The COVID-19 pandemic, policy responses and stock markets in the G20

This paper analyses the impact of the Covid-19 pandemic on stock market returns and their volatility in the case of the G20 countries. In contrast to the existing empirical literature, which typically focuses only on either Covid-19 deaths or lockdown policies, our analysis is based on a comprehensive dynamic panel model accounting for the effects of both the epidemiological situation and restrictive measures as well as of fiscal and monetary responses; moreover, instead of Covid-19 deaths it uses a far more sophisticated Covid-19 index based on a Balanced Worth (BW) methodology, and it also takes into account heterogeneity by providing additional estimates for the G7 and the remaining countries (non-G7) separately. We find that the stock markets of the G7 are affected negatively by government restrictions more than the Covid-19 pandemic itself. By contrast, in the non-G7 countries both variables have a negative impact. Further, lockdowns during periods with particularly severe Covid-19 conditions decrease returns in the non-G7 countries whilst increase volatility in the G7 ones. Fiscal and monetary policy (the latter measured by the shadow short rate) have positive and negative effects, respectively, on the stock markets of the G7 countries but not of non-G7 ones. In brief, our evidence suggests that restrictions and other policy measures play a more important role in the G7 countries whilst the Covid-19 pandemic itself is a key determinant in the case the non-G7 stock markets.


Introduction
It is well known that financial markets are affected by external events such as natural disasters and environmental developments (see, e.g., Caporale et al., 2019). They also respond to pandemics, as already seen in the case of the Severe Acute Respiratory Syndrome (SARS) and Ebola Virus Disease (EVD) outbreaks. For instance, Chen et al. (2007Chen et al. ( , 2009) employed an event study approach and found a negative impact of SARS on tourism and the wholesale and retail sector in Taiwan, but a positive one on the biotechnology sector, which meant that it was still possible to adopt profitable investment strategies by rearranging portfolios. Ichev and Marinc (2018) used both event study and regression methods and found that the Ebola outbreak affected mainly stock markets closer to the birthplace of their volatility; finally, the fact that it also allows for heterogeneity by providing additional estimates for the G7 and the remaining countries separately. The layout of the paper is as follows. Section 2 outlines the econometric framework. Section 3 describes the data and the construction of the Covid-19 index. Section 4 discusses the empirical results. Section 5 offers some concluding remarks.

Modelling framework
As stated before, the aim of the empirical analysis is to investigate the effects of the Covid-19 pandemic and of policy responses on stock market returns and volatilities. For this purpose, a dynamic panel data model with fixed effects is estimated which takes the following form 2 .
x i,t = α + βx i,t− k + hCovid19 Indexi,t− 1 + θFiscal Policyi,t− 1 + φ z i,t− 1 + e t (1) where x i,t stands in turn for stock market returns and volatility for country i at time t at both the monthly and daily frequency. An autoregressive structure is allowed with up to one lag (k = 1) for monthly data and five lags (k = 5) for daily data; insignificant lags are dropped. h and θ measure the impact of the Covid-19 index (Covid19_Index) and of fiscal policy (Fiscal_Policy) measures respectively on stock market returns (or volatility). z t-1 is a vector including the exogenous variables described in Section 3, namely a stringency index, lockdown measures, and short-term shadow rates.
Various model specifications are estimated. The Covid19_Index and Fiscal_Policy measures (our main variables of interest) are included in all cases. Model 1 and 2 examine their impact on stock market returns and volatilities. The set of regressors includes in turn a 0-1 dummy for lockdown measures (Lockdown) and a stringency index (Stringency_Index) (0-100) as possible determinants. Model 3 adds an interaction variable between the Covid-19 index and lockdown periods (i.e., . Both sets of models are estimated using monthly and daily data in turn. Finally, we control for heterogeneity by also performing the analysis separately for the G7 countries and the other countries in the sample. The estimated coefficients with the associated robust t-statistics are presented in Tables 3-6.

Data sources and description
This section describes the variables included in the econometric model, specifically stock market return and volatility (the dependent variables), a Covid-19 index and a fiscal variable, and also a set of exogeneous variables including a stringency index, a dummy for lockdown measures and the short-term shadow interest rate as a proxy for monetary policy responses.

Stock markets returns and volatility (dependent variables)
We use stock market returns (Stock_Return) and volatilities (Stock_Volatility) in turn as the dependent variables. Both series have been obtained at the daily (for working days) and monthly frequencies from Bloomberg. The sample period goes from March 2, 2020 to February 17, 2021 to match the Covid-19 data (see the following section 3.2). The list of all G20 stock market indices considered is displayed in Table 2, panel B. Stock market volatility (σ n ) is calculated as the realized volatility: We set Δt (the increment by time period) as one working day or one month for the daily and monthly frequencies respectively. P t is the stock market index at time t, where t stands for either the day or the month. n is the nth day or month at the point of estimation for the corresponding parameter. μ n is the estimated drift parameter (i.e., the realized mean).

The Covid-19 index
The source for the Covid-19 data is Our World in Data (https://ourworldindata.org/coronavirus), from which we collect the following daily series for the 20 main economies in the world (G20) 3 : new deaths from Covid-19 per million (new_deaths), intensive care unit (ICU) Covid-19 patients per million (icu_patients), hospitalized Covid-19 patients per million (hospital_patients), new Covid-19 tests per thousand (new_tests), and population for each country (population) between January 1, 2020 and February 18, 2021. The reported figures concern events that happened one day before, and thus the actual sample to consider goes from December 31, 2019 to February 17, 2021. Further, we remove the data for weekends when daily deaths, hospitalized patients, testing, etc., are normally lower because of delayed or missing Covid-19 reports. Then we obtain a balanced panel for the period from March 2, 2020 to February 17, 2021.
We create a Covid19_Index i,t based on the population weighted daily infection rate (Weighted_Infection i,t : share of the population Table 1 List of countries.
Note: × denotes our inclusion of the corresponding country and stock index in G20, G7 or Non-G7 countries sampling.
The following table shows the list of G20 countries and their corresponding stock indices used in our analysis. (population) newly infected by the Coronavirus on each day (new_cases)) and severity (Severity i,t : a daily measure of the relative health situation of that population) for country i at day t: We use a Balanced Worth (BW) methodology Villar, 2018, 2020) to measure Severity on the basis of the different possible outcomes of Covid-19 infections including new_deaths, icu_patients, hospital_patients and new_tests categories. 4 We evaluate Severity for various populations affected by the virus, G = {1,2, …, g} over a set of health conditions C = new_deaths, icu_patients, hospital_patients, new_tests ordered from worst to best. a j,c = nj,c nj is the share of people within population j with health condition c. n j and n j,c are the number of individuals in population j and those with health condition c resulting from the virus, respectively.
We then calculate the probability p j,k that an individual of population j exhibits a worse health condition than one of population k, with the health condition categories being ordered from worst to best: e j,k = e k,j is the probability of a tie between individuals of population j and k. Accordingly, we define the probability q j,k of an individual of population j being under a worse health condition than one in population k as follows: The following table shows the summary statistics for the monthly data for the G20 (Panel A), G7 (Panel B) and non-G7 (Panel C) countries. Stock returns (Stock_Return) and volatility (Stock_Volatility) are calculated as percentage returns and realized volatility, respectively, according to section 3.1. Fiscal policy (Fiscal_Policy) is the additional spending and forgone revenue) as a percentage of GDP. The stringency index (Stringency_Index) is a composite measure based on 9 response indicators (e.g., school closures, workplace closures, and travel bans) ranging between 0 and 100 where higher value indicates stronger restriction. Shadow short rate (Shadow_Short_Rate) is the short-term policy rate at the zero lower bound (zero or slightly negative) value. Lockdown (Lockdown) (new_tests), and population for each country (population). We show he mean, median, standard deviation (Std.), 25th percentile (25th per), 75th percentile (75th per) and total number of observations (N).
Note that p j,k + p k,j + e j,k = 1. Then the severity measures for the two populations j and k (s j and s k , respectively) are proportional to the corresponding probabilities of being relatively worse off, namely: This pairwise severity comparison between two populations can be extended to a comparison among more than two populations by taking expectations as follows: In equation (8), the numerator is the average relative Covid-19 severity of population j with respect to the rest, and the denominator is the average relative Covid-19 severity of the populations other than j compared to population j.
The vector of s j severity values is the BW which measures the relative severity of Covid-19 for different populations. This is obtained as the dominant eigenvector of a Perron matrix M: The Perron matrix M columns add up to (g − 1) and if M is irreducible this implies the existence, positivity and uniqueness of the BW vector Villar, 2018, 2020). In our analysis, each country i uses its own collection of populations G while the vector of s j severity values based on the BW method above is used to produce Severity i,t . We have implemented the algorithm from the Ivie website http://web2011.ivie.es/balanced-worth/to obtain the BW vectors.
In order to make our Covid-19 index comparable across the globe with a normalized figure between zero and one, we use a min-max normalization to create NCovid19_Index i,t as follows: Table 6 Correlation matrixdaily frequency. The following table shows the Pearson's correlation matrix between the daily frequency regressors for G20 (Panel A), G7 (Panel B) and non-G7 (Panel C) countries. ***, ** and * denote significance at 1%, 5% and 10% levels, respectively.
The Min(Covid19 Index) and Max(Covid19 Index) are the minimum and maximum Covid19 Index, respectively, across our sample period and countries.
We then apply the Christiano-Fitzgerald filter (Christiano and Fitzgerald, 2003) to smooth the normalized Covid-19 index (NCo-vid19_Index i,t ) and calculate the trend component (CF NCovid19 Index i,t ) using band-pass approximations: We isolate the trend component CF NCovid19 Index i,t with minimum and maximum oscillation periods p l and p u , respectively, where 2 ≤ p l < p u < ∞. We set p l = 2 and p u = 5 to allow the oscillation period to be between minimum two and maximum five days, respectively, as our daily data excludes weekends. The process CF NCovid19 Index i,t has power only in frequencies in the interval {(a, b) ∪ (− b, − a)}∈ (− π, π). The process NCovid19 Index i,t has power only in the complement of this interval in (− π, π). a and b belong to the interval 0 < a ≤ b ≤ π and are related to p l and p u by The random walk filter approximation of CF NCovid19 Index i,t is Ĉ F NCovid19 Index i,t computed as follows: where the filter weights are as below: The Christiano-Fitzgerald filter is suitable for different data frequencies and its random walk assumption optimizes the approximation better than other filters including the Hodrick-Prescott, Baxter-King ones and the Trigonometric Regression (Christiano and Fitzgerald, 2003;Baum, 2006). We display the computed CF_NCovid19_Index for each of the G20 countries in Fig. 1 where the absolute values on the y-axis are comparable across countries as they are already normalized using equation (10). For instance, one can see that this index peaked at 0.6 in the US compared to 0.03 in Australia and 0.00015 in China. 5 The various peaks in individual countries (for instance, two in Italy and three in South Korea) clearly correspond to different Covid-19 waves.

Fiscal policy measures
For the fiscal support measures taken by national governments in response to the Covid-19 pandemic the source is the International Monetary Fund (IMF)'s database of fiscal policy responses to Covid-19. Specifically, we collect the above_the_line_measures (i.e., additional spending and forgone revenue) as a percentage of GDP at three points in time, namely June 12, September 11 and December 31 in 2020. 6 For the sample period from January 1 to February 17 in 2021 we use extrapolated data.

Stringency index and lockdown measures
The Stringency_Index is collected from Our World in Data (https://ourworldindata.org/coronavirus) along with the other Covid-19 data. This index is a composite measure based on 9 response indicators (e.g., school closures, workplace closures, and travel bans) ranging between 0 and 100 where higher values indicate stricter measures. We then collect the lockdown dates from (1) the Global Covid-19 Lockdown Tracker in Aura Vision (https://auravision.ai/covid19-lockdown-tracker), (2) the Covid-19 Government Measures Dataset in ACAPS (https://www.acaps.org/covid-19-government-measures-dataset) and (3) various online news articles. Common dates across these three lockdown data sources are selected to create a lockdown dummy variable for each country which is equal to one for the lockdown periods and zero otherwise at the daily frequency, and equal to one if any date within the corresponding month includes the lockdown period and zero otherwise at the monthly frequency.

Short-term shadow rates
We use the short-term shadow rates (Shadow_Short_Rate) for each sample country to investigate the impact of monetary policy during the Covid-19 pandemic. These have been chosen as a quantitative measure of the overall stance of monetary policy when the conventional policy instrument (the short-term policy rate) is at the zero lower bound (zero or slightly negative valuesee Kuusela and Hännikäinen, 2017). We use the Morgan Stanley reported shadow short rates for the countries for which they are available, and the US one as a proxy in the other cases. Table 1 shows the list of G20 countries and the split between G7 and non-G7. Table 2 reports the sources and descriptions for the variables used to construct the Covid-19 Index (Panel A) and the others including fiscal policy, the stringency index, lockdowns and short-term shadow rates (Panel B).
Tables 3 and 4 display summary statistics for both the daily and monthly and data, winsorized at the 1st and 99th percentiles. The CF_NCovid19_Index indicates that the Covid-19 pandemic has affected more severely the G7 countries, where there have been more frequent lockdowns (Lockdown) 7 but less stringent restrictions (Stringency_Index) as well as a stronger fiscal stimulus (Fiscal_Policy) and lower shadow rates (Shadown_Short_Rate) compared to the non-G7 countries. Further, during the Covid-19 pandemic, the G7 countries experienced lower stock returns (Stock_Return) whilst the non-G7 countries exhibited higher stock market volatilitiy (Stock_Volatility). Finally, the correlation matrix for the monthly (Table 5) and daily ( Table 6) series implies that there are no multicollinearity issues.

G20 countries
The estimates from the dynamic panel data model with fixed effects given by equation (1) indicate that the impact of the Covid-19 pandemic (CF_NCovid19_Index) has decreased stock market returns whilst increased stock market volatility in all G20 countries ( Table 7). As already explained, our CF_NCovid19_Index is a composite BW measure of Covid-19 severity comprising related new deaths (New_death), intensive care unit admissions (Icu_patients), hospitalizations (Hospital_patients), and Covid tests (New_tests), which are weighted by the infection rate (New_cases) per population in each country (Population). Our results for stock market returns are consistent with the negative effect of Covid-19 confirmed cases and total deaths previously found for the Chinese stock market Fig. 1. Covid-19 indices for the G20 countries. 7 Given the way the lockdown dummies are constructed (see Section 3.4) we base our comparison on the daily variable. (Al-Awadhi et al., 2020), and the negative impact of Covid-19 related health news on the stock returns of the 20 worst hit countries reported by the Center for Disease Control and Prevention (CDC) as of March 30, 2020 (Salisu and Vo, 2020). The estimated increase in stock market volatility resulting from the Covid-19 pandemic is also in line with the conclusion reached by Baker et al. (2020) and Albulescu's (2020) according to whom this has increased global financial uncertainty proxied by the VIX. The finding that during lockdown periods corresponding to particularly severe Covid-19 conditions (CF_NCovid19_Index × Lockdown) stock market returns are lower is consistent with the results of Davis et al. (2021) indicating that the reduction in economic activity caused by lockdowns has a negative effect on returns, especially during periods when the epidemiological situation is at its worst. A Covid-19 related fiscal stimulus in the form of additional spending and forgone revenue (Fiscal_Policy) has a positive impact on stock market returns. 8 Government restrictions (Stringency_Index) including school closures, workplace closures, and travel bans during the Covid-19 pandemic reduce returns in a service-oriented economy as already found by Baker et al. (2020). The shadow short rate (Shadow_-Short_Rate), a proxy for a near-zero central bank policy rate during unconventional monetary policy periods (Krippner, 2020), is estimated to have a significant negative impact on stock market returns and a positive one on volatility. This finding confirms the importance of including this measure of the monetary policy stance during period characterised by near-zero interest rates since conventional rates, for instance, could account for at most one third of the V-shaped trajectory of the stock market rebound in mid-March of 2020 and could not explain the drop in stock prices during the Covid-19 pandemic periods (see Cox et al., 2020). Table 8 reports the results for the G7 countries. In this case there appears to be a significant negative impact on stock market returns of government restrictions (Stringency_Index) rather than the severity of Covid-19 (CF_NCovid19_Index), whilst both increase stock Note: CF_NCovid19_Index × Lockdown is the interaction term controlling for the effect of the Covid19_Index during lockdown periods only (Lockdown = 1).

G7 countries
The following table shows the Covid-19 impact (CF_NCovid19_Index) on stock returns (Stock_Return) and volatility (Stock_Volatility) for the G20 countries based on monthly (Panel A) and daily (Panel B) frequency data. We use the dynamic panel regression model with fixed effect including an autoregressive term AR(1) to generate these results. We report the F-statistics, R 2 and number of observations (N). The p-values are in the brackets. ***, ** and * denote significance at 1%, 5% and 10% levels, respectively. 8 The fiscal variable captures the effect of the fiscal stimulus announcement only, not the subsequent transmission to the real economy. market volatility. In other words, measures such as mandatory business closures, commercial activity restrictions and social distancing rather than the Covid-19 severity itself seem to have made stock prices plunge during the pandemic. The same conclusion was reached by Baker et al. (2020), who pointed out that even the much higher excess mortality rates of previous Spanish Flu  and influenza pandemics (1957-58 and 1968) only left mild traces on stock markets whilst restrictions normally have a significantly more pronounced effect. Lockdowns during periods of severe Covid-19 conditions (CF_NCovid19_Index × Lockdown) mainly affect stock market volatility as opposed to returns. Covid-19 related fiscal policy measure (Fiscal_Policy) are effective in boosting stock market returns without increasing volatility. By contrast, a higher shadow short rate (Shadow_Short_Rate) appears to have a negative impact on stock returns while increasing volatility. Table 9 shows the estimates for the non-G7 countries. Unlike in the previous case, for this subgroup the severity of Covid-19 (CF_NCovid19_Index) not only increases volatility but also reduces returns significantly. Lockdowns under severe Covid-19 conditions (CF_NCovid19_Index × Lockdown) also have both those effects and so do restrictions such as workplace closures, travel bans, social distancing, etc. (Stringency_Index). However, a fiscal stimulus (Fiscal_Policy) only increases stock market volatility. According to Auerbach et al. (2021), although such measures are useful in the event of a slump, their marginal effect on the economy decreases with higher inequality, and in fact the average Gini coefficient for the non-G7 countries (41.78) is higher than for the G7 ones (34.27) (see Appendix I), which supports this argument. The near zero policy rate (Shadow_Short_Rate) is not very effective either in boosting returns but unlike the fiscal measures does not increase volatility. Finally, all results (for the G20 as a whole and the two subgroupssee Tables 7-9) are robust across the two frequencies, daily and monthly (see Panels A and B respectively), in the sense that the coefficients signs (though their significance) are the same.  (1) to generate these results. We report the F-statistics, R 2 and number of observations (N). The p-values are in the brackets. ***, ** and * denote significance at 1%, 5% and 10% levels, respectively.

Conclusions
This paper examines the impact of the Covid-19 pandemic on stock market returns and their volatility in the case of the G20 countries. In contrast to the existing empirical literature, which typically focuses only on either Covid-19 deaths or lockdown policies, our analysis is based on a comprehensive dynamic panel model accounting for the effects of both the epidemiological situation and restrictive measures as well as of fiscal and monetary responses; moreover, instead of Covid-19 deaths it uses a far more sophisticated Covid-19 index based on a Balanced Worth (BW) methodology (see Villar, 2018, 2020), and it also takes into account heterogeneity by providing additional estimates for the G7 and the remaining countries (non-G7) separately.
Our analysis produces a number of interesting findings and confirms the importance of distinguishing between different sets of countries. In particular, whilst for the G20 as a whole it would appear that the epidemiological situation has had a significant impact on both stock market returns and volatility (negative and positive, respectively), the estimation for the G7 and non-G7 subgroups reveals some key differences between these two sets of countries. Specifically, we find that the stock markets of the G7 are affected negatively by government restrictions more than the Covid-19 pandemic itself. By contrast, in the non-G7 countries both variables have had a negative impact. Further, lockdowns during periods with particularly severe Covid-19 conditions have decreased returns in the non-G7 countries whilst increased volatility in the G7 ones. Fiscal and monetary policy (the latter measured by the shadow short rate) have had positive and negative effects, respectively, on the stock markets of the G7 countries but not of non-G7 ones. In brief, our evidence suggests that restrictions and other policy measures have played a more important in the G7 countries whilst the Covid-19 pandemic itself has been the key determinant of stock market movements in the non-G7 economies during the period in question, the implication being that the focus should be on measures directly affecting the economy in the G7 and instead on ameliorating the epidemiological situation in the non-G7 ones. Note: See notes Table 7.
The following table shows the Covid-19 impact (CF_NCovid19_Index) on stock returns (Stock_Return) and volatility (Stock_Volatility) of the non-G7 countries based on monthly (Panel A) and daily (Panel B) frequency data. We use the dynamic panel regression model with fixed effect including an autoregressive term AR(1) to generate these results. We report the F-statistics, R 2 and number of observations (N). The p-values are in the brackets. ***, ** and * denote significance at 1%, 5% and 10% levels, respectively.