key: cord-0821385-wo3t5izq authors: Deev, Oleg; Plíhal, Tomáš title: How to calm down the markets? The effects of COVID-19 economic policy responses on financial market uncertainty date: 2022-01-11 journal: Res Int Bus Finance DOI: 10.1016/j.ribaf.2022.101613 sha: 6900609629dd5f37effc9bb4ed2dc71623028443 doc_id: 821385 cord_uid: wo3t5izq Financial markets during the COVID-19 pandemic are characterized by a prolonged period of increased uncertainty. In this paper, we analyze how the announcements of policy interventions and responses, to buffer short-term economic impact of the pandemic and offset financial turmoil, have affected the level of realized volatility in 23 countries. Under the augmented heterogeneous autoregressive model framework, we show that the international calming effect of COVID-19 economic policy actions originates from the US macroprudential policy announcements. The COVID-19 pandemic brought an unprecedented level of extreme volatility onto financial markets (Ali et al., 2020; Baker et al., 2020) last seen more than a decade ago during the global financial crisis. The increased volatility is attributed to the information on the number of cases and fatalities globally (Zhang et al., 2020; Ashraf, 2020b; Baig et al., 2021) and in the US (Albulescu, 2021), reproductive number (Díaz et al., 2022) , the non-pharmaceutical interventions (such as school and workplace closing, cancelled public events, closed public transport, public information campaigns, restrictions on internal movement, and international travel controls) (Zaremba et al., 2020; Ashraf and Goodell, 2021; Bakry et al., 2021) , investor sentiment (Huynh et al., 2021) and simply fear (Lyócsa et al., 2020) . Extreme sensitivity of financial markets due to media scrutiny and fears of disastrous economic conse-10 quences and associated negative impacts on corporate profitability pushed markets into a high-volatility, low-price state (Mamaysky, 2021) . To offset possible negative effects of staled economic activities and market fears, governments, central banks and supervisory authorities around the world have started an unprecedented amount of policy interventions and responses. The combined policy reaction does not only aim to calm the financial turmoil (similar to the global financial market) but also to prevent temporary disruptions from inflicting permanent damage to the economy (Alberola et al., 2020). Since March 2020, governments have intensified their fiscal policy actions to buffer the short-term impact of the COVID-19 economic shock. The wide array of fiscal measures to support households and firms include loans and guarantees especially to SMEs, wage and employment subsidies, tax cuts and capital injections to strategic firms. In accordance with their mandates, central banks have promptly responded by concerns that price dislocation can cause significant damage to central players and thereby trigger financial crises (Bevilacqua et al., 2021) . Prudential authorities have supported the flow of credit to firms, households and governments by relaxing banking market constraints on the use of capital buffers and liquidity. Many actions have complimented each other creating an unprecedented complex stray of actions to decrease the uncertainty. The aim of the paper is to empirically examine how economic policy actions have affected the financial market uncertainty around the world during the onset of the COVID-19 pandemic. We analyze the impact of the announcements of approved actions in fiscal, monetary and macroprudential policies on the realized 30 volatility in 23 countries. Realized volatility is a common proxy of asset-market-base uncertainty in the financial literature due to free availability of high-frequency high-quality data for a large number of national stock markets 1 (e.g. Andersen and Bollerslev, 1998; Andersen et al., 2001 Andersen et al., , 2003 Andersen et al., , 2007 Corsi, 2009; Corsi and Renò, 2012; Bollerslev et al., 2018; Cascaldi-Garcia et al., 2021) . To estimate the effect of government action announcements on the overall level of volatility, we use an augmented model of the standard realized volatility heterogeneous autoregressive model (RV-HAR) of Corsi (2009). To our knowledge, the paper is the first complex study to evaluate the impact of broad government economic actions to relieve the consequences of the COVID-19 pandemic on the stock market's volatility. Previous studies have hinted on the possibilities of policy actions to offset the economic and financial impact of COVID-19 (Topcu and Gulal, 2020; Zhang et al., 2020) . The limited perspective of government support 40 announcements on market returns might be found in Ashraf (2020a) . He shows that the announcements of government income support and debt/contract relief for households programs are likely to weaken the stock markets' negative reaction to the growth in COVID-19 confirmed cases. We also address cross-country variations in market reactions to government interventions (one day before, during the day of the announcement, and one day after) basing our analysis on 28 stock market indices in 23 countries, the spillover impact of US and EU authorities' actions on other markets and the importance of specific interventions, namely asset purchases, credit facilities, liquidity enchantments, fiscal stimulus and macroprudential policies. Our analysis broadly relates to two strands of literature: the financial implications of the COVID pandemic and the impact of economic policies on stock market risk. The pandemic as an extraordinary event 50 brought a lot of attention of researchers in financial economics depicting the related global uncertainty (Baker et al., 2020) . Miescu and Rossi (2021) find that the COVID-induced shocks and structural uncertainty shocks are highly correlated and generate qualitatively and quantitatively comparable dynamic responses of key fi-1 For the purposes of our analysis, the alternative measures of market uncertainty have clear disadvantages. Uncertainty measures often require individual options data, which are not publicly available. Related implied volatility measures represent expectations of market participants about the volatility development over the remaining life of the underlying options (usually 30 days). Instead, the focus of this analysis is on immediate spot market reactions based on high-frequency information. For a comprehensive review of alternative uncertainty measures see Cascaldi-Garcia et al. (2021) nancial and economic indicators (under the structural vector-auto-regression framework). Wang and Wang (2021) provides evidence for sharply declining market efficiency of S&P 500 Index during February-March 2020. Vera-Valdés (2021) shows that volatility measures of 21 stock indices are characterized by growing long memory parameters following the pandemic. Kizys et al. (2021) demonstrate herding behaviour in stock markets during the pandemic while also showing that the stringency of both non-pharmaceutical and financial response mitigates investor herding behaviour by way of reducing multidimensional uncertainty. For instance, Demir and Danisman (2021) show that the COVID-related reduction in bank stock prices is 60 mitigated by several government response policies, such as income support, debt contract relief and fiscal measures. There are significant regional differences in the magnitude and impact of the COVID-related uncertainty. Harjoto et al. (2021) emphasise the differences in market reactions to COVID-19 cases and mortality rates in developed and emerging countries. In the investigation of 47 national stock markets, Engelhardt et al. (2021) show that the stock markets' volatility during the COVID-19 pandemic is significantly lower in high-trust countries (as measured by World Values Survey, where respondents declare confidence in government and societal trust). Szczygielski et al. (2021) demonstrate within the ARCH/GARCH framework the resilience of Asian markets, while European, North and Latin American markets experience a weakening of the uncertainty over time (the analysis covers the pandemic period till June 2020). On the other hand, Bakry et al. (2021) show that government stringency actions increase volatility in emerging and decrease it in developed markets. Within the second strand of literature, there is a profound evidence that macroeconomic measures and announcements about such measures have an impact on financial markets. For instance, Caporin and Poli (2017) show that macroeconomic news announcements provide useful information for volatility forecasting of S&P100 firms. Fiordelisi and Galloppo (2018) document the reactions of 12 stock exchanges around the world to monetary and fiscal policy announcements. Huang (2018) suggests that news announcements are significant determinants of disagreement and uncertainty in stock and bond futures. Moreover, Collingro and Frenkel (2020) show that financial market participants respond more strongly to monetary policy after the Global Financial Crisis and that policy communication is more effective in countries that had previously 80 faced a severe economic downturn. Thus, developed markets might be more susceptible to policy actions than emerging markets. While most of literature focuses on the effects of scheduled macroeconomic news announcements, policy announcements during the initial phase of pandemic were not expected. The economic understanding behind the impact of unanticipated monetary policy action on stock prices is provided in Bernanke and Kuttner (2005) . The event study of Heyden and Heyden (2021) evaluates the impact of first announced fiscal and monetary policy measures on the abnormal returns of US and European firms during the initial phase of the pandemic. The findings implicate that fiscal policy potentially adds to uncertainty among investors, while responses from central banks can have a reassuring character and help to calm markets. The event 3 J o u r n a l P r e -p r o o f study of Rahman and Al Mamun (2021) on Asia Pacific financial markets indicates that government stimulus packages calmed the markets as well. Klose and Tillmann (2021) provide an aggregate empirical analysis of the responses of European financial markets (via changes in stock and bond yields) to policy announcements in the spring of 2020. Fiscal stimulus announcements are found to be leading to higher yields, while the effect is heterogeneous depending on the severity of the pandemic. Moreover, simultaneous announcements of fiscal and monetary policy actions are particularly effective. On the other hand, Wei and Han (2021) suggest that the emergence of pandemic has significantly weakened the transmission of both conventional and unconventional monetary policies to financial markets. The later documented heterogeneity of policy responses among countries might be a result of previous policy implementations, both conventional and unconventional. In fact, in investigation of the monetary policy reaction function of central banks during the pandemic, Yilmazkuday (2021) shows that emerging 100 markets or countries without a zero bound on their interest rates were able to reduce interest rates, whereas advanced economies or countries with a zero bound on their interest rates were not. Recent studies also focus on the effect of Economic Policy Uncertainty on market volatility (Tiwari et al., 2019; Antonakakis et al., 2013) . Pre-COVID evidence suggests that policy responses in one country (especially the US) might affect the uncertainty in other countries. For instance, Mei et al. (2018) show that economic policy uncertainty in the US helps predict the volatility of the European stock markets. (Bekaert et al., 2013; Miranda-Agrippino and Rey, 2020) document international spillovers of Fed conventional policies. In a broader study, Chen et al. (2016) find that US unconventional monetary policy has spillover effects to both advanced and emerging economies. In this regard, the COVID-19 pandemic represents an interesting case study into the economic policy -volatility relationship both nationally and cross-border. For instance, Chinese stocks showed a higher degree of volatility sensitivity to economic policy uncertainty during COVID-19 lockdown (Yang and Yang, 2021) . In the during COVID study, Bevilacqua et al. (2021) evaluate the impact of Fed policy interventions on stock market fear in the US and internationally. They found that market liquidity, foreign exchange policies and macroprudential policies have significant impact on the risk term structure derived from daily option prices. In contrast, credit to households, businesses, and the public sector had no discernible impact on market fear in the main US stock market index. Authors believe that the market seems to either have expected those policies and priced them in or regarded them as inconsequential for asset prices. The most obvious channel of international spillovers of US policies is the Fed's US dollar swap lines, which are aimed to Such actions normally leading to lower returns were during pandemic associated with higher stock returns. We perform our analysis on 28 stock indices from 23 countries around the world. Table 1 presents the extensive list of stock indices. The analyzed period ranges from January 2020 to the end of July 2021. The volatility data for all indices are obtained from the Realized Volatility Library provided by Oxford-Man 130 Institute of Quantitative Finance 2 . These data are freely available and are based on high-frequency data from the Thomson Reuters DataScope Tick History database. From available volatility metrics, we choose the five-minute realized variance (RV), which is a standard in academic literature. Our choice is also supported by findings of Liu et al. (2015) , who compared almost 400 different volatility estimators and concluded that it is challenging to beat 5-minute RV significantly. Realized variance is formally defined as follows: where r t,i represents intra-day asset returns, and N is a number of returns on a given day (sampling frequency). Moreover, RV is annualized (assuming 252 trading days), and, as commonly practiced, we also apply the logarithmic transformation of this measure of variance (e.g. Andersen et al., 2001 Andersen et al., , 2003 Andersen et al., , 2007 Corsi and Renò, 2012; Taylor, 2017) . We refer to this variation measure as realized volatility or simply volatility. were no actions approved. Table 2 indicates how many days included at least one action announcement for each analyzed country. The actions in the dataset are divided into 13 categories. Commonly, more than one action was announced each day. Therefore, the total number of days with action in the last row is not a sum of all the rows above but the total number of days with any action announcement. Here we also divide policy announcements into two sub-periods. The main focus is on the first half of 2020, from January 20 to the end of July 2020. This period is characterized by a large number of policy actions. Moreover, as shown in Figure 1 , it is connected to the unprecedented levels of volatility due to overall uncertainty about economic consequences of the pandemic. The second selected period (August 2020 -July 2021) consists of a significantly lower number of the announcements mostly related to fiscal 160 policies. 4 Moreover, after summer of 2020, most of policy action announcements were related to the renewal or prolongation of the already existing measures, thus, can largely be considered as anticipated. To reflect our initial observations on volatility levels and the number of policy announcements we primarily focus our analysis on the first wave of the COVID-19 pandemic, accompanied by the most uncertainty and panic of March-April 2020. During this period, all policy announcements were not anticipated by the market. After this period, the situation has settled, while market participants became aware of the broad range of implemented responses. We also add two months before and after this period to include some calmer periods into the model and get a longer time series. Therefore, the main results of our study are based on the period between January and July 2020. To fulfill our goal, we apply the heterogeneous autoregressive model (HAR) developed by Corsi (2009). The model is easy to estimate, offers a good economic interpretation, and allows the inclusion of other regressors. Moreover, it provides an outstanding performance in-and out-of-sample (e.g., Corsi, 2009; Kourtis et al., 2016; Horpestad et al., 2018) . The model is specified as follows: where RV D t is the realized variance for the previous day, RV W t is the average realized variance for the last 170 five trading days (one week), and RV M t is the average realized variance for the past twenty-two trading days (one month). The selection of the model components should represent investors' behavior with different investment horizons (Müller et al., 1997) . 5 The model is estimated using ordinary least squares, where standard errors are obtained via heteroskedasticity-and autocorrelation-consistent estimator of Newey and West (1994) . Table 2 includes only two categories of announcements due to the overall low number of observations in other categories. 5 The HAR model specification is also supported by Audrino and Knaus (2016), who applied the least absolute shrinkage and selection operator (LASSO) to recover the lag structure of the HAR model. The out-of-sample forecasting evidence shows an equal performance of the HAR model and the LASSO approach. Note: The table indicates the number of days with at least action divided by 13 categories. The last row shows the total number of days with any action. It is common that during the same day more than one action were announced. We add policy actions announcement to the HAR model as interactive terms. All dummy variables in the model are multiplied by (RV D t ) to control for the level of volatility. Otherwise, during the period of low volatility, the dummy variables could have a much higher impact that during the period of high volatility. Therefore, our model with variables representing actions is defined as follows: For each stock index, we include interactive terms Act t+1 , Act t , Act t−1 that indicates whether there was some action announced in the corresponding country. We include the Act variable with three different time indices. Act t+1 represents the actions performed one day after the day for which we explain volatility. From another perspective, it explains what happen with volatility one day before the action takes place. It could indicate some leakage of information about government claims before the official announcement of 180 the specific action. Act t stands for actions that happened during the day of announcement. It contains the information on whether the markets reacted immediately to new information. The last component is Act t−1 , which shows the action declared one day before the day of interest. It is suitable for predictive regression and indicates how the markets reacted the next day after the action was announced. As a first step, we investigate the statistical properties of our dependent variables represented by the volatility of stock market indices (Table 3) We next test the calming effect of policy action announcements and provide evidence of their impact on market uncertainty. The results of our full model estimation for each stock index are presented in Table 4 . RV D , RV W , and RV M are the main components of the HAR model, and they fulfill the role of control variables in our analysis. In almost all models, RV D and RV W are positive and highly statistically significant. These two components are the most important for explaining the realized volatility of each day. Logically, the latest information (from the previous day and week) has the highest explaining power. On the other hand, RV M has lower importance in most models while still statistically significant most of the time. The coefficient is negative suggesting a mean reversion of volatility during the observed period. Analyzing the coefficients of interactive terms around the day of action announcements, our results confirm that the volatility increases during the day of the action announcement (Act t ). The coefficients are 210 positive and significant in almost half of the analyzed stock indices with the highest reaction for SPX (0.072). Logically, volatility rises during the day of the announcement because all market participants update their expectations according to the new information. The more important effect is whether the volatility goes down back to the previous levels, which means (10)). Kurt. is a measure of unbiased kurtosis obtained using Fisher's definition of kurtosis (kurtosis of normal distribution = 0) and the result is normalized by N-1. Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelation-consistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1. that the situation calms down quickly. The calming effect on the markets is present mainly in the United States. All stock indices in the US exhibit lower volatility than during the announcement day. Moreover, the magnitude of the coefficients is higher than during the previous day. Therefore, the volatility decreases more than it increases the day before. We only observe similar effect in Singapore and Australia, where the volatility declines one day after the actions were announced. The opposite effect of decreasing volatility after the action announcement is found in three European 220 countries, namely Belgium, Netherlands, and Norway, which have positive and statistically significant coefficients of Act t−1 . It indicates that stock indices in these countries react negatively to policy actions because their volatility increases the next day after the action. Some significant results for the variable Act t+1 are found in several countries. The highest positive coefficients are observed for India, Japan, and Australia, where volatility tends to increase before the action announcement. This might indicate the significance of public discussion on possible policy actions, information leakage before the official announcement or overall increased policy uncertainty of market participants. Surprisingly, the volatility in India and Japan is not decreasing even after the announcement day. The other interesting anomaly is Switzerland, where volatility decreases before the announcement without the subsequent increase in uncertainty on or after the announcement day. As a next step, we investigate which types of actions are the most important (also the form of robustness check for the previous results). According to Table 2 , most actions belong to two categories, Fiscal Stimulus and Macroprudential Policy. Therefore, we include actions from only one of these categories into our model estimation. The results are shown in Tables 5 and 6. In biggest European stock markets, the fiscal stimulus mostly tends to increase the volatility during the announcement day, while the macroprudential policy effect is not significant. After the announcement, the volatility increase is observed in both cases with limited statistical significance. One day before the announcement, the fiscal stimulus actions did not affect or slightly decrease the volatility in Switzerland and Germany. In comparison, the macroprudential policy actions mostly increase volatility. In America, the picture is exactly the opposite of Europe. Most of the statistically significant coefficients are negative. In the case of fiscal stimulus actions, the volatility slightly rises one day before the action for indices DJI and RUT. After the action, the volatility declines with a higher magnitude for DJI, ISIC, and SPX. For RUT, the coefficient is also negative but is not statistically significant. The macroprudential policy provides a different stronger effect. All coefficients are positive and statistically significant on the day of the action, while the following day all coefficients are also significant but negative with a higher magnitude. It indicates that the macroprudential policy actions were able to calm down the markets in the US. The Canadian stock market seems to be more sensitive to fiscal stimulus and provides no statistically significant reaction to macroprudential policy announcements. The analysis of Asian and Australian stock markets provide mixed results. It seems that the fiscal stimulus announcement has a higher effect one day before and on the day of the action. On the other hand, macroprudential policy shows the most significant impact one day after the action with a predominantly calming effect. In general, our results suggest that the effects of policy action announcements are stronger for developed countries than emerging markets, where the magnitude of policy effects and its statistical significance tend to be lower. Given the previous evidence on the impact of the US monetary policy on other markets and similar development of volatility in the studied markets (Figure 1 ), we also test how policy action announcements in the US and Euro area affect uncertainty levels in other markets. To achieve this goal, we include additional 260 regressor US Act t−1 or EU Act t−1 into the baseline model specification. These variables represent the adoption of some action in the US or EU in the previous day. We include only lagged action variable to keep the number of regressors reasonable and emphasize the effect of causality. Table 7 reports the results for the effect of action announcements in the United States on other countries. Additional variable US Act t−1 is highly statistically significant and negative in most European countries. It indicates a strong spillover effect of the US policy announcements on European stock markets. Similar to the calming effect of action announcements in the US, the uncertainty in Europe decreases. Same reaction is observed in Canada. On the contrary, Asian and Australian markets are not affected by the United States' actions. Table 8 presents the spillover effect that stems from the action announcement in the Euro area. The EU 270 action announcements do not influence the volatility in the United States and three other American markets. The increase of uncertainty after EU policy announcements is only found for the Swiss and two Asian stock markets (Indian NSEI and Singapore). Overall, the influence of the EU actions is minimal and much weaker than the US. To complement our main analysis, we also investigate the subsequent one-year period from the beginning of August 2020 till the end of July 2021. All tables with results are available in Appendix A. In general, the power of economic policy announcements to affect market uncertainty has decreased significantly and evident only for few countries. The diminishing effect is the most obvious in the United States, where compared to previous results (Table 4 ) overall announcements lost their power to influence stock market volatility. In the separate analysis of fiscal and macroprudential policy actions, many countries did not announce any new policies during the selected time period (depicted by zeros in regression coefficients for action dummy variables). For fiscal policy, the results are mixed and generally weaker than during the first half of 2020, Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelation-consistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1 . Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelation-consistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1 . Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. US Act. t−1 represents the actions from the US from the previous day, also multiplied by (RV D t ). R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelation-consistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1 . Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. EU Act. t−1 represents the actions from the EU from the previous day, also multiplied by (RV D t ). R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelationconsistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1. especially for the United States. Results on macroprudential policy announcements confirm our previous findings of the calming effect in the US and Asian countries. Nevertheless, we also observe a difference in the results of spillover effects. The United States lose its power to calm European markets, as the US news is not statistically significant for almost all cases. Surprisingly, EU actions (mostly related to fiscal stimulus) seem to gain significance in the US stock market and increase volatility in blue-chip indices of S&P500 and NASDAQ perhaps due to global nature of business of included companies. In this paper, we study market volatility during the COVID-19 pandemic and its reaction to economic policy announcements in 23 countries. Policy announcements crucially affected the financial market uncertainty in the initial phase of the pandemic. In at least 13 countries, the COVID-19 economic policy announcements increase financial market uncertainty at the day of the announcement. The local calming effect of policy announcement is strongly evident only for the US, when the decrease in market volatility the day after the announcement is higher than the increase in volatility at the announcement day. The market uncertainly in the US reacts mostly to the macroprudential policy announcements (with no statistically significant reaction to fiscal stimulus actions). Some local reactions to fiscal stimulus actions are only observed in Europe and Canada. Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelation-consistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1 . Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelation-consistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1 . Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. US Act. t−1 represents the actions from the US from the previous day, also multiplied by (RV D t ). R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelation-consistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1 . Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. EU Act. t−1 represents the actions from the EU from the previous day, also multiplied by (RV D t ). R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelationconsistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1 . Covid-19-induced shocks and uncertainty Us monetary policy and the global financial cycle. The Review of 410 Economic Studies Volatilities of different time resolutions -analyzing the dynamics of market components Automatic lag selection in covariance matrix estimation Outlasting the pandemic: Corporate payout and financing decisions during covid-19 How resilient are the asia pacific financial markets against a global pandemic? Pacific-Basin Finance Journal 69 The only certainty is uncertainty: An analysis of the impact of covid-19 uncertainty on regional stock markets Realised variance forecasting under box-cox transformations The policy uncertainty and market volatility puzzle: Evidence from wavelet analysis The impact of covid-19 on emerging stock markets The persistence of financial volatility after covid-19 Covid-19 and financial market efficiency: Evidence from an entropy-based analysis The impact of covid-19 pandemic on transmission of monetary policy to financial markets Economic policy uncertainty, covid-19 lockdown, and firm-level volatility: Evidence from china Covid-19 and monetary policy with zero bounds: A cross-country investigation Note: a, b, c, d in superscript denote significance at the 15% 10%, 5%, and 1%, levels, respectively. The values in bold show all statistically significant coefficients at the 15% level. Const. represents a constant. RV D is realized volatility from the previous day, RV W and RV M is the average realized volatility from the previous week (5 days) and month (2 days) respectively. Act t+1 , Actt, Act t−1 are dummy variables multiplied by (RV D t ). It represents action that were performed after, during, or before each day, respectively. R 2 represents R-squared. The models are estimated using ordinary least squares (OLS) and the standard errors are obtained via heteroskedasticity-and autocorrelation-consistent (HAC) estimator (Newey and West, 1994) . List of countries and stock indices is presented in Table 1 .