the virus. The larger number of newly confirmed cases reflected that the crisis was more seriously evolving; thus, managers faced increased levels of uncertainty when making decisions to resume production. Therefore, newly confirmed cases in a firm’s location exacerbate concerns about infection risk in a factory causing a long-term shutdown, which reduces the overall value of resuming production early. We thus hypothesized the following:

H3: The number of newly confirmed cases amplifies the negative reaction of investors to manufacturers’ announcements about early production resumption.

Moreover, we argue that when facing significant liquidity stress, managers have a compelling rationale to resume production earlier, as the immediate financial needs of the firm outweigh potential long-term risks. In such situations, investors may perceive early production resumption as a vital and necessary move to maintain business continuity, protect their investments, and ensure the company's short-term financial stability. This perception can lead to a more positive stock market reaction following announcements of early production resumption.

From an investor's perspective, liquidity pressures can cause a shift in priorities, placing greater emphasis on short-term cash flow over potential infection risks. Investors may view early production resumption decisions more favorably in these circumstances, as they recognize the urgent need to address the firm's financial concerns and secure its survival. This understanding aligns with recent studies that found increased debt levels can lead firms to prioritize financial stability, even at the expense of workers' health and safety (Pagell et al., 2019).

During the Chinese New Year shutdown amid the COVID-19 crisis, Chinese factories were obligated to cover the costs of remote workers' basic salaries (as mandated by the government), rent, interest, and other expenses, despite the suspension of income. These liquidity pressures were intensified for firms burdened with larger corporate debts. Creditors could potentially file for the liquidation of factory assets if companies failed to repay their debts during the crisis, leading to bankruptcy (Wang et al., 2020b). Given these potential consequences, investors may view early production resumption as a necessary measure to protect their investments, leading to a more positive reaction to such announcements.

H4: Liquidity stress attenuates the negative reaction of investors to a manufacturer’s announcement about early production resumption.

4. Data and methods

4.1. Data collection

To test our hypotheses, we sampled firms listed on the Shanghai/Shenzhen Stock Exchange. We first obtained a list of Chinese public firms from the China Stock Market and Accounting Research (CSMAR) database. We searched each firm’s name with “resumption of work” and “resumption of production” as keywords in announcements from the media report databases of CSMAR. In the initial search, the start date was set to January 23, 2020, the day when Wuhan city was locked down at 10 a.m. On the same day, 31 provinces implemented the public health level-1 response and shut down industries. The State Council of China announced that 98% of large-scale enterprises had resumed operations by March 28, 2020 (Gov.cn, 2020). We thus set our end date to March 28. We obtained 299 production resumption announcements from this period (January 23–March 28). Our research window starts on January 23, thus the assumption is that the production of Chinese manufacturers was disrupted on this date. This can be validated because this is the time during the Chinese New Year break when migrant workers were going back home for family reunions. Therefore, Chinese manufacturers would be running with minimal capacity during this period.

We then discarded seven firms that were not in manufacturing sectors (Category C in the industrial classification). To ensure the accuracy of the information in all announcements, an author assisted by two research assistants double-checked the timing of production resumption and the confounding events of 292 announcements. Then we removed two announcements accompanying confounding events including earnings announcements, management changes, lawsuits, and other operational decisions (Ding et al., 2018). We also discarded 46 firms that announced production resumption before February 2. This was because trading on the Chinese stock market was suspended for the Chinese New Year and because of the COVID-19 outbreak until February 9, 2020. Consequently, there is a lack of stock price data for the trade suspension period. Finally, we yielded 244 qualified announcements from 244 manufacturing firms for data analysis.2 Appendix Table A summarizes the data set development process. The sample firms are from 25 provinces. Appendix Table B shows the province distribution of firm announcements. The stock price and accounting data were then collected from the CSMAR database. Finally, we collected pandemic-related data for Chinese firms from the COVID-19 & Economic Research module of the CSMAR database. The module recorded COVID-19-related data including daily population, city migration, and newly confirmed cases in China. Table 1 shows the descriptive statistics of the sample firms.

Table 1.

Descriptive statistics of the sample firms in the 244 announcements.

	Total assets(RMB000,000)	Sales(RMB000,000)	Net income(RMB000,000)	Number ofEmployees(000)	ROA	Debt-to-equityratio	Price-to-bookvalue	Market value(RMB000,000)	Outstandingshare(000,000)	Stockprice(RMB)
Mean	20,930.620	16,687.650	1,064.539	9.382	0.042	0.973	1.899	32,774.130	1,232.633	19.457
Median	4,814.126	2,663.513	166.007	2.717	0.043	0.660	1.581	7,275.877	538.007	10.840
Std. deviation	68,430.660	65,597.590	4,203.061	25.702	0.082	1.394	1.116	119,742.600	2,454.945	64.260
Maximum	849,333.300	843,324.400	43,970.000	229.154	0.240	19.224	8.344	1527248.000	27,288.760	985.000
Minimum	522.642	98.909	−4,691.926	0.206	−0.429	0.075	0.757	1,105.781	19.040	1.240

4.2. Investor’s reaction to production resumption

We used an event study approach to examine investor reaction to the production resumption announcement amidst the early COVID-19 outbreak. In this study, the event is the announcement by a Chinese manufacturer claiming that production would be resumed. We first defined day t as the date of manufacturers announcing production resumption decisions. Similar to prior event studies (Asquith et al., 2005, Bradley et al., 2003, Madsen and Rodgers, 2015, Masulis et al., 2007), we analyzed a five-day event window, including two days before (Day − 2), the event day (Day 0), and two days after the production resumption (Day + 2). We included the days before the production resumption to account for possible information leakage before the formal production resumption announcement. Day 2 was also included because some investors may have delayed their reactions to the announcements. Specifically, the calculation of cumulative abnormal return (CAR) was the sum of all the daily abnormal returns from Day − 2 to Day + 2 as follows:3

$${\mathit{CAR}}_i\left[-2,+2\right]=\sum _{t=-2}^2{\mathit{AR}}_{i,t}$$ (1)

To calculate abnormal return (AR_it), we first used the Fama–French four-factor model to estimate normal return (Fama and French, 1993, Carhart, 1997). This model is defined as follows:

$$R_{\mathit{it}}={\alpha }_i+{\beta }_1({\mathit{RM}}_{\mathit{it}}-{\mathit{RF}}_{\mathit{it}})+{\beta }_2{\mathit{SMB}}_{\mathit{it}}+{\beta }_3{\mathit{HML}}_{\mathit{it}}+{\beta }_4{\mathit{UMD}}_{\mathit{it}}+{\epsilon }_{\mathit{it}}$$ (2)

where $R_{\mathit{it}}$ is the stock return of firm i in Day t; ${\mathit{RM}}_{\mathit{it}}$ is the market return of the CSI 300 index; ${\mathit{RF}}_{\mathit{it}}$ is the 3-month time deposit yield; ${\mathit{SMB}}_{\mathit{it}}$ is the difference between the returns on diversified portfolios of stocks with small (the bottom 50%) and large (the top 50%) sizes based on market capitalization; ${\mathit{HML}}_{\mathit{it}}$ is the difference between the returns on diversified portfolios of stocks with high (the top 30%) and low (the bottom 30%) book-to-market ratios; and ${\mathit{UMD}}_{\mathit{it}}$ is the difference between the returns on diversified portfolios of stocks with high (the top 50%) and low prior year daily returns (the bottom 50%).4 ${\epsilon }_{\mathit{it}}$ is the error term. All variables used the data from a 200-day estimation period (from Day –210 to Day –11, prior to the event date). The estimation window stopped at Day –11 to mitigate concerns about the effect of the announcement on the estimation.

Then, we calculated the expected return for the event period (–210, –11) as follows:

$$E\left(R_{i,t}\right)={\alpha }_i^{\prime }+{\beta }_1^{\prime }({\mathit{RM}}_{\mathit{it}}-{\mathit{RF}}_{\mathit{it}})+{\beta }_2^{\prime }{\mathit{SMB}}_{\mathit{it}}+{\beta }_3^{\prime }{\mathit{HML}}_{\mathit{it}}+{\beta }_4^{\prime }{\mathit{UMD}}_{\mathit{it}}$$ (3)

where α' and β' are model parameters to be estimated using daily data from Day –210 to Day –11 prior to the event date in Equation (2). For every day in the event period, we calculated a measure for stock return abnormality as follows:

$$A{R}_{i,t}=R_{i,t}-E\left(R_{i,t}\right)$$ (4)

where AR_i,t is the abnormal return of firm i on day t, R_i,t is the actual return on stock i on day t, and E(R_i,t) is the expected return on stock i on day t if the event had not occurred.

To examine H1, we used the t-test and Wilcoxon signed-rank test to determine the statistical significance of the mean and median abnormal returns, respectively. We also employed the binomial sign test to examine if the percent of negative abnormal returns were significantly different from the null of 50%.

4.3. Heckman analysis

To examine H2 through H4, we employed Heckman two-stage analysis of the effect of production resumption timing on investor reaction and the role of infection risk and liquidity stress in the effect of early production resumption. We noted that the announcement of production resumption is observational. Whether announcing the decision or not, firms need to resume production sooner or later. This may raise concerns about self-selection bias among the announcing firms. It is possible that the firms that chose to announce production resumption had certain systematic characteristics. For example, firms with better management skills tend to increase firm transparency, and thus may be more likely to announce production resumption; concurrently, investors may have higher confidence in these firms, meaning that the firms would have a better CAR. To address this potential problem of selection bias, we used a Heckman two-stage approach (Heckman, 1979).

The first stage of the Heckman process involved a probit regression on the choice of the announcement. A detailed description of the probit model development and analysis steps was included in Appendix C and Table C. We then generated an inverse Mills ratio (IMR) variable from the probit analysis. In the second step, the IMR was included as the self-selection correction parameter in our regression analysis to check for endogeneity.

4.4. Independent variables

H2 examined the impacts of the timing of production resumption. We used the date of January 23, 2020 (Wuhan’s commencement of lockdown), as the benchmark and determined the number of days between each announcement date and the lockdown date. The mean resumption day of our sample firms is 22.131 (SD = 6.874). It ranges from 11 days to 58 days; thus, the distribution was skewed to the right. The timing of production resumption resulted in a skewed problem because the value of skewness is greater than 2 (Bulmer, 1979).5 It would result in the estimation not being robust (Hansen, 2018). To address this problem, following Marquis and Lee (2013) and Song et al. (2016), the natural logarithm of it was used to correct for skewness. Smaller timeframes indicate a rush, and thus, a riskier decision. Therefore, we reversed the scale of the variable to indicate early production resumption decisions. The calculation of early production resumption was as follows:

$$\mathit{Early}productionresumption=-log[\mathrm{p}\mathrm{r}\mathrm{o}\mathrm{d}\mathrm{u}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{u}\mathrm{m}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}announcementdate-lockdownofWuhandate]$$ (5)

We verified this variable by regressing the timing of production resumption on research and development (R&D) intensity (Ln [R&D expense/total assets]).6 R&D intensity has been widely used as an indicator of firms’ managerial risk-taking (Xu et al., 2019). The results show that R&D intensity is negatively associated with the number of days taken to resume production (coefficient =  − 1.750, p < 0.05). These results indicate that an early production resumption decision is a reasonable proxy for the signaling of an increased risk to the market.

We postulated that the impacts of production resumption timing would be contingent on infection risks and liquidity stress. Infection risk was indicated by newly confirmed cases in each firm’s location. We determined the number for each Chinese firm’s province. The data were calculated as the average numbers for the three days before the production resumption announcement. The higher the number of newly confirmed cases, the higher the infection risk.

Liquidity stress was measured by the ratio of the book value of debt to the market value of equity. This ratio captures the firm’s debt pressure, which indicates its limitations in terms of financial slack (Bhandari, 1988, Fama and French, 1993). Liquidity stress was standardized according to the industry mean and SD (two-digit industry code) in the same year (Hendricks et al., 2009).

4.5. Controls measures

We used the following control factors to ensure that our results were robust. We controlled for firm size as the logarithm of the number of firm employees and included ROA to measure firm profitability (Wiengarten et al., 2017). Firm age was used to control for operational experience (Fan et al., 2020). Larger, more profitable, and operationally experienced firms are more likely to have a stronger financial position and management capability, and thus, to be less affected by COVID-19. The ratio of operating cash flow to total liabilities was included because investors may have more serious concerns about a firm’s bankruptcy when that firm has limited cash flow for debt repayment (Capon et al., 1990). Inventory leanness was measured by the number of inventory days (Hendricks et al., 2009, Wiengarten et al., 2017). More days of inventory corresponded to more inventory slack and loosely coupled operations. We reversed the scale of inventory days to indicate leanness. Inventory leanness was standardized according to the industry mean and SD (two-digit industry code) in the same year (Hendricks et al., 2009). Capacity slack was measured by the ratio of fixed assets to sales and standardized according to the industry mean and SD (two-digit industry code; Wiengarten et al., 2017). Firms with more capacity slack may have a greater buffer to cope with disruptions. A state-owned enterprise (State) dummy, which obtained a value of 1 if the firm was controlled by the state and 0 otherwise, was incorporated (Cui and Jiang, 2012, Zhou et al., 2017). We also considered the number of announcements in the same city and industry that were made prior to the disclosure of each firm, respectively. These two variables controlled for the mimetic pressures of announcements about production resumption from firms in the same location and industry. Last, we included dummy variables for the province in which each firm had its headquarters to capture province-specific effects. The variable definitions are summarized in Appendix Table E.

5. Results

5.1. Event analysis of stock price reaction to the production resumption

Panel A of Table 2 presents the event analysis of stock price reaction to the production resumption and examines H1. The results generally show that the stock market reacts significantly positive to the production resumption. For the CAR in the full research window of Day − 2 to Day 2, the mean CAR is 1.233% (p < 0.01). The nonparametric test results provide further support. The median CAR for Day − 2 to Day 2 is 0.784% (p < 0.01). 55.738% of firms experienced a positive CAR (p < 0.05). The market reaction can be captured starting from Day − 1 to Day 0 with the mean CAR of 0.842% (p < 0.01). The effect can also be captured at the day of announcement (Day 0), with the mean AR of 0.647% (p < 0.01). This shows that the investor generally favors the announcement of production resumption and expects the instant cash flow gain, supporting H1.

Table 2.

Event analysis of production resumptions.

Event day(s)	N	Mean	p	Median	p	% Positive	p
Panel A: Sample group (244 obs)
Day − 2 to Day − 1	244	0.388%	0.180	0.129%	0.469	52.459%	0.241
Day − 1 to Day 0	244	0.842%	0.008	0.400%	0.039	53.689%	0.138
Day 0	244	0.647%	0.005	−0.089%	0.364	47.951%	0.759
Day 0 to Day 1	244	0.905%	0.007	0.292%	0.068	55.738%	0.042
Day 1 to Day 2	244	0.294%	0.314	0.027%	0.518	50.000%	0.526
Day − 2 to Day 2	244	1.233%	0.010	0.784%	0.009	58.197%	0.012
Panel B: Control group (244 obs)
Day − 2 to Day − 1	244	−0.159%	0.516	−0.299%	0.108	43.443%	0.983
Day − 1 to Day 0	244	−0.137%	0.514	−0.125%	0.202	46.311%	0.888
Day 0	244	−0.169%	0.288	−0.250%	0.033	40.984%	0.998
Day 0 to Day 1	244	−0.162%	0.482	−0.235%	0.145	46.311%	0.888
Day 1 to Day 2	244	0.345%	0.153	−0.164%	0.943	46.311%	0.888
Day − 2 to Day 2	244	−0.024%	0.951	−0.433%	0.361	45.902%	0.911
Panel C: Mean, median, and % positive differences between sample and control groups
Day − 2 to Day − 1	244	0.558%	0.138	0.118%	0.187	52.049%	0.282
Day − 1 to Day 0	244	0.944%	0.010	0.766%	0.017	56.967%	0.017
Day 0	244	0.746%	0.006	0.422%	0.004	58.607%	0.004
Day 0 to Day 1	244	0.876%	0.022	0.783%	0.035	55.328%	0.055
Day 1 to Day 2	244	−0.048%	0.889	0.088%	0.991	50.410%	0.475
Day − 2 to Day 2	244	1.156%	0.042	1.065%	0.034	56.557%	0.024

However, there may be endogeneity in estimating the CARs. To examine whether the market reaction depends on the production resumption announcement or other unobservable factors, we followed Ding et al. (2018) and Paulraj and de Jong (2011) and used the matched-pair design to account for the unobservable. Following Harris and Bromiley (2007), the firms with production resumption announcements (i.e., the sample group) were one-to-one matched to the 244 sample firms without production resumption announcements (i.e., the control group) based on nearest-neighbourhood annual sales in 2019 in the same two-digit standard industry classification code industry. Panel B of Table 2 presents the CARs based on 244 control firms. We used parametric t-tests, the Wilcoxon signed-rank test, and the binomial sign test to test whether the CARs were significant. The results suggest that CARs in the control group (Panel B) are not significant in our event window (Day − 2 to Day 2; p greater than 0.1). This is the opposite of the results in the sample group, suggesting that the market reaction is significantly influenced by the production resumption announcement, but not affected by other unobservable factors. In addition, Panel C of Table 2 indicates the CAR differences between the sample and control groups. The results suggest that the sample group’s CARs are significantly higher than those of the control group, supporting H1 and suggesting that investors reacted positively to the production resumption announcement amidst the early COVID-19 outbreak.

In addition, our sample announcements were clustered in and around the one-month period after Wuhan’s lockdown, which could result in the problem of time clustering and bias in the statistical inference results (Brown & Warner, 1980). To address the clustering bias, we considered crude dependence adjustment, as advocated by Brown and Warner (1985). Table 3 presents the adjusted abnormal returns. The mean adjusted ARs for Day 0 and the mean adjusted CARs for Days [0, 1] and Days [−2, +2] remain positive and significant, respectively. The results are consistent with those presented in Table 3, supporting H1.

Table 3.

Crude dependence adjustment.

Event day(s)	N	Mean	t-value	p-value
Day − 2 to Day − 1	244	0.388%	1.759	0.080
Day − 1 to Day 0	244	0.842%	3.817	0.000
Day 0	244	0.647%	2.933	0.004
Day 0 to Day 1	244	0.905%	4.102	0.000
Day 1 to Day 2	244	0.294%	1.333	0.184
Day − 2 to Day 2	244	1.233%	3.535	0.000

5.2. Heckman analysis results

Table 4 presents the descriptive statistics and the correlations among the variables subjected to regression analysis. Concerns about multicollinearity are not critical because most of the correlation coefficients among the key variables are lower than 0.48 and no variance inflation factor (VIF; see Appendix Table D) is larger than 2.3 (Leiblein et al., 2002, Marquis and Qian, 2014). For the second stage of the Heckman analysis,7 bootstrapping analysis was applied to all models in Table 5 to correct for the finite-sample bias of the OLS estimator (Beran, 1983, Chang et al., 2006, Horowitz, 2001). According to the central limit theorem, small samples may not be sufficient for the assumption of the OLS estimator that error terms should be normally distributed. Therefore, the bootstrap procedure is appreciably better under the finite-sample situation since it resamples the residuals using the model’s own assumption (Beran, 1983, Freedman and Peters, 1984). We also clustered the standard errors according to industry code to account for industry variations.

Table 4.

Descriptive statistics and correlations.

No.	Variable	1	2	3	4	5	6	7	8	9	10	11	12	13
1	CAR %
2	Early production resumption (log)	−0.072
3	Infection risk	0.023	0.105
4	Liquidity stress	−0.025	0.041	0.015
5	Inventory leanness	−0.011	0.029	0.008	0.08
6	ROA	−0.020	0.242***	0.011	−0.313***	0.054
7	Operating cash flow to debt ratio	−0.032	−0.046	−0.021	−0.463***	0.246***	0.424***
8	Capacity slack	−0.008	−0.103	−0.020	0.04	−0.228***	−0.224***	−0.058
9	Firm size	−0.025	0.187***	0.027	0.472***	0.271***	0.034	−0.153**	−0.155**
10	Firm age	0.01	−0.011	0.068	−0.041	0.019	−0.082	0.018	0.011	−0.039
11	Announcements in same city	0.08	−0.367***	−0.060	−0.090	0.024	−0.024	0.048	0.107*	−0.128**	−0.121*
12	Announcements in same industry	−0.088	0.067	0.037	−0.013	−0.134**	0.056	−0.006	−0.002	0.071	0.089	−0.129**
13	State	0.059	0.093	0.116*	0.188***	0.104	0.021	−0.120*	−0.085	0.204***	0.137**	−0.115*	0.026
	Mean	1.233	−0.155	−0.008	0.095	0.134	0.042	0.246	−0.070	8.061	20.189	15.094	11.701	0.107
	Std.dev.	7.381	0.805	1.034	0.96	0.751	0.081	0.313	0.741	1.319	5.262	14.883	11.814	0.309

Table 5.

Heckman analysis.

	Cumulative abnormal stock return (Fama–French four-factors model) [Day − 2, Day + 2]
Independent variables	(1)	(2)	(3)	(4)
Early production resumption	−1.026*	−1.397**	−1.583***	−1.955***
	(0.588)	(0.599)	(0.567)	(0.553)
Early production resumption × Infection risk		−3.342***		−3.347***
		(0.495)		(0.497)
Early production resumption × Liquidity stress			1.565***	1.567***
			(0.391)	(0.395)
Infection risk	0.548	−2.302	0.628	−2.226
	(2.100)	(4.738)	(1.658)	(4.216)
Liquidity stress	−0.136	−0.145	−0.085	−0.094
	(0.860)	(0.869)	(0.957)	(0.962)
Inventory leanness	0.212	0.342	0.363	0.493
	(0.521)	(0.552)	(0.401)	(0.440)
ROA	11.555	11.112	12.094	11.651
	(11.199)	(11.099)	(11.442)	(11.317)
Operating cash flow to debt ratio	−1.489	−1.512	−2.300	−2.324
	(2.041)	(1.996)	(1.983)	(1.932)
Capacity slack	−0.060	0.127	0.065	0.253
	(0.525)	(0.618)	(0.572)	(0.670)
Firm size	0.892	0.954	1.014	1.075
	(0.948)	(0.910)	(0.905)	(0.870)
Firm age	0.008	0.030	0.028	0.050
	(0.041)	(0.048)	(0.040)	(0.045)
Announcements in same city	−0.028	−0.025	−0.036	−0.033
	(0.043)	(0.042)	(0.046)	(0.044)
Announcements in same industry	−0.077**	−0.079***	−0.101**	−0.102***
	(0.033)	(0.029)	(0.044)	(0.039)
State	2.116	1.804	2.138	1.826
	(1.675)	(1.838)	(1.630)	(1.779)
IMR	7.885	7.969	8.332*	8.416*
	(5.135)	(5.007)	(5.030)	(4.915)
Intercept	−21.937	−23.482	−23.971	−25.521*
	(16.258)	(15.472)	(16.033)	(15.279)
Province dummy	YES	YES	YES	YES
R2 (%)	9.3	11.3	12.1	14.1
Number of observations	244	244	244	244

Notes: ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Standard errors clustered at the industry level are reported in parentheses.

Table 5 uses the CAR of the Chinese manufacturers as a dependent variable and tests H2, H3, and H4. Model 1 in Table 5 includes all the control variables as well as the hypothesized effect of manufacturers’ early production resumption on manufacturer investor reactions (H2). The impact of early production resumption8 on CAR is negative (coefficient =  − 1.026) and statistically significant at the 90% confidence interval, as predicted. We defined those firms with one standard deviation above the mean in early production resumption as the firms that resumed production early. As Appendix Fig. B indicates, in moving the level of early production resumption from the mean to one standard deviation above the mean, the CAR decreases by 0.826 (0.805*−1.026). Therefore, the results of Model 1 support H2 by confirming that manufacturers’ investors react less positively to early production resumption announcements.

Model 2 in Table 5 examines whether H2 would be moderated by infection risks.9 The interaction term (early production resumption × infection risk) is negative (coefficient =  − 3.342) and statistically significant at the 99% confidence interval. As Appendix Fig. C indicates, in moving the infection risk from a low level (mean − SD) to a high level (mean + SD), the negative effect (slope) of decisions for early production resumption on manufacturers’ investor reactions is enhanced by 6.911. The change in the slope is 494.72% of the effect of the early production resumption decision (−1.397 in Model 2 of Table 5). Therefore, the results in Model 2 provide support for H3.

Model 3 in Table 5 examines whether H2 would be moderated by standardized liquidity stress. The interaction term (early production resumption × liquidity stress) is positive (coefficient = 1.565) and statistically significant at the 99% confidence interval. As Appendix Fig. D indicates, in moving liquidity stress from a low level (mean − SD) to a high level (mean + SD), the negative effect (slope) of early production resumption decisions on manufacturers’ investor reactions is attenuated by 3.005. The change in slope is 189.98% of the effect of the early production resumption decision (−1.583 in Model 3 of Table 5). Therefore, the results in Model 3 provide support for H4.

6. Robustness checks

6.1. Alternative event window

The five-day period may be quite extensive and affect our results. To address this issue, we used a shorter event window (−1,1) and reran all models. The results, reported in Table 6 , suggest that our findings are robust. Specifically, the results show that the earlier production resumptions reduce the positive reaction of investors. Such concerns were exacerbated by more locally confirmed cases of COVID-19, but were less salient for manufacturers with high debts (liquidity pressure).

Table 6.

Heckman analysis (robustness check 1: three-day event window).

	Cumulative abnormal stock return (Fama–French four-factors model) [Day − 1, Day + 1]
Independent variables	(1)	(2)	(3)	(4)
Early production resumption	−1.475***	−1.695***	−1.731***	−1.952***
	(0.362)	(0.342)	(0.416)	(0.394)
Early production resumption × Infection risk		−1.984***		−1.986***
		(0.722)		(0.720)
Early production resumption × Liquidity stress			0.719**	0.720**
			(0.306)	(0.310)
Infection risk	0.046	−1.646	0.083	−1.611
	(2.262)	(4.204)	(2.047)	(3.959)
Liquidity stress	0.171	0.165	0.194	0.188
	(0.642)	(0.645)	(0.694)	(0.695)
Inventory leanness	−0.202	−0.125	−0.133	−0.056
	(0.843)	(0.882)	(0.820)	(0.861)
ROA	5.711	5.448	5.959	5.696
	(9.151)	(9.104)	(9.184)	(9.116)
Operating cash flow to debt ratio	−0.927	−0.941	−1.300	−1.314
	(1.411)	(1.364)	(1.332)	(1.275)
Capacity slack	0.177	0.288	0.235	0.346
	(0.598)	(0.697)	(0.630)	(0.729)
Firm size	0.619	0.656	0.675	0.712
	(0.884)	(0.851)	(0.863)	(0.833)
Firm age	−0.075**	−0.062*	−0.066*	−0.053
	(0.037)	(0.038)	(0.035)	(0.034)
Announcements in same city	−0.057	−0.055	−0.061	−0.059
	(0.046)	(0.046)	(0.046)	(0.046)
Announcements in same industry	−0.027	−0.028	−0.038	−0.039
	(0.024)	(0.020)	(0.030)	(0.026)
State	1.034	0.849	1.044	0.859
	(1.577)	(1.739)	(1.550)	(1.712)
IMR	4.432	4.481	4.637	4.687
	(5.665)	(5.463)	(5.579)	(5.387)
Intercept	−11.342	−12.260	−12.277	−13.196
	(17.241)	(16.191)	(16.970)	(15.956)
Province dummy	YES	YES	YES	YES
R2 (%)	12.5	13.6	13.4	14.5
Number of observations	244	244	244	244

6.2. Alternative CAR

To facilitate a comparison between this study and the previous event study, we adopted a market model to recalculate CAR for Chinese manufacturers. Specifically, we used the market model to arrive at normal returns (Beaver, 1981, Brown and Warner, 1980, Brown and Warner, 1985, Dyckman et al., 1984). This model is defined as follows:

$$R_{i,t}={\alpha }_i+{\beta }_i{R}_{m,t}+{\epsilon }_{i,t}$$ (3)

where R_i,t is the return of firm i on day t and R_m,t is the market return on day t. The market return was calculated by the Shanghai Shenzhen 300 index. The estimation period was from day − 210 to Day − 11 prior to the event date. We then reran the regression analysis based on the CARs derived from the market model, and the analysis is presented in Table 7 . The results are largely identical to our main results and provide additional support for all the hypotheses.

Table 7.

Heckman analysis (robustness check 2: market model).

	Cumulative abnormal stock return (Market model)[Day − 2, Day + 2]
Independent variables	(1)	(2)	(3)	(4)
Early production resumption	−1.454***	−1.925***	−1.770***	−2.243***
	(0.545)	(0.583)	(0.470)	(0.493)
Early production resumption× Infection risk		−4.246***		−4.249***
		(0.431)		(0.423)
Early production resumption× Liquidity stress			0.890**	0.892**
			(0.361)	(0.374)
Infection risk	1.284	−2.336	1.330	−2.293
	(0.802)	(2.177)	(1.011)	(1.891)
Liquidity stress	−0.436	−0.448	−0.407	−0.419
	(0.874)	(0.891)	(0.917)	(0.931)
Inventory leanness	0.220	0.385	0.305	0.471
	(0.816)	(0.861)	(0.778)	(0.828)
ROA	10.843	10.280	11.149	10.586
	(12.206)	(12.104)	(12.248)	(12.111)
Operating cash flow to debt ratio	−1.492	−1.521	−1.953	−1.984
	(1.953)	(1.903)	(1.911)	(1.867)
Capacity slack	0.266	0.504	0.337	0.575
	(0.695)	(0.780)	(0.724)	(0.815)
Firm size	0.924	1.002	0.993	1.071
	(1.204)	(1.157)	(1.173)	(1.128)
Firm age	−0.055	−0.027	−0.044	−0.016
	(0.075)	(0.075)	(0.073)	(0.072)
Announcements in same city	−0.029	−0.025	−0.033	−0.029
	(0.046)	(0.045)	(0.046)	(0.045)
Announcements in same industry	−0.063	−0.065	−0.076	−0.078
	(0.051)	(0.047)	(0.058)	(0.053)
State	2.199	1.803	2.212	1.815
	(2.060)	(2.134)	(2.039)	(2.104)
IMR	8.197	8.302	8.451	8.557
	(6.175)	(5.850)	(6.069)	(5.754)
Intercept	−21.174	−23.138	−22.330	−24.298
	(21.586)	(20.104)	(21.176)	(19.716)
Province dummy	YES	YES	YES	YES
R2 (%)	11.9	15.1	12.8	15.9
Number of observations	244	244	244	244

6.3. Considering the systemically distorted stock market returns during the pandemic

To consider the systemically distorted stock market returns during the pandemic, following Hendricks et al., (2020), for each of the 5 days in our event period (Day − 2 to Day + 2), we replaced a single day return by randomly drawing the 200 daily returns of the CSI 300 Index from Day − 210 to Day − 11. We then estimated the abnormal returns for each of the 244 firms in our sample by using these randomly drawn market returns and the market model with alphas and betas estimates based on previous days (−210, −11). We repeated this process 1,000 times to obtain 1,000 abnormal returns for every event day for each firm in our sample. We averaged the abnormal returns of each firm from the 1,000 trials. The average abnormal return was used to rerun our models, and the results are presented in Table 8 . The results are consistent with our main findings.

Table 8.

Heckman analysis (robustness check 3: considering the systemically distorted stock market returns during the pandemic).

	Cumulative abnormal stock return (market model;Hendricks et al., 2020) [Day − 2, Day + 2]
Independent variables	(1)	(2)	(3)	(4)
Early production resumption	−2.758***	−3.388***	−3.095***	−3.726***
	(0.542)	(0.398)	(0.608)	(0.467)
Early production resumption × Infection risk		−5.669***		−5.672***
		(0.674)		(0.680)
Early production resumption × Liquidity stress			0.947***	0.950***
			(0.302)	(0.315)
Infection risk	1.856	−2.977***	1.904	−2.931**
	(4.035)	(0.962)	(4.319)	(1.251)
Liquidity stress	−0.727	−0.744	−0.696	−0.713
	(1.075)	(1.102)	(1.101)	(1.127)
Inventory leanness	0.141	0.361	0.233	0.453
	(0.866)	(0.930)	(0.812)	(0.882)
ROA	19.515	18.763	19.841	19.089
	(13.676)	(13.556)	(13.768)	(13.608)
Operating cash flow to debt ratio	−3.120	−3.160	−3.612*	−3.652*
	(2.037)	(2.002)	(2.100)	(2.085)
Capacity slack	0.497	0.815	0.573	0.891
	(0.698)	(0.801)	(0.713)	(0.822)
Firm size	1.318	1.422	1.391	1.496
	(1.339)	(1.280)	(1.321)	(1.264)
Firm age	−0.131	−0.093	−0.119	−0.081
	(0.135)	(0.138)	(0.133)	(0.135)
Announcements in same city	−0.015	−0.010	−0.020	−0.015
	(0.064)	(0.064)	(0.065)	(0.064)
Announcements in same industry	−0.051	−0.054	−0.065	−0.069
	(0.087)	(0.082)	(0.093)	(0.088)
State	1.705	1.176	1.719	1.189
	(2.564)	(2.661)	(2.549)	(2.631)
IMR	10.197	10.338	10.468	10.610
	(7.013)	(6.580)	(6.935)	(6.507)
Intercept	−24.990	−27.612	−26.221	−28.847
	(24.007)	(21.986)	(23.860)	(21.832)
Province dummy	YES	YES	YES	YES
R2 (%)	17.9	21.7	18.6	22.4
Number of observations	244	244	244	244

6.4. Considering province and industry variations

Last, despite there is no single province accounts for more than 20% of the sample. This distribution demonstrates that the clustering issue at the province level is not severe. To further account for variations across provinces and industries, we have adopted cluster-robust standard errors in our analysis, clustering them according to both province and industry code. The results reported in Table 9 remain consistent with our main findings, even after incorporating these adjustments.

Table 9.

Heckman analysis (robustness check 4: considering province and industry variations).

	Cumulative abnormal stock return (Fama–French four-factors model) [Day − 2, Day + 2]
Independent variables	(1)	(2)	(3)	(4)
Early production resumption	−1.026***	−1.397***	−1.583**	−1.955***
	(0.370)	(0.377)	(0.641)	(0.494)
Early production resumption × Infection risk		−3.342***		−3.347***
		(0.807)		(1.184)
Early production resumption × Liquidity stress			1.565***	1.567***
			(0.564)	(0.571)
Infection risk	0.548	−2.302	0.628	−2.226
	(7.456)	(10.799)	(2.831)	(1.437)
Liquidity stress	−0.136	−0.145	−0.085	−0.094
	(0.279)	(0.352)	(0.345)	(0.422)
Inventory leanness	0.212	0.342	0.363	0.493
	(0.619)	(0.548)	(0.543)	(0.438)
ROA	11.555	11.112	12.094	11.651
	(20.181)	(20.646)	(20.014)	(20.434)
Operating cash flow to debt ratio	−1.489	−1.512	−2.300	−2.324
	(2.439)	(2.394)	(2.163)	(2.044)
Capacity slack	−0.060	0.127	0.065	0.253
	(0.556)	(0.550)	(0.496)	(0.517)
Firm size	0.892**	0.954***	1.014***	1.075***
	(0.398)	(0.349)	(0.348)	(0.292)
Firm age	0.008	0.030	0.028	0.050
	(0.064)	(0.064)	(0.087)	(0.083)
Announcements in same city	−0.028	−0.025	−0.036	−0.033
	(0.084)	(0.084)	(0.089)	(0.088)
Announcements in same industry	−0.077	−0.079	−0.101	−0.102
	(0.090)	(0.094)	(0.089)	(0.094)
State	2.116	1.804	2.138	1.826
	(1.370)	(1.379)	(1.395)	(1.324)
IMR	7.885*	7.969*	8.332*	8.416*
	(4.330)	(4.165)	(4.470)	(4.384)
Intercept	−21.937*	–23.482**	–23.971**	−25.521**
	(11.477)	(10.017)	(12.181)	(10.563)
Province dummy	YES	YES	YES	YES
R2 (%)	9.3	11.3	12.1	14.1
Number of observations	244	244	244	244

7. Discussion, conclusion, and limitations

This study has examined the investor reaction to the decisions of early production resumption amid the early part of the COVID-19 crisis. Our event study reveals a positive reaction to 244 production resumption announcements in the Chinese manufacturing sector. However, the reaction is less positive among those who resumed production early, which demonstrates the risk concerns in the early resumption. These negative stock price changes were amplified by the number of locally confirmed cases of COVID-19 but attenuated by liquidity pressures for Chinese manufacturers. We have discussed the theoretical contributions to the OM literature in terms of supply chain disruptions and implications for manufacturers and policymakers.

7.1. Theoretical contributions

This study first contributes to the literature of supply chain disruption management. Conventional wisdom considers that a quick disruption recovery is a good disruption recovery (Chen et al., 2019). However, this study challenges this assumption by proposing that risk can emerge if a firm returns to normal too early, especially in the context of an evolving crisis such as COVID-19. This view aligns with the simulation studies concluding that the timing of the closure and opening of facilities can be a major factor in determining overall supply chain performance during a pandemic (Ivanov, 2020). This study captures the risk derived from quick recovery through investors’ stock-selling behavior. Specifically, the risk led to a 1.380% decline in stock prices for those Chinese manufacturers who resumed early, as opposed to those who resumed at an average pace.

This research can also contribute to the area of crisis management in OM literature. Differ from the current literature focus on the strategic response to the crisis (e.g., Bundy et al., 2017, Raithel and Hock, 2021), this research focuses on the tactical responses that received relatively less attention (Klöckner et al., 2023). We provided understanding to the short-term actions taken to respond to crisis. Specifically, we focus on the timing of production resumption during an evolving crisis (i.e., COVID-19). Our results illustrate the concerns about the production resumption soon after a disruption amidst a dynamic, complex and uncertainty environment.

These risks also illustrate the need to reconsider risk management for supply chain disruption. This is in line with the call for risk management research amidst the COVID-19 crisis (Choi, 2021). Conventionally, a risk management plan (identification, assessment, mitigation, and control) should be prepared before a disruptive event (Ho et al., 2015). The literature also proposed the need to execute recovery measures when disruption occurs (Ivanov et al., 2017). However, if recovery decisions themselves can generate risk, risk management is also needed at this stage. Therefore, we propose a revision to the current risk and recovery management model for disruption, as shown in Fig. 1 . First, the speed of recovery should not be the most predominant indicator of the recovery performance. The risks after recovery should be integrated as an equally important indicator. Second, to minimize the risks after recovery, the risk management process should be executed together with the recovery measures. This process can turn “unknowns” in the “new normal” (after disruption events) into “knowns” for managers to control, which can reduce the likelihood and impacts of future disruptions. Future research is encouraged to explore the disruption recovery risks in other longer-term crises, such as unrest and wars.

Fig. 1.

Revised disruption recovery model.

However, we understand that, in disruption, managers’ attention is largely focused on the resumption of cash flow because of the pressure for firm survival (Chen et al., 2019). This can be demonstrated by the mostly positive investor reaction (1.615%) to the announcement of production resumption by manufacturers. In addition, the results show that quick production resumptions were preferred when manufacturers had higher liquidity pressures.

The attention to short-term cash flow and operation continuity may lead managers to downplay the recovery risks and resume production while underprepared. This study explores the factors that can shift such attention from short-term cash flow to recovery risks. Our results show that delayed production resumptions were preferred with more newly confirmed local COVID-19 cases. This illustrates that the risks should be visible and prominent so that the threat of infection (during recovery) can receive enough attention. We propose that risk identification and assessment for disruption recovery can make this information viable to managers and other stakeholders (Larson & Gray, 2013). Future research may also investigate alternative factors that may enhance managers’ attention to risk management in disruption recovery, such as culture (Pidgeon, 1991), the resources committed (Power, 2004), and capital structure (Froot & Stein, 1998).

7.2. Practical implications

This study offers valuable insights for factory managers tasked with resuming production following a disruption caused by evolving crises. Traditional thinking suggests that investors would prefer production to recommence as quickly as possible. However, our findings indicate that a hasty recovery may also generate concerns among investors. To alleviate these concerns, managers must prioritize the occupational health and safety of workers during production resumption in times of evolving crises. This can be achieved by developing and implementing a comprehensive protocol specifically designed to safeguard worker safety during an evolving crisis such as COVID-19. For example, introducing staggered work shifts and providing personal protective equipment to all employees.

Furthermore, managers should enhance communication with shareholders to assuage their apprehensions regarding production resumption. This entails keeping investors and employees well-informed about resumption plans and the safety measures being implemented. Clear and transparent communication, such as regular updates through email newsletters or dedicated web portals, can help reduce anxiety, foster trust, and elicit a more favorable response from investors.

Lastly, it is crucial for firms to continuously monitor the status of the evolving crisis when planning production resumption during such uncertain times. Our findings reveal that the severity of the crisis can create additional complexities for managers, which may consequently heighten investor concerns. As a result, a robust risk management process is essential in this context to mitigate risks during resumption and alleviate investor apprehensions. For instance, managers could establish a dedicated task force to assess the evolving crisis, explore alternative production strategies, and develop contingency plans to maintain operational continuity.

Government is important in assisting and guiding factories to resume with order. We observed that firms loaded with substantial debt tended to resume production in a rush. Thus, the government should assist these firms and provide bridging loans for them to mitigate their liquidity concerns. Specifically, our results estimate that firms with book debt/market value ratios of higher than 243.08% would create a sentiment among investors favoring early resumption. Thus, governments should help these firms to reduce their debts to below this level and avoid the rushed reopening of factories under stockholder pressure. In addition, common wisdom considers that the stock market should favor a quick economic reboot after a lockdown, whereas our results suggest the opposite. An overly hasty reboot may cause negative sentiment and worry among investors about the future risk of lockdown. These implications may also be generalized from COVID-19 to other types of evolving crises that could occur in the future.

7.3. Limitations and future research directions

This study has several limitations that future research could address First, our sample comprises production resumption decisions from the early COVID-19 outbreak in China when knowledge of the virus was limited, which may have affected risk evaluation and decision-making processes. Infection risks and impacts may be perceived differently as the understanding of the virus and precautionary measures evolve. Future research could explore how improved knowledge of the virus and the development of effective prevention strategies influence production resumption decisions and investor reactions over time.

Second, our sample focuses exclusively on listed firms with sufficient resources to survive the COVID-19 crisis. Small and medium-sized enterprises (SMEs) generally have fewer resources and may face greater liquidity pressures. Future research should investigate how SMEs navigate production resumption decisions during evolving crises and whether their decision-making processes and investor reactions differ from those of larger firms. This could also include examining the role of government support and industry assistance in enabling SMEs to build more resilient supply chains.

Third, we used COVID-19 as an example of an evolving crisis in this research. However, future research could extend the empirical analysis to other evolving crises, such as climate crises or geopolitical conflicts (Fan et al., 2022a, Fan and Xiao, 2023), to better understand how firms respond to different types of disruptions and the implications for supply chain resilience. This would provide a more comprehensive understanding of the generalizability of our findings and offer insights into how firms can adapt their operations and supply chain management practices to various crisis scenarios.

Furthermore, future studies could investigate the role of digital technologies, such as artificial intelligence, big data analytics, and blockchain, in enhancing supply chain visibility and facilitating more effective decision-making during evolving crises. This research could provide valuable insights into how firms can leverage emerging technologies to build more resilient supply chains and mitigate the risks associated with disruptions.

Lastly, research could explore the long-term effects of production resumption decisions during evolving crises on firm performance and supply chain resilience. By examining how different resumption strategies impact a firm's operational efficiency, financial performance, and overall competitiveness, researchers could develop more informed recommendations for managing disruptions and enhancing supply chain resilience in the face of evolving crises.

CRediT authorship contribution statement

Di Fan: Conceptualization, Methodology, Resources, Writing – original draft, Writing – review & editing, Supervision, Funding acquisition. Yongjia Lin: Methodology, Validation, Formal analysis, Investigation, Writing – original draft, Writing – review & editing, Supervision, Funding acquisition. Maggie X. Fu: Writing – original draft, Supervision, Funding acquisition. Andy C.L. Yeung: Resources, Supervision. Xuanyi Shi: Formal analysis, Investigation.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A.

Table A1.

Sample selection.

Steps	Description	Discard	Announcement
1	Select announcements with “resumption of work” and “resumption of production” as keywords from media report databases of CSMAR from January 23–March 28, 2020, for all listed firms on the Shenzhen/Shanghai stock exchange		299
2	Minus: non-manufacturing firms	7	292
3	Minus: confounding events	2	290
4	Minus: non-trading period (Jan. 24, 2020 ∼ Feb. 2, 2020)	46	244

Fig. A1.

Timeline of early COVID-19 outbreak in China.

Appendix B.

Table B1.

Province distribution of firm announcements.

Province	Freq.	Percent
Shanghai	9	3.69
Inner Mongolia	5	2.05
Beijing	11	4.51
Sichuan	11	4.51
Tianjin	4	1.64
Ningxia	1	0.41
Anhui	5	2.05
Shandong	17	6.97
Shanxi	1	0.41
Guangdong	47	19.26
Xinjiang	6	2.46
Jiangsu	20	8.2
Jiangxi	2	0.82
Hebei	2	0.82
Henan	4	1.64
Zhejiang	48	19.67
Hainan	2	0.82
Hubei	10	4.1
Hunan	8	3.28
Gansu	5	2.05
Fujian	14	5.74
Guizhou	4	1.64
Liaoning	3	1.23
Chongqing	4	1.64
Shanxi	1	0.41
Total	244	100

Fig. B1.

The effect of the manufacturer’s early production resumption decision on the manufacturer’s investor reactions (H2).

Appendix C. Selection model of Heckman two-stage analysis

The first stage of the Heckman process involves a probit regression on the choice of the announcement. To construct the data set, we collected data for all the Chinese-listed manufacturing firms from the CSMAR database. This binary variable, production resumption, was used as the dependent variable. Specifically, “1″ was coded for the firms with an announcement from January 23–March 28, 2020 (290 firms) and “0” otherwise (2,070 firms).

We then selected independent variables based on their influence on the choice of the announcement. First, the disclosure of production resumption may be pressured by announcements from industry competitors. We thus included industry production resumption, measured by the ratio of the number of announced to non-announced firms in each firm’s two-digit industry classification code (developed by the China Securities Regulatory Committee). Ali et al. (2014) argued that corporate disclosures in more concentrated industries may provide more reliable information than similar disclosures in less concentrated industries. Rivals may use this information to adjust their strategies to the detriment of the disclosing firm. Consequently, industry concentration is negatively related to corporate disclosure. We included industry concentration, measured by the sale-based Herfindahl–Hirschman index (HHI) to control for the competition from potential entrants. The HHI is calculated as the sum of the squared market share based on the industry. A higher value means competition from potential entrants is less intense. These two industry factors are exogenous; thus, the model can fulfill the exclusion restriction of the selection model development for Heckman’s two-stage analysis.

Inventory slacks were measured by the number of inventory days and standardized according to the industry mean and standard deviation (the two-digit industry code) in the same year (Hendricks et al., 2009). Firms with more inventory slack may have less motivation to announce because of uncertainty about investors’ opinions concerning production resumption. Firm size, as the logarithm of the number of firm employees and performance, measured by return on assets (ROA), was included because larger and more profitable firms experience greater demand for, and benefits from, information acquisition (Boone & White, 2015). Leverage is measured by the ratio of the book value of debt to the book value of assets and is included because highly leveraged firms tend to have more variable earnings, which may lead firms to refrain from announcing to avoid investors’ attention. Tobin’s Q, as a proxy for growth options (Bamber & Cheon, 1998), is measured by the ratio of the sum of the book value of debt and the market value of equity to the book value of assets. Investors may pay more attention to firms with greater growth options; thus, these firms have more pressure to be transparent. Appendix Table C presents the results of the probit analysis. The model provided a satisfactory estimate of the likelihood of announcement initiatives for production resumption (Chi² = 45,212.61, p < 0.01).

Table C1.

First-step probit analysis (n = 2361).

Dependent variable: Production resumption (“1″ = yes; “0” = no)
	Coef.	Robust SE	p
Industry production resumption	3.290	0.548	0.000
Industry concentration	−0.583	0.322	0.070
Inventory slack	−0.068	0.040	0.089
ROA	0.705	0.422	0.093
Firm size	0.178	0.048	0.000
Leverage	0.141	0.204	0.475
Tobin’s Q	−0.001	0.017	0.976
Intercept	−3.527	0.345	0.000
Province dummy			Included
Chi2	45381.28		0.000

Fig. C1.

Interaction slopes for manufacturer’s early production resumption and infection risk (H3).

Appendix D.

Table D1.

VIF.

Variable	VIF
Early production resumption	1.30
Infection risk	1.10
Liquidity stress	1.74
Inventory leanness	1.34
ROA	1.80
Operating cash flow to debt ratio	1.66
Capacity slack	1.16
Firm size	2.30
Firm age	1.07
Announcements in same city	1.23
Announcements in same industry	1.06
State	1.12
IMR	2.29
Mean VIF	1.47

Fig. D1.

Interaction slopes for manufacturer’s early production resumption and liquidity stress (H4).

Appendix E.

Table E1.

Variable definitions.

Variables	Description	Initial Data Source	Reference
Early production resumption	Reversed natural log-transformed the number of days between the announcement date and the lockdown date of Wuhan	CSMAR	–
Infection risks	Newly confirmed cases in each firm’s location	CSMAR	Gu et al. (2020)
Liquidity stress	The ratio of the book value of debt to the market value of equity, standardized according to the industry mean and standard deviation (two-digit industry code)	CSMAR	Bhandari, 1988, Fama and French, 1993, Schmidt et al., 2020
Inventory leanness	The number of inventory days (scale reversed), standardized according to the industry mean and standard deviation (two-digit industry code)	CSMAR	Hendricks et al., 2009, Wiengarten et al., 2017
ROA	Return on assets	CSMAR	Wiengarten et al. (2017)
Operating cash flow to debt ratio	The ratio of operating cash flow to total liabilities	CSMAR	Capon et al. (1990)
Capacity slack	The ratio of fixed assets to sales, standardized according to the industry mean and standard deviation (two-digit industry code)	CSMAR	Wiengarten et al. (2017)
Firm size	The logarithm of the number of firm employees	CSMAR	Wiengarten et al. (2017);
Firm age	The logarithm of the number of firm age	CSMAR	Fan et al. (2020)
Announcements in same city	Number of announcements in the same city that were made prior to the focal firm’s announcement	CSMAR	–
Announcements in same industry	Number of announcements in the same industry that were made prior to the focal firm’s announcement	CSMAR	–
State	A state-owned enterprise dummy SOE, taking a value of 1 if the firm is controlled by the state and 0 otherwise	CSMAR	Cui and Jiang, 2012, Zhou et al., 2017