Fears for COVID-19: The crash risk of stock market
11 / Fears for COVID-19: The crash risk of stock market
Zhifeng Liu , Toan Luu Duc Huynh , Peng-Fei
Dai (1) Management School, Hainan University, Haikou, China (2)
College of Management and Economics, Tianjin University, Tianjin, China (3)
Institute of Research and Development, Duy Tan University, Danang 550000, Vietnam (4)
Faculty of Business Administration, Duy Tan University, Danang 550000, Vietnam (5)
Chair of Behavioral Finance, WHU – Otto Beisheim School of Management, Vallendar, Germany (6)
Supply Chain and Logistics Optimization Research Center, University of Windsor, Windsor, Canada
ABTRACT
This paper investigates the impact of COVID-19 epidemic on the Chinese stock market crash risk.We first estimate conditional skewness of the return distribution from the GARCH-S model as theproxy of the equity market crash risk for the Shanghai Exchange Stock Market. Then, weconstruct a fear index for COVID-19 using the data from Baidu Index. Our findings show that theconditional skewness reacts negatively to daily growth in total confirmed cases, indicating that theepidemic increases the crash risk of stock market. Furthermore, we find that the fear sentimentalso exacerbates the crash risk. In particular, the fear sentiment plays a significant role in theimpact of COVID-19 on the crash risk. When the fear sentiment among people is high, the stockmarket crash risk is affected by the epidemic more seriously. Evidence from the daily deaths andglobal cases shows the robustness.
Keywords : COVID-19; Fear sentiment; Investor sentiment; Stock market crash risk; Skewness.
JEL Classification:
G10; G32.
Acknowledgement:
This research was supported by the National Natural Science Foundation of China (71861008), theNatural Science Foundation of Hainan Province (718QN221, 2019RC151), and the ScientificResearch Foundation of Hainan University (kyqd(sk)1809, kyqd1634). /
1. Introduction
During the COVID-19 epidemic, there was a significant decline in stock marketreturns, which has attracted the attention of many scholars recently. Previous studieshave mainly examined the impacts of COVID-19 on the return or volatility of thestock market, and some conclusions have been supported by empirical evidence(Baker et al. (2020), Al-Awadhi, Al-Saifi, Al-Awadhi and Alhamadi (2020), Phan andNarayan (2020), Ashraf (2020), Kartal et al. (2020), Ramelli and Wagner (2020),Zhang, Hu and Ji (2020a), Saeed and Ridoy (2020), Sharif, Aloui and Yarovaya(2020)). Although the COVID-19 broken out earlier in the regional area of China’sHubei Province, this negative effects have applied the unprecedented pressures to theglobal financial markets. In which, due to the comovements of global stock markets(Wen, Yang and Zhou, 2019), the global equity plummeted, which was followed by aspike of market volatility. In the same vein, Baker et al. (2020) draw the conclusionthat the level of market volatility in this time (March 2020) can be equivalent or evensurpassing in comparison with the previous crisis, namely October 1987 (BlackMonday) and December 2008 (Global financial crisis) and, before that, in late 1929(Great Crash) and the early 1930s (Great Depression). Concomitantly, Schell et al.(2020) also emphasized that this time is indeed different, implying that only theCOVID-19 exhibits the negative returns with the Public Health Risk Emergency ofInternational Concern (PHEIC) announcements. Therefore, the adverse effects ofCOVID-19 pandemic on the stock markets are still being studied from differentperspectives. Motivated by these extant literature, this paper will focus on anotherpivotal downside risk in the stock market during the epidemic: stock market crash riskand investors sentiment in China. Notwithstanding the current literature between theinfectious disease outbreaks and stock market performance, our study does not onlyquantify the stock market crash risk but also investigate the role of investors’sentiments by using the data searching tools on the Baidu and the number of infectedcases (deaths) in terms of driving the potential risks. In doing so, our study is one of / the pioneers to shed a new light on how the coronavirus proxies and investors’behaviors could predict the equity market crash risk on the onset of the pandemic.In fact, during the coronavirus outbreak, the stock market has suffered a severeshock, and the probability of the stock market crash risk is significantly higher thanusual. In the first three months of 2020 (59 trading days), there were six days with asingle-day crash of 2% or more. While in the past three years (730 trading days), thetotal number of days to experience such a decline was only 21. Some scholars areconcerned about the stock market crash during the epidemic. For example, Mazur,Dang and Vega (2020) discussed COVID-19 and the stock market crash, but theydefined the crash in terms of extreme returns and volatility.The crash risk, measured by the conditional skewness in our paper, can capturethe negative asymmetry risk and the extreme downside risk in the stock market (Chen,Hong and Stein (2001)). Academics have analyzed stock market crash risk fromdifferent perspectives. For example, Chen, Hong and Stein (2001) undertake anempirical investigation to forecast crash risk (skewness), both at firm and at wholemarket levels. And concerning the crash risk at firm level, Kim, Li and Zhang (2011a),Kim, Li and Zhang (2011b) have done a series of representative work.This paper investigates the effects of COVID-19 on the stock market crash risk.We first use the GARCH-S model to estimate the daily time-varying skewness ofstock returns, and use it as a measure of stock market crash risk. In the subsequentempirical analysis, we not only analyzed the impact of the severity of the epidemic(measured by the number of daily confirmed cases) on the crash risk, but alsodiscussed the impact of people's panic during the epidemic. We found that the moresevere the epidemic and the more people panic about the epidemic, the greater thecrash risk of the stock market. Furthermore, we further studied the interactionbetween the severity of the epidemic and fear sentiment. We found that the fearsentiment has contributed to the negative effect of the epidemic on the stock marketcrash risk. / This paper contributes to the extant literature by several ways. First, asmentioned above, we focus our research on the risk of stock market crashes, whichallows us to pay more attention to those asymmetric negative and extreme risks.Second, we create a fear sentiment by using the Baidu index towards the COVID-19pandemic to examine if panic for the epidemic will correlate with the market crash.Finally, we investigate the role of fear sentiment in the impact of COVID-19 on thestock market crash risk. More importantly, our findings carry some policyimplications. One of the most important policies is to alleviate the investors’ panic tomitigate the equity market crash risk. The second preventive measure could beconsidered as the way to communicate about the number of infected cases as well asdeaths. To recapitulate, our findings might provide the policymakers with the deepunderstanding of how to response to and cope with the investors’ pessimism about theequity markets in a timely and comprehensive manner during such a financialdownturn.The rest of this paper is organized as follows: Section 2 acknowledges thecurrent literature while section 3 describes the data and methodology. Following this,section 4 presents the empirical results of the impacts of COVID-19 on stock marketcrash risk. Section 5 concludes.
2. Brief literature review
It is essential to construct the solid and sound theoretical framework how theCOVID-19 pandemic adversely influence the financial markets. Goodell (2020)compares that the markets are likely to react with the pandemic in the same way withthe other forms of disasters such as catastrophic disasters (Gao et al., 2020) orterrorism (Wang and Young, 2020). There is a common trait that investors’ riskpreferences or their mood towards these events might vary considerably, which meansan increase in fear-induced feelings. While the previous events were occurred inspecific region or area with the partial disruption, the existence of COVID-19pandemic locked the travelling status as well as economic transactions in the global / scope. Hence, the effects of pandemic on the total economy will not only enormouslyinfluence domestic demands but also completely limit the supply, which is expectedto dip the future cash flows of the firms. This phenomenon first manifests itself as thepublic mood about the future pessimistic feelings.The COVID-19 pandemic is considered as the unique public health crisis event interms of its global scope since the influenza pandemic in 1918. Thus, there are manyunknown perspectives which need to be examined. One of the biggest concerns is thefinancial crashes. While Mazur et al. (2020) claimed that the financial market crash inMarch 2020 was triggered by the government’s reactions. Interestingly, the negativeeffects are more pronounced in some specific industries such as crude petroleum, realestate, entertainment, and hospitality sectors. This paper also confirms what Mishkinand White (2002) found. The equity market crash could results in a drop by 20-25%in the main index during the previous crisis due to the sequence of panic selling. Thus,our motivation is to examine the determinants including the investors’ sentiment andthe pandemic status could predict the equity market crash risk while Giglio et al.(2020), Wen et al. (2019a), Wen et al. (2019b) and Zhang et al. (2020) reveal thatinvestors’ expectations with the short run might correlate with the stock market crashrisk. It is noticeable that the previous studies (Giglio et al., 2019; Giglio et al., 2020)also confirm that the probability of equity market crash before the crisis is lowerbecause the investors tend to be more optimistic about stock market returns.Notwithstanding the evident findings, the further investigation of COVID-19 is stillpromising since we take no stance on whether the likelihood of market crash riskwould significantly change in the two sub-period, including before- and after- thepandemic. Given the foregoing discussion and argument, we accordingly hypothesizethe following: H : The Chinese equity market crash risk in the COVID-19 pandemic is higher thanin the previous period. To examine the first hypothesis, we divide our samples into two sub-periods and / employ the statistical test as well as the illustration. The aforementioned literature(such as Giglio et al., 2019; Giglio et al., 2020; Gabaix, 2012; Wachter, 2013) wouldbe the sound and solid framework for our constructions of the first hypothesis. H : There is no relationship between investors’ sentiment and stock market crash onthe onset of COVID-19 pandemic. Although there are a mounting literature examining how investors overreact orunderreact in the COVID-19 pandemic (see more at Schell et al., 2020; Aslam et al.,2020; Yarovaya et al., 2020), it is still unanswered what drives the Chinese stockmarket crash risk. It is marginally relevant to consider that the combination ofeconomic uncertainty and behavioral factors positively contributes to financial assetcrash risk (for example, Bitcoin (Kalyvas et al., 2020), Chinese stock market (Jin etal., 2019; Luo and Zhang, 2019). Noticeably, the concern how the aforementionedfactors drive the stock market during the diseases period is still open. In doing so, weuse the two proxies such as the fear index for COVID-19 aggregating the data fromBaidu Index and the actual figures of pandemic situation to predict the changes instock market crash risk index, constructed by employing the GARCH-S. Thesubstitution of different proxies would be our alternative approach to check whetherour findings and results are robust or not.Overall, the existing literature still has a gap for us to fulfill with two main folds. First,answering the research question ‘
How is the equity market crash risk before and afterthe COVID-19 pandemic? ’ would benefit to not only the practitioners to be cautiousabout the extreme shocks in the market but also the understandings of empiricalevidence for academia. The second question needs to be answered is “
What drives theChinese stock market crash risk? Do the investors’ fears and/or the current situationof coronavirus statistical numbers matter the Chinese stock market crash risk? ’.Based on the above arguments, our corresponding questions are closely related to the / line of research on the financial markets’ reactions under the COVID-19 pandemic.However, only few works so far have examined the effect of investors’ emotion,especially fear sentiment, on systematic risk in emerging economies on the onset ofcoronavirus outbreaks. The majority of extant studies only have kept their eyes on theadvanced markets; for example, the United States, or European markets while only ascarcity of papers address the phenomenon in emerging economies, which also sufferfrom the severity of downside in the COVID-19 outbreaks. Hence, our study wouldshed a new light on how the level of Chinese stock market crash risk change in thisdifficult time.
3. Data and methodology3.1. COVID-19 related variables
The proxy we used in this paper to measure the severity of the COVID-19epidemic is the logarithmic growth rate of daily confirmed cases ( rCases ). We alsoconstruct an alternative variable using the logarithmic growth rate of daily deaths( rDths ) to run the robustness check. The data is retrieved from the CSMAR database.We create a COVID-19 induced fear sentiment index ( fearSent ) utilizing the datafrom the Baidu index following the idea of Da, Engelberg and Gao (2011). If thesearching volume of COVID-19 related keywords is high, it means people are fear oreven panic about the epidemic. Specifically, we define the fear index as the log ofsearching volume plus one. Figure 1 displays the trend of the daily confirmed casesand fear sentiment. Moreover, we set a dummy variable
D_fear for fear index: if thesearching volume is greater than the median of the 2020 sample, the value of thisdummy is one, and zero otherwise. / Figure 1. Daily confirmed cases and fear sentiment3.2. Measuring stock market crash risk
Our market returns were collected from the value-weighted market return ofShanghai A shares, used frequently in the extant literature (see more at Ashraf (2020),Al-Awadhi et al. (2020)). To measure the crash risk of stock market, we follow thework of Chen, Hong and Stein (2001), which associate it solely with the conditionalskewness of the market return. They calculate the six-month horizon skewness fromthe daily returns, however, we use the GARCH-S (GARCH with skewness) model toestimate the daily skewness.
21 2 120 1 1 2 120 1 1 2 1 ; ~ 0,; ~ 0,1 ; ~ 0, t t tt t t t t t tt t tt t t r h I hh hs s (1)Where t r is the value-weighted market return of Shanghai A shares, t is theresidual and t is the standardized residual. t I denotes the information set at the / period t. t h is the conditional heteroscedasticity with a classical GARCH(1,1)structure. t s represents the conditional skewness process, and we specify it as bothautoregressive and dependent on lagged return shocks. To estimate the GARCH-Smodel, following Leon, Rubio and Serna (2005), we use a Gram-Charlier seriesexpansion and truncate at the third moment. And it should be noted that, due to thehigh non-linearity of the likelihood function, we use the starting values of parametersestimated from the simple GARCH (1,1) model.The market data we used is also from the CSMAR database and the sampleperiod in our paper is from January 1, 2017 to March 31, 2020. Table 1 provides themarket returns descriptive statistics of the whole sample and sub-samples, includingthe unconditional skewnesses. We can see that the skewness of the whole sample is-0.71, while during the COVID-19 epidemic (January 2020 – March 2020), this valueis -1.47, compared to -0.30 in the 2017-2019 sample period. Table 1. Summary of descriptive statistics for market returns
Sub-sample Obs. Mean Min Max Std. Dev. Skewness
Whole sample 789 0.00009 -0.075 0.055 0.011 -0.712Jan 2017 – Dec 2019 731 0.0002 -0.053 0.055 0.010 -0.302Jan 2020 – Mar 2020 58 -0.0016 -0.075 0.031 0.017 -1.469
Notes:
We divide our sample into two sub-samples consisting of the prior- and post- the COVID-19 pandemic.
When looking at the statistical evidence, we do not reject the mean differencebetween two sub-samples including before (0.0002) – after (-0.0016) in Table 1 (t-stat= 1.25, = 0.211). It implies that there is no difference in market return when theCOVID-19 pandemic happened. However, with regard to the skewness index,representing the market crash risk, we observe the significant difference in mean / between two periods. To be more precise, the level of market crash in 2020 issignificantly higher than the previous period (t-stat = 2.50, = 0.01). Therefore, wedo not reject the first hypothesis, implying the higher extreme volatility in the Chineseequity market when having the COVID-19 pandemic.Table 2 further reports the estimation results of GARCH-S model. As expected,the results indicate a significant presence of conditional skewness. Specifically, thecoefficient of lagged skewness is positive and significant (0.148 with a t-statistic58.079), indicating that the skewness is persistent and high conditional skewness isfollowed by high conditional skewness. We also find the coefficient of the shock toskewness is positive and significant (0.036 with a t-statistic 15.373), which is similarwith that in the variance case. In general, the majority of coefficients are significant,implying the appropriateness of our models when using GARCH-S to estimate theskewness of the market return. Table 2. Estimation results of GARCH-S modelParameter Value Parameter Value μ β α β α β α / Notes : ***, **, * represent statistical significance at 1%, 5%, and 10% levels, respectively. Thet-statistics are presented in the brackets.
Figure 2 presents the trajectory of conditional skewness, from which we canvisually see that the skewness is time-varying and clustering. In particular, there are alarge number of cases where the skewness is negative, indicating that the crash risk atthese points is high. The largest negative value of skewness occurred during theCOVID-19 outbreak in our sample period, that is, on February 4, 2020, and it reached-0.76.
Figure 2. The conditional skewness3.3. Model specifications
We conduct a simple time series model to examine the relationship betweenCOVID-19 outbreak and stock market crash risk. Our dependent variable is the crashrisk, i.e., conditional skewness calculated from the estimation results of GARCH-Smodel. Due to the persistence of skewness, we add the lagged skewness term in ourmodel. The benchmark specification of the regression model is specified as: + t t t t Skew c Skew rCases (Eq. 2)where t Skew is the conditional skewness derived from the GARCH-S model, and / rCases is the logarithmic growth rate of daily confirmed cases. In addition, c isconstant term, and are the coefficients of the one-period lagged term and thelogarithmic growth rate of infected cases, respectively. Finally, is the error term inthe estimation.In our paper, we also consider whether the COVID-19 induced fear sentimentindex ( fearSent ) affects the crash risk. And thus, we estimate the following model: + t t t t Skew c Skew fearSent (Eq. 3)In which, Equation 3 also has similar denotations, which are presented above.We only substitute the rCase by fearSent to consider how investors’ sentiment couldpredict the stock market crash risk. In order to further investigate the interaction effectbetween the daily confirmed cases and the fear sentiment, we also add the interactionterm in our model. Similarly, we choose that , represent the coefficients for fearsentiment and interaction term while the other components in equation 4 are the sameas the previous equations. Our model specification is as follows: + + + t t t t t t t Skew c Skew rCase fearSent rCase fearSent (Eq. 4)We also employ the Granger causality test to detect the causal relationshipbetween the crash risk of the stock market and the fear sentiment. The model forGranger causality test is specified as follows: titpi pi iitit fearSentSkewcSkew (Eq. 5) tjtpi pj iitit fearSentSkewcfaerSent
21 12 (Eq. 6)where p is the largest lag order which is determined through the VAR model andthe Bayesian information criterion. The hypotheses tests for Granger causality arepresented in Table 6. Finally, in our robustness section, we use the alternative proxy / for the growth rate of confirmed cases as well as death cases to predict the equitymarket crash risk. In addition, due to the integration of the financial markets, we alsorun the further estimations to predict the equity market crash risk with the number ofinfected cases and deaths in the global scope. Our justification is that the Chineseinvestors do not only react with the local information but also the worldwide news,which might drive their behaviors to the market crash risk. tttt sGlobalCaserSkewcSkew (Eq. 7) tt 1tttt fearSentsGlobalCaser fearSentsGlobalCaserSkewcSkew )( )( (Eq. 8)The equation (2) and (4) are changed into equation (7) and (8) respectively. By doingthat, we could examine how our determinants are robust. Our findings will beillustrated and summarized in the following sections.
4. Findings and results4.1. COVID-19 and stock market crash risk
The estimation results of model (2) are reported in Table 3. In column (1), wecan see that the coefficient of t rCases is negative and significant. It is consistentwith our exception that the COVID-19 outbreak has a negative impact on stockmarket crash risk. This result also reflects the reality well. With the rapid spread of theepidemic, the value of listed companies has generally been affected, and the stockmarket has entered a period of obvious downturn, accompanied by large declines oreven crashes. Surprisingly, there is no predictive power of the other lagged termsincluding rCases (t-2) and rCases (t-3) for the changes in market crash risk. Thus, thisfinding emphasizes the role of information, particularly the number of cases, in theprevious trading day on the market shocks. In terms of explanatory power, theR-squared in these markets are around 7 per cent indicating that the lagged variables / of logarithmic growth rate of daily confirmed cases can be explained by changes inmarket skewness, representing the stock market crash risk. Table 3. The effects of COVID-19 on stock market crash riskVariables (1) (2) (3)
Intercept -0.001(-0.45) -0.001(-0.45) -0.001(-0.45)Skew (t-1) (t-1) -0.083***(-5.26) -0.082***(-5.17) -0.082***(-5.12)rCases (t-2) -0.005(-0.34) -0.005(-0.34)rCases (t-3) -0.001(-0.05)N 787 786 785R Notes: ***, **, * represent statistical significance at 1%, 5%, and 10% levels,respectively. The t-statistics are presented in the brackets. Table 3 summarizes theestimated results for the Equation 2.
Our findings also confirm the extant literature that the stock markets are likely tobe sensitive to the information about the growth in number of confirmed cases ascompared to the growth in number of deaths (Ashraf, 2020; Albulescu, 2020). Thus,apart from the United States market, the new infection cases reported at Chinese levelamplify the Chinese market crash risk.
We consider the role of COVID-19 induced fear sentiment in this section. The / motivation is that people's panic about the epidemic may remain at a high level,although the number of confirmed cases at this time is not very large. For example, asearly as January 20, 2020, Academician Zhong Nanshan publicly confirmed thehuman-to-human transmission of COVID-19 on TV. A few days later, on January 23,the central government of China announced the lockdown of Wuhan. Although thenumber of confirmed cases publicly disclosed at that time was still at a low level,people quickly fell into panic. That is to say, the fear sentiment may have an impacton the stock market ahead of the impact from the confirmed cases.The results of model (3) are shown in Table 4. We use the variable fearSent inthe results of the first two columns. In addition, we also use the dummy variable, d_fear , to re-estimate model (3), and the results are displayed in the last two columns.We can see that all the coefficients about fear sentiment are negative and significant,indicating that COVID-19 induced fear sentiment will cause a significant crash instock market. Table 4. The effects of COVID-19 induced fear sentiment on stock market crashriskVariables (1) (2) (3) (4)
Intercept 0.045**(2.48) 0.038**(2.13) -0.0002(-0.05) 0.0001(0.01)Skew (t-1) (t-1) -0.080***(-5.09) -0.081***(-5.15)Fear sentiment -0.005***(-2.61) -0.004**(-2.24) -0.042***(-2.61) -0.038**(-2.38)N 788 787 788 787R / Notes: ***, **, * represent statistical significance at 1%, 5%, and 10% levels, respectively. The t-statistics arepresented in the brackets. Table 3 summarizes the estimated results for the Equation 3.
While Duan et al. (2020) employed the textual analysis of 6.3 million messageson social media to conclude that the Chinese stock market is likely to overact with thegrowth sentiments, our findings are also consistent with the aforementioned study byusing the construction methods of Da, Engelberg and Gao (2011) with Baidusearching engine. Interestingly, Burgraf et al. (2020) applied the same method toindicate that the Bitcoin market significantly changes when the investors’ sentimentsfluctuate. However, one of the most novel point from our study is to examine wherethe fear sentiment stands in over the pandemic. It is noticeable that the Chinese stockmarket crash worsens when incorporating the fear sentiment. Our results are robustwhen controlling the other variables such as the lagged term of skewness (theprevious term of market crash risk), the number of infected cases. To sum up, ourfindings do reject the second hypothesis, which means that there exists therelationship between investor sentiment and Chinese market crash risk during theCOVID-19 outbreaks. Although the extant literature confirms this linkage in thenormal market condition, our study also sheds a new light on this relationship on theonset of pandemic, one unique event shaking the investors’ sentiments.
The above results show that both daily confirm cases and fear sentiment canincrease the risk of stock market crashes. In this section, we try to further explore theinner links between these impacts to the underlying mechanisms.Table 5 gives the results about the interaction effect between daily cases and fearsentiment. The coefficients of the interaction terms are negative and show theirsignificance. It means that fear sentiment will further amplify the negative impact ofconfirmed cases on stock market crash risk. In other words, because of fear, thenegative impact of COVID has been exacerbated. It shows the importance of / preventing investors from falling into panic and maintaining optimism during theepidemic. Table 5. The interaction effect between COVID-19 and fear sentimentVariables (1) (2)
Intercept 0.036**(2.05) -0.0002(-0.07)Skew (t-1) (t-1) (t-1) Fear Sentiment -0.032***(-2.76)rCase (t-1) D_Fear Sentiment -0.124***(-3.77)R Notes: ***, **, * represent statistical significance at 1%, 5%, and 10% levels, respectively. The t-statistics arepresented in the brackets. Table 3 summarizes the estimated results for the Equation 5. In the first column, weuse the variable, Fear Sentiment, and in column (2), the dummy variable, D_Fear Sentiment, is used. Thetotal observations are 787 over our research period.
It is important to consider the existence of interaction term in our regression dueto two main reasons. First, the investors’ fear exhibit the dynamics pattern with thefatality ratio. It means that when the number of infected cases increase, the investors’sentiments might be induced by the fears. Second, the feelings of fear would mitigatethe risky behaviors on the onset of pandemic. This might lead to the decrease ininfected cases. Therefore, looking at the interaction variable constructed by the / aforementioned components would offer some insights, especially how this factorincreases or decreases the equity market crash risk.There are three main conclusion which could be drawn from the regression inTable 5. First, the interaction variable increases the likelihood of market crash risk at1% significance level. This can be exemplified that both factors amplify the negativeimpact on the equity market shocks. Second, our results are robust although wesubstitute the fear emotion with continuous or binary variable. It does not onlyemphasize the dynamics pattern but also confirms the existing role of investors’emotion on the systematic risk. Third, to compare the previous regression in Table 3and 4, the explanatory level, captured by R , is substantially improved. It implies thatthe interaction variable could positively contribute to the explanatory feature of thechanges in the equity market crash risk.One of the worth noting points is that our existing findings mainly stem from thecorrelation. We are cautious about confirming the sound tone for causal relationshipbefore obtaining any statistical evidence. In doing so, we perform the Granger causaltest to examine whether the fear emotion could cause the market risk crash or not. Ofcourse, the opposite direction is also examined. Table 6 reports the results of the Granger causality test. To examine thehypothesis of each causality, we conduct the F-test. As we can see from the table 6,the fearSent is the Granger cause of
Skew which implies the fear sentiment causes thecrash risk of stock market. On the contrary, the crash risk of stock market is not theGranger causes of the fear sentiment.
Table 6. Results of Granger causality testDirection of causality F-test P-value / fearSent → Skew ***
Skew → fearSent Notes : ***, **, * represent statistical significance at 1%, 5%, and 10% levels, respectively. The nullhypothesis for Granger causality is summarized as ‘ fearSent does not cause the Granger causality to Skew ’( fearSent → Skew ) and the remaining hypothesis is that ‘
Skew does not cause the Granger causality tofearSent ’ (
Skew → fearSent) . Interestingly, we only observe the mono-direction in the Granger causalitybetween fear sentiments and the market crash risk. To be more precise, the fearsentiment is the factor which causes the changes in the market crash risk while we donot have any evidence in the opposite direction. In doing so, we come into conclusionthat the investors’ attitudes towards the uncertainties in terms of fear, macroeconomicsand microeconomics will be stimulus to the risk of market crash risk. Our findingsalso confirm the extant literature of fear and stock market dynamics (such as Bitcoinmarket, Chen et al. (2020); the financial markets, Sharif et al. (2020); energy market,Salisu et al. (2020)). By examining the causal relationship, our policy implicationwould be focused on how to alleviate the investors’ panic to maintain the stability ofthe market. This is important, indeed.
An alternative proxy to measure the severity of the COVID-19 epidemic is the growthrate of daily deaths (rDeaths). We use it to redo our empirical analysis and the resultsare shown in Table 6. In general, the results are consistent with the previous.Furthermore, we substitute the number of global cases for that of the cases of Chinafor another robustness checks since the people in China does not just pay attention tothe progress of COVID-19 in China but also to the global progress. Table 7 and Table8 show the results of the corresponding regressions, which illustrates that ourconclusions remain robust. Thus, we could draw the policy implications from what we / found from the previous results. / Table 6. Robustness results from daily deathsVariables (1) (2) (3) (4) (5)
Intercept -0.001(-0.55) 0.04**(2.29) -0.0005(-0.19) 0.0357**(1.97) -0.0001(-0.05)Skew (t-1) (t-1) -0.115***(-5.06) -0.112***(-4.97) -0.108***(-4.70) 0.520**(2.00) 0.021(0.53)
Fear Sentiment -0.005**(-2.42) -0.029*(-1.83) -0.004**(-2.08) -0.023(-1.44)rDeaths (t-1) Fear Sentiment -0.044**(-2.44) -0.193***(-3.98)R Notes: ***, **, * represent statistical significance at 1%, 5%, and 10% levels, respectively. The t-statistics are presented in the brackets. Model(1), (2), (3), (4), and (5) represent the differences of the set of
Fear Sentiment proxy including none of use, Fear Sentiment, dummy for FearSentiment, Fear Sentiment and dummy for Fear Sentiment, respectively. The total observations are 787 over our research period. / Table 7. Robustness results from daily global casesVariables (1) (2) (3) (4) (5)
Intercept -0.001197(-0.40) 0.033680*(1.83) 0.000024(0.01) 0.020837(1.12) -0.000165(-0.05)Skew (t-1) (t-1) -0.039522***(-3.93) -0.035903***(-3.51) -0.036809***(-3.64) 0.162300***(3.20) -0.007415(-0.59)
Fear Sentiment -0.004065*(-1.92) -0.034670**(-2.14) -0.002498(-1.17) -0.023728(-1.46)r GlobalCases ( (t-1) Fear Sentiment -0.016869***(-3.99) -0.082836***(-3.94)R Notes: ***, **, * represent statistical significance at 1%, 5%, and 10% levels, respectively. The t-statistics are presented in the brackets. Model(1), (2), (3), (4), and (5) represent the differences of the set of
Fear Sentiment proxy including none of use, Fear Sentiment, dummy for FearSentiment, Fear Sentiment and dummy for Fear Sentiment, respectively. The total observations are 787 over our research period.
Table 8. Robustness results from daily global deaths / Variables (1) (2) (3) (4) (5)
Intercept -0.001254(-0.42) 0.035187*(1.91) 0.000004(0.00) 0.035596*(1.93) 0.000248(0.08)Skew (t-1) (t-1) -0.049846***(-3.85) -0.045567***(-3.48) -0.046631***(-3.59) 0.137102(0.55) -0.100354**(-1.97)
Fear Sentiment -0.004246**(-2.01) -0.035512**(-2.19) -0.004297**(-2.03) -0.036279**(-2.24)r GlobalDeaths ( (t-1) Fear Sentiment -0.012501(-0.74) 0.057492(1.09)R Notes: ***, **, * represent statistical significance at 1%, 5%, and 10% levels, respectively. The t-statistics are presented in the brackets. Model (1), (2),(3), (4), and (5) represent the differences of the set of
Fear Sentiment proxy including none of use, Fear Sentiment, dummy for Fear Sentiment, FearSentiment and dummy for Fear Sentiment, respectively. The total observations are 787 over our research period. /
5. Conclusions
This paper has examined the relationship between the COVID-19 and the stockmarket crash risk. First, we confirm that COVID-19 will increase the likelihood of thestock market crash risk. It not only means that it will bring about a decline in stockmarket returns, but also means aggravating the negative symmetry of stock marketreturns and increasing the possibility of extreme downturns in stock prices. Second,we find that even when the number of confirmed cases is not very large, people's fearabout the virus will increase the crash risk of the stock market.It reminds us that preventing the spread of panic during the epidemic is helpfulin reducing the crash risk. Finally, we also find that fear sentiment can not onlydirectly increase the crash risk, it will also boost the negative impact of COVID-19 onthe stock market crash risk. Thus, our study also draws two main policy implications.First, the closer observation from lawmakers on the financial markets with thedynamics of fear and the number of cases is necessary. It is due to the fact that thegovernor could manage how to immediately support the market when the shake offear is overwhelming. By doing this, the market crash risk might be managed in someurgent cases. Second, investors are likely to be sensitive with not only the localinformation (Chinese domestic infected cases or deaths) but also the global news.Therefore, the clear and timely communication regarding the COVID-19 pandemicwould bring the effective prediction to the market. More importantly, both investorsand governor might be cautious about the market crash risk when the number of cases(or deaths) significantly rise. Then, the hedging or safe-haven strategies could beimplemented such as the suggestions of studies of Colon et al. (2020), Colon andMcGee (2020). / References
Albulescu, C. T. (2020). COVID-19 and the United States financial markets’ volatility.
Finance ResearchLetters , 101699.Al-Awadhi, Abdullah M., Khaled Al-Saifi, Ahmad Al-Awadhi, and Salah Alhamadi (2020) Death andcontagious infectious diseases: Impact of the COVID-19 virus on stock market returns,
Journal ofBehavioral and Experimental Finance ' reaction to COVID-19: cases or fatalities? Research inInternational Business and Finance
Technological Forecasting and Social Change ,120261.Baker, Scott R., Nicholas Bloom, Steven J. Davis, Kyle Kost, Marco Sammon, and Tasaneeya Viratyosin,2020, The unprecedented stock market reaction to COVID-19,
The Review of Asset Pricing Studies .Burggraf, T., Huynh, T. L. D., Rudolf, M., & Wang, M. (2020). Do FEARS drive Bitcoin?. Review ofBehavioral Finance. Ahead-of-print. https://doi.org/10.1108/RBF-11-2019-0161Chen, C., Liu, L., & Zhao, N. (2020). Fear sentiment, uncertainty, and bitcoin price dynamics: The case ofCOVID-19. Emerging Markets Finance and Trade, 56(10), 2298-2309.Chen, Joseph, and Harrison Hong, and Jeremy C. Stein (2001) Forecasting crashes: trading volume, pastreturns, and conditional skewness in stock prices,
Journal of financial economics
61, 345-381.Conlon, T., Corbet, S., & McGee, R. J. (2020). Are Cryptocurrencies a Safe Haven for Equity Markets? AnInternational Perspective from the COVID-19 Pandemic.
Research in International Business and Finance ,101248.Conlon, T., & McGee, R. (2020). Safe haven or risky hazard? Bitcoin during the COVID-19 bear market.
Finance Research Letters , 101607.Da, Zhi, and Joseph Engelberg, and Pengjie Gao (2011) In search of attention,
The Journal of Finance
Finance Research Letters,
Accounting &Finance , 58(5), 1291-1318. / Kalyvas, A., Papakyriakou, P., Sakkas, A., & Urquhart, A. (2020). What drives Bitcoin’s price crash risk?.
Economics Letters , 191, 108777.Kartal, Mustafa Tevfik, Özer Depren, and Serpil Kilic Depren. (2020). The determinants of main stockexchange index changes in emerging countries: evidence from Turkey in COVID-19 pandemicage.
Quantitative Finance and Economics , 4(4), 526-541.Kim, Jeong-Bon, and Yinghua Li, and Liandong Zhang (2011) CFOs versus CEOs: Equity incentives andcrashes,
Journal of Financial Economics
Journal of Financial Economics
Research inInternational Business and Finance , 51, 101112.Mazur, Mieszko, and Man Dang, and Miguel Vega. (2020) COVID-19 and the march 2020 stock marketcrash. Evidence from S&P1500,
Finance Research Letters
The Quarterly journal of economics , 127(2), 645-700.Gao, M., Liu, Y. J., & Shi, Y. (2020). Do people feel less at risk? Evidence from disaster experience.
Journal of Financial Economics . https://doi.org/10.1016/j.jfineco.2020.06.010.Giglio, S., Maggiori, M., Stroebel, J., & Utkus, S. (2019). Five facts about beliefs and portfolios (No.w25744).
National Bureau of Economic Research .Giglio, S., Maggiori, M., Stroebel, J., & Utkus, S. (2020). Inside the mind of a stock market crash (No.w27272).
National Bureau of Economic Research .Phan, Dinh Hoang Bach, and Paresh Kumar Narayan (2020). Country responses and the reaction of the stockmarket to COVID-19 — A preliminary exposition,
Emerging Markets Finance and Trade
56, 2138-2150.Ramelli, Stefano, and Alexander F. Wagner (2020) Feverish stock price reactions to covid-19.
NationalBureau of Economic Research
Saeed Sazzad Jeris and Ridoy Deb Nath. (2020). Covid-19, oil price and UK economic policy uncertainty:evidence from the ARDL approach.
Quantitative Finance and Economics / wavelet-based approach, International Review of Financial Analysis
Journal of Behavioral and Experimental Finance , 100349.Wachter, J. A. (2013). Can time‐varying risk of rare disasters explain aggregate stock market volatility?.
TheJournal of Finance , 68(3), 987-1035.Wang, A. Y., & Young, M. (2020). Terrorist attacks and investor risk preference: Evidence from mutualfund flows.
Journal of Financial Economics . https://doi.org/10.1016/j.jfineco.2020.02.008.Wen, F., Xu, L., Ouyang, G., & Kou, G. (2019a). Retail investor attention and stock price crash risk:Evidence from China.
International Review of Financial Analysis , 65, 101376.Wen, F., Xu, L., Chen, B., Xia, X., & Li, J. (2019b). Heterogeneous institutional investors, short selling andstock price crash risk: Evidence from China.
Emerging Markets Finance and Trade , 56(12), 2812-2825.Wen, F., Yang, X., & Zhou, W. X. (2019). Tail dependence networks of global stock markets.
InternationalJournal of Finance & Economics , 24(1), 558-567.Yarovaya, L., Matkovskyy, R., & Jalan, A. (2020). The Effects of a 'Black Swan' Event (COVID-19) onHerding Behavior in Cryptocurrency Markets: Evidence from Cryptocurrency USD, EUR, JPY and KRWMarkets.
EUR, JPY and KRW Markets . http://dx.doi.org/10.2139/ssrn.3586511Zhang, Dayong, and Min Hu, and Qiang Ji. (2020a). Financial markets under the global pandemic ofCOVID-19,
Finance Research Letters