COVID-19 pandemic and capital markets: the role of government responses

The literature on behavioral finance has extensively explored whether investor sentiment affects trading decisions, leads to irrational trading behavior, and affects stock prices (e.g., Baker and Wurgler 2006; Beer and Zouaoui 2012; Chau et al. 2016; Cormier et al. 2010; Fang and Peress 2009; He et al. 2020; Hong and Stein 2007, 1999; Kaniel et al. 2008; Long et al. 1990; Palomino 1996; Shleifer and Vishny 1997). Bollen et al. (2011) find that investors’ trading behavior is directly shaped by their perceptions about future market development. Anxiety increases investors’ risk avoidance and contributes to pessimism (Baker and Wurgler 2006; Cen and Liyan 2013).

One relevant driver of investor perceptions and expectations is news. Tetlock (2007) shows that news media pessimism predicts downward pressure on market prices, even leading to a revision of fundamentals. Subsequent studies find that bad news leads to more intense or even panic-driven, long-term stock market effects (Cohen et al. 2018; Jung et al. 2018; Miller and Skinner 2015). A subordinate literature stream covers the formation and consequences of investor overreactions. In a fundamental study, de Bondt and Thaler (1985) investigate whether, and following which rules, the stock market overreacts. They present experimental evidence that in probability revision problems, people show a tendency to overreact, i.e., they overweight recent information and underweight base rate data. Barberis et al. (1998) link investor overreactions to news. They analyze the impact of good and bad market-relevant news announcements as a moderator of investor overreaction and stock return developments. Results show that investors overreact to consistent patterns of good (bad) news with a correlation to the length of good (bad) news series. Michayluk and Neuhauser (2006) analyze the role of investor overreaction during the market decline following the 1997 Asian financial crisis. They conclude that overreaction is sensitive to news announcements and is traceable for 1 week after the announcement. On a different note, Gennaioli et al. (2015) argue that a single piece of bad news in a series of good news items will not lead to a change in the underlying beliefs of investors regarding certain stocks. Nevertheless, a continuous series of bad news results in investor overreaction because previously ignored bad news is remembered, 'leading to a sharp rise in the perceived probability of a crisis and a collapse of prices' (Gennaioli et al. 2015, p.312).

Another influencing factor on investor sentiment is trust. Georgarakos and Pasini (2011) find an association between trust in financial markets and investors’ trading behavior. Based on a portfolio model using survey data, they show that in countries where investors exhibit a high level of trust toward the capital market, the stockholding share is significantly higher than in low-trust countries. Lesmeister et al. (2018) analyze whether shareholders extend or reduce their monitoring activity (e.g., by shareholder votes) relative to the general trust that they experience. The study reveals that trust reduces the amount of shareholder monitoring activity. Moreover, when exposed to external shocks, such as terrorist attacks, investors’ trust decreases by an increase in announced fatalities.

As financial crises caused by epidemics or pandemics are not without precedent, few studies have taken the occasion of previous health crises to expand the knowledge of their power to impact investor sentiment. Funck and Gutierrez (2018) examine the impact of Ebola headline news on media-highlighted stocks. They employ the VIX-Investor Fear Gauge (VIX), introduced by Whaley (2009), to proxy for investor pessimism during Ebola outbreak announcements. They reveal that stock prices tend to reverse themselves within one day after the announcement, supporting the traditional theory of investor overreaction. Avian influenza (bird flu), initially reported in China in March 2013, caused a loss in the agriculture sector by $6.5 billion due to changes in prices, consumer confidence and trade volumes. Jiang et al. (2017) explored the impact of daily avian influenza case announcements on stock prices. They find peaks in investors’ overreacting behavior within the initial outbreak announcements; that is, investors seem to act more reasonably in time when the shock of the first outbreak news has been overcome.

Ichev and Marinč (2018) focus on the 2014–2016 Ebola outbreak. They analyze the effect of mass media news announcements about Ebola outbreak events on firms’ stock returns. They find that investors act irrationally to the news on the Ebola outbreak and that Ebola outbreak events unequally affect investors’ sentiment about stock returns, depending on the distance of the outbreak event from the markets. Negative effects on financial markets are stronger for firms that operate in countries with a larger Ebola exposure. They conclude that a firm’s geographic proximity to the Ebola outbreak event increases the impact on its stock returns.

In a recent study, Engelhardt et al. (2020) find that COVID-19-induced stock market declines in 64 countries are mainly associated with greater news attention and less with rational expectations. Following this research, we expect the growth rates of COVID-19 cases and deaths in firm revenue-relevant countries to depress investor sentiment by provoking panic and pessimism, resulting in temporary market overreactions and leading to a decline in stock returns during the pandemic.

With rising COVID-19 cases and deaths, several governments worldwide took action to mitigate the repercussions of the disease and introduced diverse sets of countermeasures, i.e., government responses. In addition to restrictive policies to contain the spread of the disease (e.g., travel restrictions, stay-at-home requirements, school closings, public transport closings), stimulus packages were implemented to mitigate the economic impact of the pandemic (e.g., income support, debt relief, fiscal measures). Moreover, proactive measures were set in motion to reduce infection rates (e.g., testing policies, vaccination policies, contract tracing) (Hale et al. 2021). Following the literature on investor sentiment, we argue that government responses to COVID-19 can reinforce or mitigate the negative impact of the COVID-19 pandemic on stock prices. Governments may reduce uncertainty among investors, regain investors’ trust and lead to a less severe decline, or even rise, of firms’ stock returns, depending on the aim and scope of the policy. Moreover, we expect investors to rate government responses differently depending on a firm’s exposure to COVID-19; this is how severely those firm-relevant countries are affected, which directly contribute to the firm’s revenues.

We obtain daily data on worldwide confirmed COVID-19 cases and deaths per country from January 1, 2020 to March 15, 2021 from the European Centre for Disease Prevention and Control (ECDC). We calculate the daily growth rates of both cases and deaths in each country, relative to a country’s population, by dividing the number of daily cases and deaths per million in country i on day t by the same measure of the previous day t−1. We receive a subsample of 180 countries and 563 observation days per country. Our sample is unbalanced since, for some sample countries, no cases or deaths were reported until April 2020.

To measure firm-level stock market reactions to the COVID-19 pandemic, we obtain daily stock prices for 511 firms that make up the S&P 500 index during our observation period from January 1, 2020 to March 15, 2021. To solely assign COVID-19 cases and deaths and government responses of firm-relevant countries to a firm, we obtain country-specific sales revenues [REVENUE_i,c,t] for the compounding S&P 500 firms for 2019 from the FactSet Geographic Revenue (GeoRev) Database. We assign all countries to a firm that contributed to the firm’s revenues, leading to a combination of multiple countries for each firm per observation day. We only include values with a certainty score of 70 or above, as provided by GeoRev. The certainty score is based on source metadata and ranges from 1 (low certainty) to 80 (declared value). This proceeding enables us to isolate information that is assumed to be of decision-relevant significance to each firm’s investors. See Appendix 1 for a list of countries included and the number of firms that realize revenues in the respective country.

We employ data from the OxCGRT, published by Hale et al. (2021), to measure country-specific government responses to the pandemic. Our dataset covers 16 individual ordinal scaled indicators that represent the strictness of various government policies in response to the COVID-19 pandemic. It should be noted that these values do not reflect the effectiveness of each policy. All indicators can be classified into three groups, representing the scope of the policy, i.e., containment and closure policies, health system policies, and economic policies. We include the following indicators:

Containment and closure policies

School closings; workplace closings; cancelations of public events; restrictions on gatherings; closing of public transport; stay-at-home requirements; restrictions on international movement; international travel controls

Health system policies

Public information campaigns; testing policy; contact tracing; facial covering policy; vaccination policy; protection of elderly people

Economic policies

Income support; debt and contract relief

Please see Appendix 2 for a full list of indicators, including detailed descriptions and scale coding. To reflect the extent of each government’s efforts, we calculate an index for each of the three policy groups.¹ Since all indicators are ordinally scaled and ranked with different values set as maximums, we calculate subindices to normalize each indicator to an equally spaced scale between 0 and 100. The three indices, i.e., containment and closure index, health system index, and economic support index, are then calculated as simple averages of the normalized individual subindices. Moreover, we aggregate all three indices to create a summarized government response index. We merge our government response data with the firm-country dataset, including growth rates in the number of cases and deaths. Our final sample consists of 10,060,911 daily firm-country-specific observations.

To test whether government responses to the COVID-19 pandemic moderate the impact of confirmed and announced COVID-19 cases and deaths on firms’ daily stock returns, we estimate the following regression model using firm- (Firm FE) and day-fixed (Day FE) effects*:where i, c, and t index firm, country, and day, respectively.

Our dependent variable [AR_i,t+1] is measured as adjusted abnormal logarithmic returns using a single-index market model, where expected returns are estimated with market betas using the firm’s daily stock returns and the S&P 500 index returns of the respective period (e.g., Brown and Barry 1984; Dai and Zhu 2020; Jain 1986; Sharpe 1963). Specifically, we define abnormal returns as \(AR_{t}^{i} = R_{t}^{i} - E\left( {R_{t}^{i} } \right)\), where \(R_{t}^{i}\) represents the daily return for firm i on day t. We estimate the firm’s expected return as \(E(R_{t}^{i} ) = \alpha_{i} + b_{i} E(R_{t}^{market} )\), with \(R_{t}^{market} = (P_{t}^{market} - P_{t - 1}^{market} )/P_{t - 1}^{market}\), where \(P_{t}^{market}\) represents the S&P 500 index closing price on day t.² Following prior research on the stock market impact of infectious diseases, our model parameters are estimated over a 90-trading-day estimation period, starting one day prior to day t to prevent unusual effects on the measurement day from interfering the estimation (e.g., Liu et al. 2020; Wang et al. 2013).

COVID-19_c,t represents either the growth rate of the announced cumulative number of confirmed COVID-19-positive cases [CASES_c,t] or deaths [DEATHS_c,t] associated with or caused by the disease per million in country c for day t. RESPONSE_c,t corresponds to each of our four daily government response indices, i.e., containment and closure index [CONTAINMENT_CLOSURE_c,t], health system index [HEALTH_SYSTEM_c,t], economic support index [ECON_SUPPORT_c,t], and the summarized government response index [GOV_RESPONSE_c,t]. We use a set of control variables based on prior literature. We control for abnormal returns one day prior to the measurement day by using abnormal returns [AR_i,t] in autoregression to consider stock return autocorrelation (e.g., Kraft and Kraft 1977; Smirlock and Starks 1988). Zaremba et al. (2020) and Ali et al. (2020) reveal a significant impact of the COVID-19 pandemic on market volatility. Hence, we include a firm’s annualized volatility of logarithmic stock returns during our observation period [VOLATILITY_i,t]. We control for the number of analysts following a firm’s share [FOLLOWING_i,t] as well as for the percentage of institutional holdings [INSTITUTIONAL_i,t], with both variables to be measured annually (e.g., Bartov et al. 2018; Chen et al. 2013; Hong et al. 2000). A broad number of studies investigate the relationship between firm fundamentals and stock market returns. We follow these findings by including firm size as proxied annually by the logarithm of a firm’s total assets [SIZE_i,t] and the price-to-book ratio as a firm’s daily market price per share divided by the share’s book value [PTB_i,t] (e.g., Fama and Franch 1993; Griffin and Lemmon 2002; Pontiff and Schall 1998). Moreover, corporate governance research finds that the number of a firm’s board members shapes board integrity and effectiveness and, thus, reinforced by the perception of investors, affects returns (Cheng 2008; González et al. 2019). Hence, we include the quarterly board size [BOARD_i,t] in our set of controls. Davison (2020, p. 2) finds that 'the stocks returns of firms who are relatively unable to transition their business to comply with social distancing are much more responsive to changes in their level of leverage going into the pandemic'. We, therefore, process a firm’s leverage change during our observation period as the yearly differences in the ratio of a firm’s book value debt and total assets [LEVERAGEi,t]. Hu and Zhang (2021) show that firm performance deteriorates during the COVID-19 pandemic. However, the effect is smaller for firms in countries with better health systems, more advanced financial systems, and better institutions. We address these findings by controlling for the quarterly change in a firm’s return-on-assets [ROA_i,t], calculated as operating income before depreciation over total assets. Several studies discuss a firm’s economic, social, and governmental scores (ESG scores) to indicate the resilience of stock prices during the COVID-19 pandemic. However, the results of preliminary studies are controversial. Albuquerque et al. (2020) find that stocks with higher ESG scores have significantly higher returns, lower return volatility, and higher operating profit margins during the first quarter of 2020. Broadstock et al. (2021) add that ESG performance mitigates financial risk during a financial crisis and that high-ESG portfolios generally outperform low-ESG portfolios. In contrast, Demers et al. (2020) conclude that higher ESG scores do not immunize stocks. We control for ESG scores [ESG_i,t] using the most prevalent Refinitiv ESG SCORE³ that weekly measures a company’s relative ESG performance, commitment and effectiveness across 10 main themes, including emissions, environmental product innovation and human rights (Refinitiv 2020). Since the discussed literature on investor sentiment assigns an important role in affecting investor sentiment to news media, we control for three variables covering the spread, perception, and sentiment of COVID-19 news in each country. Data were obtained from the RavenPack Coronavirus Media Monitor. MEDIA_COVERAGEc,t calculates the daily percentage of all news agencies in country c on day t covering the topic of COVID-19. The index is computed as the daily number of distinct news agencies that mention COVID-19, divided by the total available number of news agencies in the country. MEDIA_HYPE_c,t measures the percentage of news that is currently reporting about COVID-19 in country c on day t, regardless the originating news agency. The index is computed as the daily number of reports that mention COVID-19, divided by the total daily number of reports. MEDIA_SENTIMENT_c,t measures the level of sentiment that news reports express towards a firm that is mentioned in the report alongside COVID-19. Specifically, it reflects the daily average of the difference between the number of positive and negative news reports. RavenPack determines positive or negative sentiment by 'systematically matching stories usually categorized by financial experts as having a positive or negative financial or economic impact. The algorithm produces a score for more than 6900 categories of business, economic, and geopolitical events, ranging from earnings announcements to natural disasters.' (Hafez et al. 2020). The index ranges between − 100 and 100, where a value of 100 is the most positive sentiment, − 100 is the most negative, and 0 is neutral. All variables are defined in Appendix 3.

Table 1 reports descriptive statistics for all variables in our main regression model specifications. All variables are winsorized at the 1st and 99th percentiles to mitigate the influence of outliers. The mean value of firm-specific abnormal returns is 0.001 [AR_i,t+1]. The mean annualized volatility of stock returns is 31.336 [VOLATILITY_i,t], which is remarkable since the S&P 500’s annualized average volatility from 1926 through 2017 is 15.2.⁴ The mean share of revenues that a firm derives from a country is 1.126% [REVENUE_i,c,t], with a maximum share of 53.175%. This maximum coincides with the average revenue share that S&P 500 firms realize in the United States, as expected. As the median value is 0.083 and the 75th percentile shows a value of 0.301, the highest revenue firms are allocated to a small number of countries. On average, a firm is followed by 22 analysts [FOLLOWING_i,t], and the mean share of institutional owners [INSTITUTIONAL_i,t] is 82.829%. Firm size [SIZE_i,t], price-to-book-ratio [PTB_i,t], and the level of leverage [LEVERAGE_i,t] show mean values of 223.730 bn, 5.730, and − 0.973, respectively. A firm’s board comprises approximately 11 members. Unsurprisingly, concerning firm performance, the average return on assets is negative at – 1.656%. Thomson Reuters Refinitiv calculates a mean ESG score [ESG_i,t] of 63.582 for our sample firms. Turning to country-specific data, on average, a country is exposed to a daily growth of 5.926% in COVID-19 cases per million people [CASES_i,t] and a daily growth in COVID-19-related deaths per million people of 0.129% [DEATHS_c,t]. We calculate our index for a country’s containment and closure policies [CONTAINMENT_CLOSURE_c,t] to average 49.616. Values for our index covering a country’s efforts to support the health system [HEALTH_SYSTEM_c,t] as well as the economy [ECON_SUPPORT_c,t] rank closely at 54.449 and 43.965, respectively. Our summarized measure for a government’s effort in all fields [GOV_RESPONSE_c,t] exhibits a mean value of 54.940. Our controls for the spread, perception, and sentiment of COVID-19 news in each country reveal the following descriptive insights: On average, 55.511% of all news agencies in a country cover the topic of COVID-19. The percentage of news that reports on COVID-19 each day in a country is 46.118. The mean level of sentiment toward a country mentioned in the news alongside COVID-19 is − 5.930, with an overall sample maximum of 50.950 of 100 index points and a minimum of − 97.210. The median remains negative at − 2.460.

Figure 3 shows the S&P 500 stock market index and the intensity of the global government response index [GOV_RESPONSE_c,t] to COVID-19 for our observation period from January 1, 2020 to March 15, 2021. The intensity of global government responses is calculated by accumulating all 16 policy indicators per day over all 180 countries. The scale is normalized on a range of 0–100, where 100 represents the maximum intensity for the observation period. From March 5, 2020 on, worldwide initiatives to contain the spread of the pandemic are increasingly put in place. Hence, the global government response index grows exponentially. The growth accelerates from March 13, 2020, when the WHO declares the COVID-19 outbreak to officially be a pandemic. At the same time, the S&P 500 stock market index experiences a decrease that seemingly mirrors the government response’ s development. After a peak on April 17, 2020, the response index values diminish throughout the summer months, starting from approximately the date when the S&P 500 index recovers to its pandemic pre-announcement value. Towards winter, global government responses increase again slightly despite a continuously bullish stock market.

Figure 4 shows the S&P 500 stock market index and the intensity of accumulated global government responses to COVID-19 for our observation period, split for policies in the fields of containment and closure [CONTAINMENT_CLOSURE_c,t], health system [HEALTH_SYSTEM_c,t], and economic support [ECON_SUPPORT_c,t]. As before, the scale is normalized on a range of 0–100, where 100 represents the maximum intensity for the observation period. All three indices follow the path of the accumulated, global government response index, as illustrated in Fig. 3. However, health system initiatives are implemented later compared to containment and closure policies or economic support. Containment and closure policies experience the strongest decline, with health system initiatives still increasing. In the course of the bullish stock market, economic support and containment and closure policies are reduced, while health system support is still being extended.

Table 2 presents the Pearson–Spearman correlations. Both CASES_c,t and DEATHS_c,t are significantly negatively correlated with AR_i,t+1 (Pearson − 0.042; Spearman − 0.029 and Pearson − 0.027; Spearman – 0.022). This is a first indicator of the negative impact that rising COVID-19 cases and deaths cause on stock prices. GOV_RESPONSE_c,t shows a significant and positive correlation with AR (Pearson 0.023; Spearman 0.019), as CONTAINMENT_CLOSURE_c,t does (Pearson 0.064; Spearman 0.055). In contrast, correlation coefficients for the remaining two indices, HEALTH_SYSTEM_c,t and ECON_SUPPORT_c,t, are significantly negative regarding AR_i,t+1 (Pearson − 0.032; Spearman − 0.041 and Pearson − 0.029; Spearman − 0.018). These findings imply that a split investigation of government responses is imperative since different government policies may provoke different market reactions. All four government response indices are significantly and positively correlated to both CASES_c,t and DEATHS_c,t, exhibiting high magnitudes (e.g., COV_RESPONSE_c,t is correlated to CASES_c,t with Pearson 0.0552; Spearman 0.427). This illustrates the sensitivity of government interventions worldwide to rising COVID-19 cases and deaths. Turning to our controls, we find interesting values for our news media measures. As expected, both MEDIA_COVERAGE_c,t and MEDIA_HYPE_c,t are positively correlated with CASES_c,t and DEATHS_c,t (e.g., MEDIA COVEERAGE_c,t and CASES_c,t Pearson 0.192; Spearman 0.140). However, MEDIA_SENTIMENT_c,t is negatively correlated with COVID-19 proxies (CASES_c,t: Pearson − 0.078; Spearman − 0.110, DEATHS_c,t Pearson − 0.124; Spearman − 0.123). This leads to the interpretation that the amount of news that mentions COVID-19 and specific countries increases as COVID-19 cases and deaths grow. At the same time, the sentiment expressed in the news media toward countries becomes negative as COVID-19 numbers grow. Overall, the absence of high correlations among our variables suggests that there are no multicollinearity concerns.

First, we examine whether the impact of COVID-19 on firms’ daily abnormal stock returns is moderated by the summarized government responses of countries that contribute to a firm’s revenue. Table 3 presents our main regression results for Eq. (1).

Models I and II examine the raw impact of the growth rate of the announced number of confirmed COVID-19-positive cases [CASES_c,t] or deaths [DEATHS_c,t] per million associated with or caused by the disease on abnormal returns [AR_i,t+1]. As expected, results show that the stock market responds significantly negatively to both COVID-19 proxies. However, the daily growth in the number of country-specific deaths is of greater relevance to investors since the coefficient for DEATHS_c,t (− 0.063) is larger in magnitude than the coefficient for CASES_c,t (− 0.038). These findings are contrary to previous studies on the impact of COVID-19 cases and deaths on the stock market that expose cases to mainly drive the stock market (e.g., Alexakis et al. 2021; Ali et al. 2020; Erdem 2020). This contrast is most likely explained by the longer observation period that is considered in our study. Specially, when analyzing the consequences of diseases for the first time, a longer observation period allows for incorporating the different stages it passes. More specifically, in the initial period, the spread of the pandemic was mainly measured (and publicly discussed) by the increasing number of cases. In the course of the pandemic, this perspective changed due to worldwide high levels of cases, and deaths became a more important focus of interest for health organizations, governments, and news media. In addition, as is shown in Fig. 1, the increase in the number of deaths was delayed compared to the increase in the number of cases. Hence, due to short observations covering the initial period of the pandemic, most preliminary studies lack enough data to reveal robust and interpretable results concerning deaths.

Our firm specific controls perform as expected and support the findings of the previous studies on stock market reaction we discussed. Turning to our media-related controls, we find MEDIA_COVERAGE and MEDIA_HYPE to show positive effects on stock returns. This is unsurprising since prior research on behavioral finance has well explored that the excessive presence of news, regardless of the expressed sentiment, leads to a higher attention of investors, and thus, gives positive momentum to the stock market development (e.g., Andrei and Halser 2015; Engelhardt et al. 2020). At the same time, MEDIA_SENTIMENT shows a significant and negative impact on stock returns. As this variable incorporates both positive and negative sentiment, and descriptive statistics reveal a mean of − 5.930, and thus, predominantly negative sentiment throughout our observation period, this result follows prior research that finds negative news to negatively influence the stock markets (e.g., Cohen et al. 2018; Jung et al. 2018).

In Models III and IV, we add our summarized government response index [GOV_RESPONSE_c,t] to analyze whether the entirety of a country’s responses to the pandemic moderates the association shown in the previous models. Across both models, we find positive and statistically significant coefficients on the moderation of cases and deaths by government responses. This indicates that the entirety of government policies positively influences investor sentiment, retriggers optimism, restores investor trust and eventually mitigates the decline of stock prices. In other words, market participants seem to appreciate governments’ efforts to contain the consequences of the pandemic. Our control variables follow similar patterns as in the previous models. Within each regression, F tests indicate that the coefficients on the CASES/DEATHS and GOV_RESPONSE variables are significantly different from each other in all specifications, suggesting that both variables add explanatory value to our model.

Since we find summarized government responses to play an important role in stock market development during the COVID-19 pandemic, we are now interested in dividing our GOV_RESPONSE_c,t index into the different scopes of government responses, measured by our 16 country-specific indicators, to see whether different scopes of responses mitigate or reinforce the negative stock market impact of the pandemic. Table 4 presents the results of our regression models covering the moderating effects of government containment and closure policies [CONTAINMENT_CLOSURE_c,t], the support of the country’s health system [HEALTH_SYSTEM_c,t], and the support of the economy [ECON_SUPPORT_c,t].

In Models I and II, we estimate the additional moderating effect of containment and closure policies in coherence with CASES_c,t and DEATHS_c,t, respectively. Both interaction coefficients are positive and statistically significant. Hence, governments mitigate negative market impacts of the pandemic by taking actions to contain the spread of the disease, e.g., school closings, workplace closings, closings of public transport or issuing stay-at-home orders.

Government economic support [ECON_SUPPORT_c,t] in Models V and VI has similar effects. Investors seem to appreciate the efforts of governments to mitigate the economic consequences of the pandemic by relieving debts and contracts or by supporting incomes. They most likely adjust their perceptions about market development and, in consequence, positively adjust their investment decisions.

Interestingly, government support of a country’s health system [HEALTH_SYSTEM_c,t], shown in Models III and IV, causes further declines in abnormal stock returns as the interaction coefficients are negative and significant. For the interpretation of this counterintuitive effect, it is helpful to reinvestigate Fig. 4. The pathway of the index for government efforts globally in supporting health systems [HEALTH_SYSTEM_c,t] differs from the remaining two indices. From the date the WHO pronounced COVID-19 a pandemic on March 13, 2020 until the S&P 500 recovered to a preannouncement value on March 30, 2020, the growth of the health system index was significantly smaller, suggesting a smaller contribution to stock market declines. As the situation progressed, with recovering S&P 500 values, the health system index continued growing, while the other two indices showed persistent declines. Analyzing the index composition, one explanation may be the initial weakness and the delay of efforts in strengthening the health system. For example, widespread testing initiatives were implemented late due to the absence of reliable tests. This more obviously holds true for vaccination campaigns. Because health system policies are mainly implemented when the stock market was on a path of recovery and was gaining momentum, investors apparently feared restrictions or policies in the health sector that would interfere with the boom again. This uncertainty most likely provoked pessimism and, in consequence, negative stock market effects. Results from F tests indicate a significant difference in the coefficients on CASES/DEATHS and the government response measures.

The coefficients of our sets of firm and country-specific control variables remain stable in significance, magnitude, and direction with indistinguishable differences from our previous regression covering summarizes government responses.

In contrast to most studies on the economic consequences of the COVID-19 pandemic we discussed, we investigate COVID-19 related impacts on firm-level rather than aggregating an entire economy. This approach allows a detailed investigation of the linkage between the COVID-19 pandemic, worldwide government regulations, and multinational firms’ characteristics.

One major field of interest for both research and practice is whether some firms are more severely affected by the COVID-19 pandemic and by government responses than others. Recent studies show various firm-specific characteristics to moderate the extent of the declines in stock returns associated with the pandemic. For example, Ding et al. (2021) find that firms with high exposure to supply chain disruptions show greater stock price declines.

We aim to analyze whether country-specific sales revenues at the firm level influence the effect of governmental responses to COVID-19-associated stock market effects. Specifically, we are interested in whether investors react differently to a growth in the number of COVID-19 cases and deaths and related government responses when a firm is more severely affected by this growth on the sales side.

We employ the comprehensive FactSet Geographic Revenue Exposure (GeoRev) Database that meters a firm’s annual sales revenue for each of the world's countries it operates in. The data is derived from a broad set of sources, e.g., summarized annual reports. In addition, 'an estimation algorithm based on GDP weighting and accounting logic is then applied to solve for any non-explicit disclosures' (FactSet 2022). The result is a detailed breakdown of a company’s revenues into any geographic country. The database also provides a certainty score that is based on source metadata and ranges from 1 (low certainty) to 80 (declared value). We only include values with a certainty score of 70 or above.

We assume, that for each firm, country-specific revenues reflect the importance of a country to the firm’s economy. The risk of losing revenue in a country may increase when the COVID-19 pandemic’s intensity is high. To combine both the importance of a country for a firm’s revenues and the risk of losses in a country, we calculate an indicator for firm-country-level sales revenue exposure to the pandemic. Thus, EXPOSURE_i,t is the country-specific sales revenues of our sample firms [REVENUE_i,c,t], weighted by country’s daily growth rates of COVID-19 cases [CASES_c,t] and deaths [DEATHS_c,t] per million. For instance, a high infection or death rate in a country that does not significantly contribute to a firm’s sales revenues may not be considered as a threat to that firm. The same may holds true for a country that highly contributes to the firm’s sales revenue but shows low infection rates.

The variable is then standardized using z-scores and normalized to a scale from 0 to 100, where higher values indicate a larger firm-specific revenue exposure to COVID-19. For example, a value of 100 would be related to a firm-day observation if the growth in COVID-19 cases or deaths among all firm-relevant countries is the highest in the countries that contribute most to the firm’s revenues. In the same way, a value of 0 would be related to a firm-day observation if there is no growth in COVID-19 cases or deaths among all countries that contribute to a firm’s revenues. We make the following adjustments to our regression model in Eq. (1):where i, c, and t index firm, country, and day, respectively.

Results are presented in Table 5, where Panels A and B observe CASES_c,t and DEATHS_c,t, respectively, as proxies for COVID-19_c,t.

In Panel A Model I, the three-way interaction reveals that the mitigating effect of summarized government responses [GOV_RESPONSE_c,t], as shown previously, is even stronger for firms that are highly exposed to COVID-19 on the sales side. The coefficient is positive and significant but shows a low magnitude.

The same holds true for governmental efforts in economic support [ECON_SUPPORT_c,t], presented in Model IV. In contrast, containment and closure policies [CONTAINMENT_CLOSURE_c,t] (Model II) seem to be rated differently by investors in the case that a firm’s revenue is highly exposed to COVID-19. While in our main regression model with split government responses by scope, containment and closure policies mitigate stock price declines due to growing COVID-19 case numbers, they seem not strong enough to do so if a firm achieves high sales revenues in countries that are highly affected by the pandemic.

Turning to health system policies [HEALTH_SYSTEMc,t] (Model III), government support seems to be rated more positively when investors are more severely affected by the pandemic. Moreover, the additional positive effect of government health system policies for firms that are highly exposed to the pandemic (coefficient 0.033) even reverses the initial negative effect (coefficient − 0.015), where the moderation of EXPOSURE_c,t was unconsidered. Thus, investors of firms with high COVID-19 revenue exposure aim for a fast recovery and positively assess health system policies, regardless of consequential restrictions. In Panel B, COVID-19_c,t is represented by DEATHS_c,t. For all models, the results are similarly directed. As a sole exception, we fail to find a significant three-way interaction for the economic support index.

In our main regression models, we employ the percentage growth rates for both COVID-19 cases and deaths to proxy for the spread of the COVID-19 pandemic. Specifically, for the increase in COVID-19 cases and deaths, we consider the growth rate of the announced cumulative number of confirmed COVID-19-positive cases per million and country and the growth rate of the announced deaths associated or caused by COVID-19 per million and country, respectively. However, this perspective assumes a linear progression of the pandemic.

In fact, as graphically visible in Fig. 1, the spread of COVID-19 can be separated into two phases: exponential growth of cases and deaths corresponding to initial global outbreaks, followed by logistic progression due to mitigated infection rates as global government responses unfold (De Silva et al. 2012). Epidemiological standard models can be used to address this nonlinearity. Comparing the performance of standard models for fast epidemics, Ma (2020) shows that the susceptible-infectious-recovered model (SIR), first published by Hethcote (1989), provides a robust estimate for the spread of a disease exhibiting a pattern similar to COVID-19. SIR assumes the immunity of recovered individuals and incorporates the number of susceptible individuals. We calculate the SIR growth rate following Ma (2020) and Furtado (2021) aswhere S is the share of susceptible individuals, I is the share of infectious individuals, and R is the share of recovered individuals. ß is the transmission rate per infectious individual, and y is the recovery rate. With the total number of individuals kept constant as the sum of S, I, and R, the expected growth rate is calculated as λ = ß – y. Descriptive statistics provide a mean of 4.531%. We recalculate our regression models, substituting our measures for the spread of the disease, i.e., CASES_c,t, by the SIR growth rate. Untabulated results remain congruent within all our regression models, indicating no distortion due to an imprecise, exponential fit of our COVID-19 proxies.

Clearly, our study covers a worldwide economic crises that, without precedent, can be expected to systematically affect almost all sectors and firms worldwide. Thus, it is particularly important to find a benchmark for a firm’s returns that does not ignore the overarching effects of the crisis and is free of systematic influences. Thus, we employ three alternative approaches for the calculation of abnormal stock returns.

First, we recalculate our models using the daily average market return of the entire US market instead of focusing on the S&P 500. Therefore, we obtain the daily index prices of the Dow Jones U.S. Total Stock Market Index for our observation period. The index measures all US equity issues with available prices and covers 4224 firms and ten sectors. Similar to our main regression, we define the market adjustment of raw returns as \(AR_{t}^{i} = R_{t}^{i} - E\left( {R_{t}^{i} } \right)\), where \(R_{t}^{i}\) represents the daily return for firm i on day t. We estimate the firm’s expected return as \(E(R_{t}^{i} ) = \alpha_{i} + b_{i} E(R_{t}^{market} )\), with \(R_{t}^{market} = (P_{t}^{market} - P_{t - 1}^{market} )/P_{t - 1}^{market}\), where \(P_{t}^{market}\) now represents the Dow Jones US. Total Stock Market Index closing price on day t. As all estimates are statistically indistinguishable from one another, evidence is provided for our main model’s results.

Second, we apply a portfolio-based approach to incorporate firm-specific risk instead of simply observing the daily market average. Specifically, following Fama and French (1995) and Kothari and Warner (2004), we compute a firm’s abnormal returns by adjusting the total returns for factors that have been found to explain cross-sectional differences in stock returns, i.e., a firm’s market capitalization and its price-to-book ratio. Similar to Brav et al. (2000), we form a set of floating portfolios of firms, with each portfolio expressing a distinct range of firm-size, i.e., a firm’s capitalization. Within each portfolio, we rank the firms by their price-to-book ratio. Based on our ranking values, we define a weighting factor \(w_{j}\) for each ranking position. Total returns \(R_{t}^{i}\) are then adjusted by the weighted total returns with varying values for each observation day: \(AR_{t}^{i} = R_{t}^{i} - R_{t}^{i, port}\). Following Sul et al., (2014), we calculate \(R_{t}^{i}\) as the natural logarithm of total returns plus one:\(R_{t}^{i} = ln\left( {R_{t}^{i} + 1} \right)\). \({\varvec{R}}_{{\varvec{t}}}^{{{\varvec{i}},\user2{ port}}}\) therefore, can be defined as \(\frac{1}{n}\sum\nolimits_{j = 1}^{n} {w_{j} \left( {R_{t}^{i} } \right)}\), where the sum of weights in each portfolio equals 1 (\(\sum\nolimits_{j = 1}^{n} {w_{j} = 1}\)). All results remain unchanged in direction and show insignificant divergences in magnitude.

In addition, the approach includes several country-specific control variables, e.g., media-sentiment variables, that continuously measure the spread and tone of pandemic-related news in a country. Following prior research that we discussed beforehand, they may be considered important drivers of investor sentiment.

However, a straightforward way of analyzing our dataset is to aggregate all data at the firm level. Specifically, on firm level, we weight a country’s daily COVID-19 related case and death numbers with its intensity of government responses. The result is a global variable for the overall strength of COVID-19 government policies for each firm. In this approach, again, all countries that contribute to the firm's sales revenues, are included. This calculation is blurred since it averages the country-specific government responses and the COVID-19 measures worldwide. However, it may still reinforce the robustness of our research design. We recalculate our regression models, using the aggregated data on firm level. Results remain unaffected and show similar magnitudes and directions.

We perform further robustness tests to validate the results of our regression models. First, we include country-fixed effects to control for country-level heterogeneity. Again, results remain unchanged in direction and magnitude. Second, to control for cross-effects indicating that abnormal stock returns influence government responses, we repeat all regressions using RESPONSE_c,t as the dependent variable. We do not find significant results, supporting our main findings. Third, we conduct an in-time placebo test using placebo time windows to ensure that our regression results are not driven by our research design (e.g., Conley and Taber 2011; Hahn and Shi 2017). We run our main regression model shifting all independent variables back and forth 10, 15 and 30 trading days, respectively, holding our dependent variable, abnormal returns, constant. We fail to find any significant results, suggesting that, assuming no treatment effects, there is no evidence of random or systematic errors due to a weak model design. Forth, the reported results remain stable when conducting random effect regressions. Fifth, we use altering data frequencies to assure that our results are not sensitive to the daily-data approach. We aggregate all variables both weekly and monthly and rerun our main regressions. Results do not reveal divergences. Sixth, we recalculate our analysis substituting our dependent variable, abnormal returns, with raw returns. Results follow similar patterns throughout all models.