Anxiety about the pandemic and trust in financial markets

Abstract The COVID-19 pandemic has generated a novel context of global financial distress. This paper enters the related scientific debate and focuses on the relationship between the anxiety felt by the population of a wide set of countries during the pandemic and the trust in the future performance of financial markets. Precisely, we move from the idea—grounded on some recent literature contributions—that the volume of Google searches about “coronavirus” can be considered as a proxy of anxiety and, jointly with the stock index prices, can be used to produce indicators of the population mood—in terms of pessimism and optimism—at country level. We analyse the “very high human developed countries” according to the Human Development Index plus China and the main stock market indexes associated with them. Namely, we propose both a time-dependent and a global indicator of pessimism and optimism and classify indexes and countries accordingly. The results show the existence of different clusters of countries and markets in terms of pessimism and optimism. Moreover, specific regimes emerge, with optimism increasing around the middle of June 2020. Furthermore, countries with different government responses to the pandemic have experienced different levels of mood indicators, so countries with less stringent lockdown measures had a higher level of optimism.


Introduction
The world has experienced the rapid and dramatic widespread of COVID-19 Zhu et al. 2020)-a pandemic generated by a coronavirus-with millions of infected and a large number of deaths. Beyond the sanitary aspects of such an infectious disease, one of the main concerns experienced by communities regards the economic impact of the measures taken for contrasting the virus (see, for example, Siddique et al. 2022, where authors have analysed the role of regional poverty during the COVID-19 pandemic in the USA).
The financial distress we have observed in the international stock marketswhose entity has been much more evident during the so-called first wave of the pandemic, in the period February-June 2020, can be reasonably interpreted also through the anxiety of the people, whose worries for the pandemic affected the expectations of financial markets' future performance.
The term anxiety in our setting demands some words to contextualise its meaning. For us, anxiety (for the pandemics) stands for a feeling of worry and/or fear about the future uncertain evolution (of the pandemics). In this respect, anxiety is, for us, synonymous of fear and worry. It consequently generates pessimism, i.e. the tendency to feel that the most negative scenarios (in the context of the pandemic, in our case) will occur. Conversely, a positive view of the future is associated with optimism.
This paper enters this debate. Specifically, it explores how the anxiety expressed by the population about COVID-19 mirrors the strategies of investing/disinvesting capital in financial markets, here represented by major stock market indexes. In particular, we discuss the relationship between anxiety about COVID-19 and the view of financial markets, aiming to investigate optimism and pessimism.
Consistently, we focus only on the first wave of the pandemic; indeed, empirical evidence suggests that financial distress is remarkably evident at the beginning of COVID-19 diffusion (see, Deb 2021where a focus on the industry of airlines is presented). The analysis deals with the country-level moods relaying on Zimmermann et al. (2020)'s conclusions that human factors should be monitored and considered at the outbreak in such a globalised world. We explore the relationship mentioned above for a large set of countries to derive the different behaviours of the populations. Fetzer et al. (2021) and Binder (2020) are remarkably relevant for contextualizing our study. The authors discuss the economic anxiety stemming from the coronavirus. Binder (2020) conducts a survey study of over 500 US consumers and shows that the serious concern about coronavirus implications leads to pessimistic expectations about macroeconomic turnaround via deterioration of the economic fundamentals. Fetzer et al. (2021) complement Binder (2020)'s perspective by also including the time dimension and the causal effect of the pandemic on the increased economic anxiety. The methodological ground of Fetzer et al. (2021) lies in the meaningfulness of Google Trends data, which is assumed to give in-depth information on the development of anxiety in the specific context of the economic outcomes (for additional supportive pieces of evidence that relates web searches and population anxiety see Rovetta and Castaldo 2020;Monzani et al. 2021;Halford et al. 2020). We adopt Fetzer et al. (2021)'s view and hypothesise that anxiety about COVID-19 is proxied by the irrepressible persistence of related web searches (on the significance such a type of data, see Cinelli et al. 2020;Choi and Ahn 2020 in the former an analysis of the infodemic is presented, in the latter the Google Trends data are used in an influenza spread forecast model). In so doing, we also follow Mertens et al. (2020), where a survey-based study over a large number of respondents confirms that media exposure and online searches are good predictors of the increasing fear of coronavirus (in this, see also the review paper by Garfin et al. 2020). Additionally, in Hisada et al. (2020), the authors have underlined the relevance of online searches in predicting emerging COVID-19 clusters of infections.
In detail, we collect and compare two datasets over the same reference period, from January 6, 2020, to June 19, 2020. On one side, we consider the daily Google Trends data. Specifically, we examine the search volumes of the word "coronavirus" along with its translations for different countries' most spoken languages. Data retrieved at a country level allow for sounding out similarities and discrepancies in the search for information practised by users in need of awareness. In our approach, such compulsive searching is intended as a proxy for the anxiety generated by the pandemic. On the other side, we consider the daily levels of the main stock indexes, including companies related to the countries. The source of financial data is Eikon -Datastream (in line with studies such as Lewis andBozos 2019, andChizema 2010). In order to have a reliable and consistent dataset (in terms of countries' features of interest), countries are chosen by using the Human Development Index (HDI) used by the United Nations Development Programme (UNDP) in the Human Development Report Office to rank countries on the basis of their human development. Specifically, we select areas with an HDI index greater than 0.8, calculated with the 2018 information. The choice of 0.8 as a threshold is appropriate because all countries with at least that level can be considered as "very high human developed countries" (see, UNDP 2019for additional information on the definition). It ensures a good enough level of connections between socio-financial entities within the countries. Namely, it guarantees the incorporation of the necessary links between citizens' cognitions of the problems, ability to get informed about them, access to the relevant resources and financial strategists presence (this choice is in line with the findings presented in Chundakkadan and Ravindran 2020about the relevance of the access to online sources to increase the response capacity of a country). Indeed, the starting point of our study is that the data on Google Trends about searches offers a reliable description of how people search for information in a given country. This statement is valid where access to the web is widespread and granted to citizens. Therefore, it is not valid for less developed countries such as those classified as "Low human development" in UNDP (2019), where a large portion of the population has poor internet access for causes such as lack of infrastructure, devices or limited digital literacy. Thus, including all the world's countries would lead to a biased analysis, with some underdeveloped countries represented by the "privileged club" of people living at the highest standards. Therefore, taking the subset of countries corresponding to those under the locution "very high human developed countries" makes the analysis less biased and more rigorous. We add China, which is ranked below the 0.8 threshold -specifically, 0.75, to such a list of nations. We reasonably do so because China is central to the phenomenon under investigation. Moreover, the countries without data on stock exchanges in our source, Datastream, have been obviously excluded from the list.
Our work departs from Fetzer et al. (2021) in two keys respects: first, the quoted paper deals with topics detected in Google Trends, and we deal with one crucial word, "coronavirus". In so doing, we have a translation task to face, as acknowledged by Fetzer et al. (2021). Nevertheless, the use of one word allows us to obtain intuitive results and is far from being restrictive in our context (we are also in line with Baig et al. 2020;Goodell and Huynh 2020). Indeed, a preliminary inspection of the Google Trends data shows that the considered word is the most relevant related to the studied pandemic; second, the quoted paper directly derives information about economic anxiety from Google Trends. Differently, we here start from the idea that the anxiety is manifested through the Google searches of the word "coronavirus" (and its translations); in doing so, we differ from Chundakkadan and Ravindran (2020), and we are in line with Bento et al. (2020) where such a keyword is employed. After that, we move to stock indexes' performances to assess the links with financial markets and with the trust in them.
Indicators synthesising the considered time series have been suitably introduced in this work to offer a broad perspective on the connections between the variables. We conceptualise such indicators focusing on specific dates and offering global information on the entire reference period. All the proposed indicators range in the unitary interval [0, 1], making the comparative analysis of different countries possible.
Several interesting results emerge. Countries and stock indexes can be clustered in terms of their resulting mood during the first wave of the pandemic period. Regularities and deviations at individual week levels can also be identified. Moreover, the analysis of the daily variations of the levels of anxiety and trust in financial markets gives insights about countries' behaviours in the period. A general trend of pessimism was concentrated in early, and mid-March 2020 when many countries adopted the lockdown, and the international community started to gauge the problem's severity. A focus on some noticeable cases of hard and weak lockdown policies has also been presented. In this respect, countries with a stricter lockdown had a more persistent and higher level of pessimism.
The obtained findings can be placed in a wide literature strand on the pandemics' effects on people's moods. In this respect, we mention, e.g. Francisco et al. (2020); Hennessy et al. (2021); Kolakowsky-Hayner et al. (2021); Yuksel et al. (2021). However, to the best of our knowledge, the analysis proposed in the present study is the first one giving an overview of the mood of citizens for a large set of countries and proposing time-varying and global studies of such a mood. Indeed, for example, Francisco et al. (2020) restrict the analysis to children and adolescents whilst we consider the entire population in each country. The authors describe the case of three countries (Spain, Italy and Portugal) and explore only different ways to measure pessimism. From a different perspective, Hennessy et al. (2021) offer an analysis based on the music as a device for measuring the mood. The authors refer to the early phases of COVID-19 in India, the UK, the USA and Italy. They carry out an interesting study on a small sample of individuals. Kolakowsky-Hayner et al. (2021) insist on a gender-based analysis of the mood for 59 countries but offered aggregate results based on a small sample of individuals. Finally, Yuksel et al. (2021) deal with several countries and implemented a survey on the quality of sleep, certainly related to mood during the pandemic. Also, the sample is small in their case, which is utterly appropriate for a survey-based study.
The rest of the paper is organised as follows. Section 2 discusses some key contributions on the roots of the anxiety for a pandemic and its links with the stock indexes' performance. Section 3 presents the employed dataset by also providing details on the data collection procedure. Section 4 illustrates the indicators used for the study. Section 5 outlines and discusses the analysis results. The last section concludes.

Literature review
The individuals' behaviours, attitudes and choices are at the core of the interest of many scientific studies given that those are the ground for a deep understanding of the economic patterns; this is even more relevant when peculiar social settings occur, such as those realised as a consequence of the pandemic. For example, sadly, social interactions represented a threat in the context of a pandemic spreading, see, Xiong et al. (2020). In this respect, Bonacini et al. (2020) discusses the effectiveness of the lockdown policies in the paradigmatic case of Italy, while social distance and freedom restrictions are the basis of Qiu et al. (2020) and Venter et al. (2020), the former provides an exploration of the influence of contagion in nearby cities in China and the latter estimates the improvement in air-pollution deriving from actives reduction indirectly caused by the pandemic. The quoted papers suggest pointing attention to the evidence that several businesses require physical interactions among the involved actors-and such interactions have been reduced by the lockdown policies and by the natural attitude of people avoiding possible sources of contagionwhile virtual connections allow another set of economic relevant activities, such as investing in financial markets. In Danisman and Tarazi (2020), the authors consider the "uncertain prospects after the COVID-19 pandemic" as a premise for including new financial technologies through fintech as a response taken in the financial sector. Zahra (2020) discusses the uncertainty of the post-COVID-19 world and the role of innovation activities in international entrepreneurship initiatives. Similarly, Dias et al. (2020) discusses changes in the online learning environment that is having disruptive innovations and changes worldwide.
In Oldekop et al. (2020), the authors remark that the global development paradigm is based on three main factors, and the first mentioned is "the interconnectedness of contemporary capitalism" across countries and its permeation with global development. This point constitutes the theoretical ground for understanding the increasing interest in financial markets' performance and catastrophes. Goodell (2020) provides a brief discussion on the financial markets reactions to rare catastrophic events of non-financial nature. The author points the readers to the plausible parallelisms between pandemics and natural disasters, terrorist attacks and even nuclear conflict. Some features of the markets manifested in such cases have been outlined by Lyócsa and Molnár (2020), associating Google searches and S&P 500 returns and volatility. Less recently, Kaplanski and Levy (2010) elaborate on how aviation disasters can generate a decline in related stock prices. Goel et al. (2017) treat the special case of terrorist attacks exploring the vulnerability of financial markets to terrorist incidents. In general, empirical evidence proves that prices collapse in concomitance to rare and unexpected disasters (see, e.g. Barro 2006;Gabaix 2012;Gourio 2012). On the same line, but from a broader perspective, several authoritative studies highlight that anxiety and negative mood might increase investors' risk aversion, hence leading to the collapse of stock prices (see, e.g . Ariel 1990; Kamstra et al. 2000Kamstra et al. , 2003Cohen-Charash et al. 2013). Interestingly, in Ho and Wyer (2023), a focus on the relationship between risk-taking, optimism and pessimism is presented, and in Buchheim et al. (2022), referring to the German context, the authors state that "firms incorporate this sentiment [optimism and pessimism] regarding the shutdown duration in their more general business outlook", confirming the conceptual framework according which the mood and the economics and financial expectations interacted, affecting each other during the pandemic.

Data 1
We now present the employed data. As we will see in detail below, the considered dataset is associated with Google Trends and the stock indexes at the country level. As a premise, stock indexes data are not always available; moreover, some countries have regions and territories whose inhabitants have limited resources to gather information from the web. In these circumstances, the validity of the Google Trends data for the intended purposes is questionable because the detected volumes of searches may not be representative of the entire population but just of a set of more privileged citizens. To avoid such sources of bias and inconsistency, we focus on a qualified set of countries whose data provide a good description of the situation of their inhabitants. At this aim-and for providing a consistent analysis-we have used the Human Development Index (HDI) adopted by United Nations Development Programme (UNDP)'s Human Development Report Office as the criterion for selecting the countries to be investigated. Indeed, HDI is a composite index made of factors like life expectancy, education, per capita income indicators, and other relevant factors whose details are recollected in Ul Haq (1995) by Mahbub ul Haq, one of the two designers of the index. HDI is used to rank countries on the basis of human development. More specifically, we take all the countries defined as "very high human developed countries", namely those having an HDI index greater than 0.8. The selection is based on data from 2018, Table 1 of UNDP (2019). China is added to the considered countries-even if the HDI of China is 0.75-because of its centrality in the COVID-19 propagation; the first known human infections were in China.
Employing Google Translate, the word "coronavirus" is translated from English to the equivalent word in the most used language in each of the considered countries. In so doing, we obtain the translations reported in Table 1.
The translated terms are employed to query the web search indicator from Google Trends. Namely, for each country, one looks for the index of search of the "coronavirus" translations in the language awarding the largest number of speakers. The period investigated captures the first wave of the pandemic; it goes from January 6, 2020, to June 19, 2020.
At the end of this process, one gets a time series matrix regarding 63 countries. In our analysis, we are interested in examining the Google Trend search indicator 1 The data that support the findings of this study are available from the corresponding author upon request from the first day a relevant search volume is recorded in each country; i.e. on the first day in which Google Trends offers a non-null value for the translated terms. See columns one, two and three of Table 1 and Fig. 1 for an idea of the main trends in the data. The most noticeable point is the high volume of searches around mid-March 2020. We associate at least one stock index with each country on the abovementioned list. Per each index, the closing prices are downloaded from Thomson Reuters Datastream. The period is the same adopted for collecting the Google Trends data (see Table 2 and Fig. 2) so that one has the same amount of data. Andorra, Bahamas, Barbados, Belarus, Brunei, Liechtenstein, Palau, Seychelles and Uruguay do not have a stock market index of reference in our data source, so we exclude them because we need to have data points for both the variables under consideration. The final list of considered countries contains 54 elements. Furthermore, we align the Google Trends and financial data so that the volume of web searches can be used in the analysis for each day in which prices are recorded. Indeed, as we will see, the indicators proposed in the paper are grounded on the joint observation of the Google Trends data and the stock indexes' performance. Therefore, it is necessary to work on these two quantities considering only the days when Google Trends data and the stock indexes' performance are registered and can be jointly elaborated. More than this, a biased and incomplete analysis would be the outcome of the study of only one of such factors. Hence, reducing the Google Trends data for having a shared time frame with the stock indexes' performance ones lets the study be free of biases and suitable to pursue the intended aim. In this respect, we notice that Google searches data are available daily. In contrast, financial data are available only when the financial markets are open, i.e. during trading days, typically not on weekends or other special dates. This explains why the Google Trends data are reduced. As a reference for the number of observations, one can look at column "N. Obs." in Table 2.

Indicators
To face the problem, we design indicators that capture the connection between anxiety about the pandemic and the outcomes of financial markets. The underlying idea relates to the synchronicity between increments and decrements of Google searches and stock index levels so that increasing (decreasing) volumes of searches and decreasing (increasing) prices are associated with pessimistic (optimistic) moods. Thus, optimism and pessimism are measured by combining the analysis of Google searches and stock indexes' performance. Namely, the connection between optimistic and pessimistic phases and the evolution of the financial markets are captured, including the assessment of bullish and bearish periods. One intuitively expects the mood indicator to lean towards optimism during a bullish period (or towards pessimism in a bearish one). However, including the Google Trends index in our proposed indicators means that a bullish (or bearish) period can only be associated with optimism or pessimism after a jointly analysing financial performances and Google searches. This joint analysis provides a clear proxy for anxiety about the pandemic. This can be clarified further by looking at the formal presentation of the indicators below and reflecting on their functioning mechanism. The employed methodology can be described after some notation is introduced.
We denote the number of considered countries by J-and J is 54 for us, see Sect. 3-and label the generic country by j = 1, … , J . Each country is associated with K stock indexes. The number of stock indexes depends on the selected country, so one should write K = K(j) . Such a dependence will be omitted when unnecessary i.e., when there is only one stock index of reference for that country. Often, K > 1 -i.e. most countries are associated with more than one stock index. However, there are cases of countries with K = 1 . The generic stock index is k = 1, … , K.
As already discussed in Sect. 3, we have daily data on prices and Google searches of the word "coronavirus" (and its translations) in a common reference period of T days. For country j, we denote the available time series of the prices of the stock index k by p j k = (p j k (1), … , p j k (T)) . Analogously, the sample of the Google searches for country j is w j = (w j (1), … , w j (T)).
Notice that the range of variation of the components of p j k and w j is different. Indeed, p j k has non-negative components without a pre-defined ceiling, while the components of w j are integer numbers ranging in [0, 100], and there exists t such that w j (t) = 100 . Time t represents the day with the maximum level of searches over the period [1, T] and depends on j. Also, such dependence will be conveniently omitted. The minimum value of the elements of w j is not necessarily null. Indeed, null search means the absence of interest for the considered word in the country ji.e. null amount of Google searches; such an occurrence does not necessarily appear over the period [1, T]. Assigning value 100 to the highest daily magnitude of Google searches over [1, T] and null value to null searches allows a easy normalisationimplemented directly by the Google Trends proprietary algorithm-of the Google search data in the range [0, 100].
For facilitate comparisons, we impose the variation range [0, 100] also to the series p j k for each j and k through a simple normalisation procedure. We denote the normalised series of the prices by p j k . First, we identify t ∈ {1, … , T} such that p j k (t) = max{p j k (t) ∶ t = 1, … , T} . Then, we set p j k (t) = 100 . Null price is associated with zero value for the normalized The analysis and comparison of the normalised financial data and Google Trends index is performed at the country level. It is implemented by conceptualising suitable indicators that provide several insights into countries' regularities and discrepancies as presented in the next sections.

Time-dependent indicators
We first propose an indicator based on the comparison between the time-dependent normalised accumulations of prices and Google searches. We consider is a period accounting for a high percentage of the price of index k and a low percentage of Google searches-where percentages have to be intended in terms of the total amount on the overall period. 2 Thus, A j ([t 1 , t 2 ];k) close to one means that [t 1 , t 2 ] is an optimistic period. Differently, A j ([t 1 , t 2 ];k) is close to zero when fraction of prices are relatively low, and Google searches of the word "coronavirus" are relatively high. In this case, [t 1 , t 2 ] is a time interval where country j has experienced anxiety about COVID-19 and a lack of trust in index k.
Notice that the case t 1 = 1 and t 2 = T is trivial and not interesting, being A j ([1, T];k) = 1∕2 for each j and k-i.e. in the middle (fair) situation between optimism and pessimism. Indeed, [1, T] is the entire period, hence is associated with the full percentages of prices and Google searches. More reasonably, the proper selection of t 1 and t 2 allows exploring elements of the considered sample in relevant sub-periods.
At a country level, we can average the A j 's in Eq.
(2) with respect to the stock indexes. In particular, we define We observe that A j ([t 1 , t 2 ]) ∈ [0, 1] , and all the comments reported above remain valid for the indicator presented in Eq. (3).

Global indicators
We here compare the considered series on the basis of the signs of their daily variations. Precisely, we assess how often an increase (a decrease) in Google searches is associated with a reduction (an increase) in the stock indexes prices. The entity of the daily variation is also taken into account. Consistently with our framework, we refer hereafter to a generic series x = (x(1), … , x(T)) , whose components range in [0, 100].
Thus, given a threshold ∈ [0, 100] and t = 1, … , T − 1 , we define the series x variation's sign between t and t + 1 at the threshold as follows: The parameter is decided a priori; it represents the entity of the daily variation to be crossed for declaring that the series have an increase (or a decrease, by taking the variation with negative sign) from time t − 1 to time t. Evidently, the case = 0 leads to (0) The comparison between the behaviours of the Google searches and the stock indexes can be performed at the country level, employing the 's defined in Eq. (4) and using the two series on interested instead of the generic x.
For each j = 1, … , J and k = 1, … , K(j) , we compare the series w j with p j k . We define By definition, the Δ 's in Eq. (5) can take values in {−2, −1, 0, 1, 2} . Such values have specific meanings to be mapped in the optimism and pessimism setting.
When Δ ( ) (t, j, k) = −2 , then we observe a decrease in the Google searches related to "coronavirus" and an increase in the price of the stock index k. This case has a straightforward interpretation in terms of optimism. Indeed, people exhibit decreasing anxiety about the pandemic disease-they weaken the number of searches on Googleand simultaneously exhibit an increasing interest in investing in the stock index. The value -1 is associated with constant Google searches and an increase in the price or decreasing level of Google searches and an invariant price. The value 0 is related to the cases of identical behaviour between Google searches and price so that they can be invariant between date t and t + 1 or both can increase/decrease. The value +1 relies on an increasing level of Google searches and invariant price or a constant level of Google searches and decreasing price. The value +2 describes the situation in which Google searches grow and price decrease. This is the other corner case associated with pessimism, in which anxiety about the spread of the disease-mirrored by the growth of Google searches-is associated with decreasing investments in the stock index.
In general, the positive values of the Δ 's describe situations of pessimism, captured by anxiety for the disease and decreased investments in the stock indexes. Conversely, the cases of negative Δ 's are related to optimism, with decreasing interest in COVID-19 and growing attention to the future evolutions of stock indexes, investing in them.
Some indicators with high information content are derived from Eq. (5). We measure the aggregated connection between the considered trends in Google searches and the price of stock index k in country j over the considered period by defining If such an indicator approaches zero, then people in country j tend to be at the highest level of optimism-in a sense expressed when the case of Δ = −2 was discussed-when analysing the Google searches of the considered word and its connections with stock index k. The converse situation appears when H ( ) j (k) is close to one, namely when we are in the presence of a high level of pessimism.
By averaging the H j 's in Eq. (6) with respect to k, we obtain an indicator describing the mood at the country level for all the connections between the considered word searches and the prices of stock indexes, as follows: Clearly, H ( ) j ∈ [0, 1] and the arguments above-opportunely cascaded for a country level view-remain valid.
We now provide a measure describing how a country has experienced optimism versus pessimism. At this aim, we consider a ratio indicator as follows: where By construction, R ( ) j (k) ∈ [0, 1] . For country j and stock index k, there is a high percentage of optimistic days with respect to pessimistic ones as the value of such an indicator approaches zero, while we are in a substantial context of pessimism when the indicator in Eq. (8) is close to one. The corner cases have a clear interpretation: when R ( ) j (k) = 0 , then all the days in the considered period present decreasing anxiety about COVID-19 coupled with increasing trust in the performance of stock index k; differently, R ( ) j (k) = 1 is associated with an entire period of increasing need of awareness on COVID-19 and decreasing price of stock index k.
Also in this case, we can focus on country j by averaging the R j 's over the stock indexes: Evidently, R ( ) j ∈ [0, 1] and the discussion reported above applies also in this more general case.
The global indicators presented above capture two aspects of the phenomenon under analysis. H ( ) j and H ( ) j (k) provide information on moods as an average of Δ 's over all the days of the considered sample. Differently, R ( ) j (k) and R ( ) j focus only on the dates where the daily variations of volumes of searches and stock index levels have had discordant behaviours. Namely, the indicators R's offer more details on the ratio between entirely optimistic days and wholly pessimistic ones, i.e. intuitively, on the proportion of the days in which the Google searches have decreased, and the indexes' prices have increased and those with an increase in searches and a decrease in prices.

Results and discussion
The normalised time series of the stock index prices are obtained via Eq. (1). The outcome of such normalisation is presented in Fig. 2, and the main statistical indicators of both original and normalised time series are shown in Table 2. The visual inspection of this Fig. 2 allows the reader to confirm the general trends of the stock markets, with a decline inducted by the incorporation of the pandemic effects of the first wave. Figure 1 and Table 1 show the increased Google searches of the translated "coronavirus" in different countries. The search activities started at a different time and with a general delay with respect to the decline recorded by the stock indexes.
As a preliminary comment, we notice that A j in Eq.
(2) and (3) compares the normalised values of Google searches and prices, while H j in Eq. (6) and (7), and the R j in Eq. (8) and (9) compare their daily increments and decrements. Thus, A j offers a view on anxiety about COVID-19 and trust in stock markets; differently, H j and R j propose an evolutive perspective on the daily variations of the Google search and the stock indexes data, presenting insights on a synthesised version of the mood.

Analysis of the global indicators
In computing the index A j ([t 1 , t 2 ];k) employing Eq.
(2), we take t 2 − t 1 constantly equal to five days, hence studying the weekly behaviour of the indicator. The outcomes per each stock index are summarized in Fig. 4 and Table 3. Moreover, the results of A j ([t 1 , t 2 ]) across the stock indexes of each country-namely, those in Eq.
(3)-are reported in Fig. 5 and Table 4. From this view, some facts emerge: • The paths have drastically changed between the 7th and the 8th weeks of the year, namely between 17/02/2020 and 01/03/2020. During this period, the international community started to take the situation seriously despite the controversial statements of national governments' heads. On 11/03/2020, WHO's Director declared, "WHO has been assessing this outbreak around the clock, and we are deeply concerned both by the alarming levels of spread and severity, and by the alarming levels of inaction. We have, therefore, made the assessment that COVID-19 can be characterised as a pandemic." WHO (2020). • Greece and South Korea have spent more than 90% of the analysed weeks in a quite positive mood, precisely reporting an A j ([t 1 , t 2 ]) > 0.5. • Cyprus and Iceland have experienced mild pessimism for quite a large number of weeks. They present A j ([t 1 , t 2 ]) < 0.5 at least 40% of the times in the studied period. • Weeks 10 and 11 are characterized by the lowest average of A j ([t 1 , t 2 ]) . Their means across the countries are, respectively, 0.485 and 0.483. • The highest number of countries experiencing a A j ([t 1 , t 2 ]) < 0.5 is met on week 11. During 16/03/2020 -20/03/2020, 81% of the analysed countries experienced a high volume of Google searches and low normalised prices. Therefore, a high level of pessimism is recorded. On the other hand, the tails (weeks 1-4 and 20-24) present a higher index level, with an increased presence of positivism in most countries during the most recent weeks.
In Table 5, the considered countries are week-wise ranked by using A j ([t 1 , t 2 ]) . Montenegro holds the first position for five weeks. Similarly, Greece, Iceland and Malta usually sit in the first four positions. This outcome suggests that Greece, Iceland and Malta experienced waves of optimism and pessimism; interestingly, for the quoted countries, consecutive weeks may have a large discrepancy in the ranking positions. Thus, one can say that the waves are impulsive and compulsive-perhaps, they are driven by news on the pandemic or statements of the Governments-and this leads to sudden changes in people's behaviour towards searching on Google and adjusting positions in the stock markets.
We also propose a focus on weekly rankings of some paradigmatic cases: Sweden, Iceland and South Korea-countries which experienced an "easy" lockdown, see Wikipedia (2020a, 2020b); Normile et al. (2020); Florida and Mellander (2021)-and Italy, UK, USA and China-which are countries that experienced harder lockdown. By inspecting Fig. 6, one can appreciate that the countries that have experienced an easier lockdown have spent more optimistic moods in recent weeks.
The results show some regularities in the behaviour across countries and indexes, as Figs. 4 and 5 testify. An initial phase of optimism was probably induced by sceptical statements from national governments and media agencies; in fact, the emergence has been underestimated by many people at its inception, see Colarossi (2020). Then, once the situation escalated, Google searches drastically increased (see Fig. 3), and the stock indexes reacted plausibly in the light of the lockdown policies implemented worldwide. The blue bands represent the raised pessimism in Figs. 4 and 5 in weeks 10-15. A general relief came in after that. In a few cases, the anxiety was boosted from the very beginning. This is clearly the case for Iceland, Malaysia, Malta, and, more mildly, for Singapore; see Fig. 5 and Table 5. Considering week 24th, the stock indexes and so the countries reporting the highest level of A j from Eq. (3) are Greece, Iceland and Malta, with values 0.527, 0.524, 0.523, respectively. On the other hand, those having the lowest values are Montenegro, Bahrain and Singapore, with 0.508, 0.507 and 0.504, respectively. Figure 6 offers a comparison of the weekly rank of the countries-based on A j ([t 1 , t 2 ])-having experienced an easy (upper panel) and hard (lower panel) lockdown. Countries with a stricter lockdown show more pervasive pessimistic moods than those with a weaker lockdown. In particular, one can notice the presence of common waves of optimism (low rank) and pessimism (high rank) over the considered period. Importantly, there is an evident countertendency among some countries, with opposite moods in peculiar sub-periods. Iceland, South Korea and Sweden show pessimism at the beginning of the pandemic and optimism for the rest of the period, with a spike of pessimism around weeks 15-16. The case of South Korea is significant and in line with the findings presented in Park and Chung (2021). The situation is more scattered for China, the UK, Italy, and the USA. However, there is optimism at the beginning for the UK, Italy and the USA, and substantial pessimism for all the considered countries in the last part of the period. China and Italy seem to exhibit similar trends during the latter portion of the period under study; a possible explanation can be found in the strict collaboration between such countries during the lockdown, which can be seen as the driver of a common mood. From a more general point of view, the results showed in Fig. 6 can be further considered in the light of the findings reported in Harring et al. (2021). Namely, the trust in stock markets is affected and affects the trust in government policies.

Analysis of the time-dependent indicators
Eqs. (6) and (8) indicators employ different levels of , which is the threshold used to capture the variations of the observed series daily. Specifically, we use = 0, 1, … , 50.
The results for H ( ) j (k) (see Eq. 6) are reported in Fig. 7 and Table 6. Stock indexes show quite similar behaviours in their links with the Google Trends indicator, mainly in the maximum values of H ( ) j (k) . Indeed, the variation range in the maxima is 0.502− 0.530, with stock indexes associated with Bahrain being  The averaged results at the country level obtained with the indicator represented by Eq. (7) are shown in Fig. 8 and Table 7.
Some cases are particularly interesting and can be noticed by visually inspecting the results: • Latvia, Montenegro, Norway, Denmark and Canada have a vast majority of H ( ) j > 0.5 manifesting a high average level of contemporaneous Google searches growth and stock indexes declines. Across the s used in calculating H ( ) j (k) , such an occurrence appears at least in the 90% of the cases. • Malta has 92% of H ( ) j < 0.5 , representing an average low level of decreasing Google searches and stock indexes increments at the same time. • The highest value of H ( ) j occurs in Bahrain, with 0.559, for = 0 . This finding is in agreement with those discussed already for H ( ) j (k) above • The smallest value of H ( ) j occurs in Italy, with 0.4, for = 0.
The R ( ) j (k) presented in Eq. (8) are calculated and reported in Fig. 9 and Table 8. The variation range in the maxima for the case of R ( ) j (k) is 0.5-0.565, with Bahrain's stock indexes having the highest values. Differences in the minimum values are also noticeable; the range goes from 0.421 to 0.5. The lowest value is associated with Italy's index once again. Remarkable differences appear for the stock indexes within the same country, in the specific case of R ( ) j (k) ; the USA is again one of the most remarkable examples of a wide variation range at a stock index level.
The results at the country level are presented in Fig. 10 and Table 9; they have been calculated through Eq. (9). The most relevant facts are listed below:  • Qatar has the highest percentage of s such that R ( ) j > 0.5 , namely 19.6% ; therefore, it is the country having contemporaneous increases in Google searches and decreases in stock index prices for a large number of thresholds s. Belgium, Spain and France follow, with 17.6% of the s leading to R ( ) j in the range (0.5,1]. • Greece, Malaysia, Argentina and New Zealand have the highest percentages of s such that R ( ) j < 0.5 , with the first two countries having 11.8% of the observations falling within [0,0.5) and the latest two ones having a proportion of 9.8%. • The lowest value of R ( ) j occurs in Italy, with 0.421, for = 0. • The highest value of R ( ) j occurs in Bahrain, with 0.565, for = 0.
By analysing the global indicators, the case of = 0 is the most relevant to be commented for the information carried out. The proposed indexes are sensible to the smallest daily variation in such a case. Bahrain, Malta, Israel, Cyprus, United Arab Emirates, Singapore, Oman and Japan have H ( =0) j > 0.5 . Thus, on average, these countries have experienced significant anxiety about COVID-19 and a small trust in the stock markets' future performances. Differently, Italy, Canada, Lithuania, Germany, the UK and Spain have the lowest positions, with H ( =0) j < 0.5 . In such countries, an optimistic mood is preponderant, on average. Notice that such a list of countries with "optimistic moods" are highly developed and had a noticeable spread of the pandemic. Reasonably, in those countries, people's optimism is connected to their trust in the healthcare system, financial industry, and the collaborative efforts of science in addressing the widespread pandemic.
For the case of R ( =0) j < 0.5 , the lowest positions are held by Russia, Switzerland, Lithuania, Romania, Germany and Italy. These countries have experienced a large number of days of contemporaneous decreases in Google searches and increase in stock index prices. Bahrain, Israel, Japan, Singapore, Oman, Malta and Iceland are the countries with R ( =0) j > 0.5 . Of course, results for H ( =0) j and R ( =0) j are often overlapping, and some countries confirm their general mood when the comparison between entirely optimistic days and wholly pessimistic ones is performed.  Interestingly, in places where the pandemic's consequences have been managed quite brightly, the general feelings have been more pessimistic than optimistic (see, e.g. the case of Israel).

Conclusions
The study investigates the relationship between the Google search volumes of "coronavirus" and the stock index prices. The first wave of the pandemic has been considered to include the financial distress that occurred in the prompt reaction to the initial events. The analysis is carried out at the country level. Thus, the word "coronavirus" has been opportunely translated with the appropriate language when needed. Such an analysis allows for mapping interrelationships between COVID-19 anxiety in nations and lack of trust in stock markets' future performance. These aspects are related to the uncertainty surrounding the evolution of the pandemic and expectations about its effects. In our framework, we follow Rovetta and Castaldo (2020); Monzani et al. (2021);Fetzer et al. (2021); Binder (2020) and hypothesise that anxiety is manifested via the intensity of the searches run on Google and related to the virus. The proposed indicators allow for capturing the changes in moods over timefor the case of the A j presented in Eq. (2) and (3)-and also facilitate classification of countries under a more global perspective on the overall considered period-see H j presented in Eq. (6) and (7) and R j in (8) and (9). Moreover, A j accounts for the values of Google searches and prices, while H j and R j compare the daily increments/decrements of such quantities.
To make research on a reasonably homogeneous setting and for a fair treatment of the considered dataset, we have taken into consideration only "very high human  (3), at country level. Also in this case, indents represent the differences in the starting date of the Google Trends data at country level developed countries"-i.e. those with an HDI greater than 0.8-and have added China for its relevance in the studied phenomenon. Some countries with HDI greater than 0.8 but without an associated stock index had to be removed from the list to respect the formulation of the mood indicators proposed.
The study allows a panoramic view of the evolution of the mood related to the pandemic in its first wave, jointly considering the behaviour of people and the stock markets. Furthermore, the country-level approach gives insights into similarities and discrepancies of the different populations regarding the link between anxiety about COVID-19 and the expectations about stock market performance.
In conclusion, we offer some considerations that emerged while designing this study. Those might be seen as open questions that keep the scientific debate ongoing.
Taking Google searches of the word "coronavirus" ( and its translations in suitable languages depending on the country) presents the limitation of narrowing the analysis to only one word. Even if it is a crucial word in the search data about the pandemic, we reckon it can be seen as a limitation. A wider selection of terms to be tested for creating an aggregated indicator of Google searches related to the pandemic might lead to a more comprehensive view of the pandemic's anxiety but also to its overestimation or to an equally biased recording of it. Selecting more words would increase the computational complexity of the empirical experiments while providing a less intuitive definition of mood indicators. Such complexity would also Fig. 6 A comparison of the weekly mood of the countries-based on the ranks of A j ([t 1 , t 2 ])-having experienced an easy/hard lockdown. The lower the rank, the higher the optimism experienced in that week by the countries characterised by colours Table 6 Main statistical indicators of H ( ) j (k) in Eq. (6)  be expressed by the inevitable discussion around semantic and contextual meanings triggered by the selection of words to include, and its resolution does not present straightforward answers. The employment of tools from the field of On the methodological front, other devices could be exploited to capture the population's mood during a pandemic. Referring to the literature discussed in the first    countries considered and/or use professional services to gather data in those countries (e.g. one has to have questions translated in all the languages). It is certainly an interesting research item to add to researchers' agenda whose working tools include primary data collection, maybe involving colleagues in various places of the  world. On the other hand, for researchers comfortable with secondary data collection, the exploration of the interconnections between stock markets and the evolution of COVID-19 is a very interesting challenge that would complete the view in this arena. Such exploration is of pivotal interest for grounding the reactions to the pandemic (with a focus on investment decisions) on the official data on COVID-19 and its complex interrelations with the patterns of the stock markets.
In designing the present study, we have considered these challenges and open questions; in fact, we carefully developed the indicators so that the non-explored areas in this field do not undermine the results presented.