Anxiety for the pandemic and trust in financial markets

The COVID-19 pandemic has generated disruptive changes in many fields. Here we focus on the relationship between the anxiety felt by people during the pandemic and the trust in the future performance of financial markets. Precisely, we move from the idea that the volume of Google searches about"coronavirus"can be considered as a proxy of the anxiety and, jointly with the stock index prices, can be used to produce mood indicators -- 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 their respective main stock market indexes. Namely, we propose both a temporal and a global measure of pessimism and optimism and provide accordingly a classification of indexes and countries. The results show the existence of different clusters of countries and markets in terms of pessimism and optimism. Moreover, specific regimes along the time emerge, with an increasing optimism spreading during the mid of June 2020. Furthermore, countries with different government responses to the pandemic have experienced different levels of mood indicators, so that countries with less strict lockdown had a higher level of optimism.


Introduction
The world is experiencing the rapid and dramatic widespread of COVID-19 [30,45]  The individuals' behaviours are at the core of the interest of many scientific studies given that, those behaviours are the ground of a deep understanding of the economic patterns linked to the pandemic. Sadly, social interactions represent a threat in the context of pandemic spread [43]. In this respect, [6] discusses the effectiveness of the lockdown policies in the paradigmatic case of Italy, while social distance and freedom restrictions are the basis of [35] and [38], 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 to point attention to the evidence that several businesses require the 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. For example, in [13], the authors consider the "uncertain prospects after the COVID-19 pandemic" as a premise for inclusion of new financial technologies through fintech as response taken in the financial sector. [44] discusses the uncertainty of the world post COVID-19 and the role of innovation activities in international entrepreneurship activities. [14] discusses changes in the online learning environment that is having disruptive innovations.
In [33], 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 the global development. This point constitutes the theoretical ground to understand the increasing interest for the financial markets performance and the catastrophes. [20] provides a brief discussion on the reactions of the financial markets to rare catastrophic events of non-financial nature. The author points the attention of 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 [39], relating Google searches and S&P 500 returns and volatility. Less recently, [28] elaborate on how aviation disasters can generate a decline in stock market prices. The special case of terrorist attacks is treated by [19], where the authors elaborate on the vulnerability of financial markets to terrorist incidents. In general, empirical evidence prove that prices collapse in concomitance to rare and unexpected disasters 2 [see, e.g. 3,17,22]. On the same line, but in 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. 1,26,27,11].
The financial distress we are observing in the international stock markets -whose 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 through the anxiety of people, whose worries for the pandemic disease affect the expectation of financial markets future performance. This paper enters this debate. Specifically, it explores how the anxiety at the entire countries population level for COVID-19 mirrors the strategies of investing/disinvesting money in financial markets. In particular, we discuss the relationship between anxiety for COVID-19 and the view of financial markets, with the particular aim of investigating 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 the diffusion of COVID-19. The analysis is carried out by dealing with the country-level moods relaying on [46] conclusions that human factors should be monitored and considered at the outbreak in such globalised world. We explore the relationship mentioned above for a large set of countries, to derive the different behaviours of the populations. [16] and [5] are remarkably relevant for contextualizing our study. The authors discuss the economic anxiety stemmed from coronavirus. [5] 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. [16] complement [5]'s perspective by including also the time dimension and the causal effect of the pandemic on the increased economic anxiety. The methodological ground of [16] lies on the meaningfulness of Google Trends data, which are assumed to give in-depth information on the development of the anxiety in the specific context of the economic outcomes [23]. We adopt [16]'s view and hypothesise that anxiety for COVID-19 is proxied by the irrepressible persistence of related web searches [on the significance such a type of data, see 10, 8, in the former an analysis of the infodemic is presented, in the latter the Google Trends data is used in a influenza spread forecast model]. In so doing, we also follow [31], where a survey 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 18]. Additionally, in [25] the authors have underlined the relevance of on line searches in predicting emerging COVID-19 clusters of infections.
In details, we collect and compare two different datasets over the same reference period which goes from January 6 th , 2020 to the June 19 th , 2020. By one side, we consider the daily Google Trends data. Specifically, we examine for the searches volumes of the word "coronavirus" along with its translations for different countries and respective most spoken languages. Data retrieved at a country level allows for sounding out similarities and discrepancies in the searching for information practised by users in need of awareness. In our approach, such a 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, which include companies related to the considered countries.
The source of financial data is Datastream [29,7]. In order to have a reliable and consistent dataset, countries are chosen by using the Human Development Index (HDI) embraced 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 the areas having an HDI index greater than 0.8 calculated with the 2018 information. The choice of 0.8 as a threshold is appropriate because all the countries having at least that level can be considered as "very high human developed countries". 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 and financial strategies designer presence (this choice is in line with the findings presented in [9] about the relevance of the access to online sources to increase the response capacity of a country). To such a list of nations, we add China, which is ranked below the 0.8 thresholds -specifically, 0.75. We reasonably do so because China is central in the phenomenon under investigation. Moreover, the countries without data on stock exchanges in our source -which is Datastream -have been obviously excluded from the list. In doing so, we include most of the countries included in [15] where restrictions on trade on medical supplies are discussed.
We move our steps from [16] in two main respects: first, the quoted paper deals with topics in Google Trends, and we deal with one crucial word. In so doing, we have a translation task to face -as also acknowledged by [16]. Nevertheless, the use of one word allows to obtain intuitive results and is far from being restrictive in our context [we are also in line with 2, 21]. Indeed, a preliminary inspection of the Google Trends data shows that the considered word is the most relevant trend related to the current pandemic; second, the quoted paper derives information about economic anxiety directly by 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 [9] and we are in line with [4] where such a keyword is employed. After that we move to the real performance of the financial markets, to assess the links with the trust on them.
Some distance measures between time series have been suitably introduced to offer a broad perspective on the connections between the considered data. We consider concepts of distance focusing on specific dates and offering global information on the entire reference period. All the proposed measures range in the unitary interval [0, 1], hence letting the comparative analysis of 4 different countries be possible.
Several interesting results emerge. Countries and markets can be properly clustered in terms of their mood during the first wave of the pandemic period. Regularities and deviations at an individual week level can also be identified. Moreover, the analysis of the daily variations of the levels of anxiety and trust in financial markets gives insights on countries behaviours in the considered overall period. A general trend of pessimism is concentrated in early and mid-March when many countries have adopted the lockdown, and the international community started to reckon the severity of the problem. A focus on some noticeable cases of hard and weak lockdown policies has also been presented. In this respect, countries with a stricter lockdown have a more persistent and higher level of pessimism.
The rest of the paper is organised as follows. Section 2 presents the employed dataset by also providing details on the data collection procedure. Section 3 illustrates the distances measures used for the study. Section 4 outlines and discusses the results of the analysis. The last section concludes.

Data
We now present the employed data. As we will see in detail below, the considered dataset is associated with the Google Trends and to the financial markets at the country level. As a premise, we have to say that data on financial markets are not always available; moreover, the access to the web is not a reliable issue is some realities. Thus, we focus on a set of countries whose data are unbiased and reliable. 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. 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 [36] by Mahbub ul Haq, one of the two designers of the index. HDI is used to rank countries on the basis of human development. 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 [37]. 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.
The respective most used language is associated with each of these countries. Then, by means of Google Translate, the word "coronavirus" is translated from English to each of those languages ( in chi. In so doing, we obtain the translations reported in Table 1.
The translated terms are employed to download the web searches indicator from Google Trends.

5
Namely, for each country, one looks for the searches of the respective "coronavirus" translations.
In this way, the magnitude of the searches by country is obtained employing the words translated in the country most common language. The period investigated captures the first wave of the pandemic; it goes from January 6 th , 2020 to June 19 th , 2020.
At the end of this process, one gets a matrix of time series regarding 63 countries. In our analysis, we are interested in examining the time series of the searches from the first day in which a relevant volume of researches is recorded in each country -i.e., in the first day in which Google Trends offers a non-null value -for the respective translated terms. See columns one, two and three of Table 1 and Figure 1 to have an idea of the main trends in the data. The most evident point regards the high volume of searches occurred during the same days around mid-March 2020.
We associate at least one stock market index with each country of the list mentioned above.
Per each index, the closing prices are downloaded from Thomson Reuters Datastream. The period is defined by the same criterion adopted in collecting the Google Trends data (see Table 2 and Figure 2), so that one has the same time span. 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. The final list of considered countries contains 54 elements.
Furthermore, we align the Google Trends data and the financial data so that, for each day in which prices are recorded, the volume of web searches can be used in the analysis. Consequently, because the financial markets are closed during non-trading days, Google Trends data is reduced accordingly. As a reference for the number of observations, one can look at column "N. Obs." in Table 2.

Distance Measures
To face the problem, we design indicators able to capture the connection between anxiety for the pandemic and expectations on the future 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 associate to pessimistic (optimistic) moods.
The employed methodology can be described after that some notation is introduced.
We denote the number of considered countries by J -and J is 54 for us, see Section 2 -and label the generic country by j = 1, . . . , J. Each country hosts K financial markets. The number of financial markets depends on the selected country so that one should write K = K(j). Such a dependence will be omitted when not necessary. Often, K > 1 -i.e. the most part of the  considered countries is associated with more than one financial market. However, there are cases of countries with K = 1. The generic financial market is k = 1, . . . , K.
As already discussed in Section 2, 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 are different. Indeed, p j k has nonnegative components, while the components of w j are integer numbers ranging in [0, 100], and there existst such that w j (t) = 100. Timet represents the day with the maximum level of searches over the period [1, T ], and depends naturally on j. Also, such dependence will be conveniently For better 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 bȳ First of all, we identifyt ∈ {1, . . . , T } such that p j k (t) = max{p j k (t) : t = 1, . . . , T }. Then, we setp j k (t) = 100. Null price is associated to zero value for the normalized series, so that we set p j k (t) = 0 when p j k (t) = 0. Evidently, one can have p j k (t) > 0 for each t = 1, . . . , T , so that one has p j k (t) > 0 for each t.
The entire series can be derived as follows where [•] is the integer part of the real number •.
The exploration and comparison of financial data and Google Trends will proceed at the country level; it will be implemented by conceptualising suitable distance measures, under different perspectives. In so doing, we provide several insights on countries regularities and discrepancies.

Time-dependent distance measures
We first build a distance measures based on the comparison between the time-dependent normalized accumulations of prices and Google searches. We consider t 1 , t 2 ∈ {1, . . . , T } with t 1 ≤ t 2 and define where At a country level, we can average the A j 's in (2) with respect to the markets. 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 (3).

Global distance measures
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) of the Google searches is associated with a reduction (an increase) of the prices of the financial markets. The entity of the daily variation is also taken into account.
Thus, given a threshold ζ ∈ [0, 100] and t = 1, . . . , T − 1, we define the sign variation of the series x between t and t + 1 at the threshold ζ as follows: 9 The parameter ζ is fixed a-priori; it represents the entity of the daily variation to be crossed for stating 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 ζ = 0 leads to δ (0) The comparison between the behaviours of the Google searches and of the financial markets can be performed at country level employing the δ's defined in (4).
For each j = 1, . . . , J, we compare the series w j withp j k , for each k = 1, . . . , K(j). We define By definition, the ∆'s in (5)  Some distance measures with high information content can be derived by (5).
We measure the aggregated connection between the considered trend in Google and the price of market k in country j over the considered period by defining By construction, H By averaging the H j 's in (6) with respect to k we obtain an indicator describing the reality at country level, for all the connections between the considered word and the prices of financial markets, as follows: Clearly, H (ζ) j ∈ [0, 1] and the arguments above -opportunely rephrased at country level -remain valid.
We now provide a measure of how a specific country has experienced optimism versus pessimism over the considered period. At this aim, we consider a ratio indicator as follows: where By construction, R Also in this case, we can focus on country j by averaging the R j 's over the markets: Evidently, R Namely, the indicators R's offer more details on the ratio between entirely optimistic days and wholly pessimistic ones, i.e. on the proportion of the days in which the Google searches have decreased, and the indexes prices have increased and those with an increase of searches and a decrease of the prices.

Results and discussion
The normalised time series of the stock indexes prices are obtained via Eq. (1). The outcome of such normalisation is presented in Figure 2 and main statistical indicators of both original and normalised time series are shown in Table 2. The visual inspection of this Figure allows the reader to confirm the general trends of the markets, with a decline inducted by 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 searching activities started at a different time and with a general delay with respect to the decline recorded in the financial markets.
As a preliminary comment, we notice that the Moreover, the A j 's in Eq. (2) and (3) compare the normalised values of Google searches and prices, while the H j 's in Eq. (6) and (7) and the R j 's in Eq. (8) and (9) compare their daily increments and decrements. Thus, the A j 's offer a view of the snapshots of anxiety for COVID-19 and trust in financial markets; differently, the H j 's and the R j 's propose an evolutive perspective on the daily variations of the Google search and the stock market data.
In computing the index A j ([t 1 , t 2 ]; k) in Eq. (2), we take t 2 − t 1 constantly equal to five days, hence studying the weekly behaviour of the index. The outcomes per each index are summarized in Figure 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 Figure 5 and Table 4. From this view, some facts emerge: • The paths have drastically changed between the 7 th and the 8 th weeks of the year, namely between 17/02/2020 and 01/03/2020. This is the period during which 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." [40].
• Greece and South Korea have spent more than 90% of the analysed weeks in a quite positive mood, namely reporting an A j ([t 1 , t 2 ]) > 0.5. 13 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 the period 16/03/2020 -20/03/2020, 81% of the analysed countries experienced a high volume of Google searches and a low level of normalised prices. Therefore, a high level of anxiety/pessimism. On the other hands, the tails (weeks 1-4 and 20-24) present a higher level of the index, with an increased presence of positivism in most of the countries during the most recent weeks.
In Table 5  We also propose a focus of weekly rankings of some paradigmatic cases: Sweden, Iceland and South Korea -the countries with an easy lockdown, see [41,42,32] -and Italy, UK, USA and China -which are countries having or having had a harder lockdown. By inspecting Figure 6, one can appreciate that the countries having experienced an easier lockdown have spent more optimistic moods during the recent weeks.
The results show some regularities in the behaviour across countries and indexes, as Figures 4 and 5 clearly testify. An initial phase of optimism was probably induced by sceptic statements from national governments and media agencies; in fact, the emergence has been underestimated by a large number of people at its inception, see [12]. Then, once the situation has escalated, Google searches have drastically increased (see Figure 3), and the markets have simultaneously reacted, plausibly also in the light of the lockdown policies implemented all over the world. The raised pessimism is represented in Figure 4 and 5 by the blue bands in weeks 10-15. A general relief came in after that. In a few cases, the anxiety boosted from the very beginning. This is clearly the case for Iceland, Malaysia, Malta and more mildly for Singapore, see Figure 5 and Table 5.  with a stricter lockdown seem to show more pervasive pessimistic moods than those with a weaker lockdown globally. 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 subperiods. 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 week 15-16. The case of South Korea is incredibly meaningful and in line with the findings presented in [34]. For China, UK, Italy and USA the situation is more scattered, but there is optimism at the beginning for UK, Italy and USA, a substantial pessimism of all the considered countries in the last part of the period. China and Italy seem to follow analogous patterns in the late part of the period; 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 Figure 6 can be further considered in lights of the findings reported in [24], namely, the trust in financial markets is affected and affects the trust in government policies.
Some cases are particularly interesting and can be noticed by visual inspecting the results: j (k), such an occurrence appears at least in the 90% of the cases.
• Malta have 92% of H (ζ) 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 (ζ) j occurs in Bahrain, with 0.559, for ζ = 0. This finding is in agreement with those discussed already for H (ζ) j (k) above.
• The smallest value of H (ζ) j occurs in Italy, with 0.400, for ζ = 0.
The R (ζ) j (k) in Eq. (8) are reported in Figure 9 and Table 8. The variation range in the maxima for the case of R The results at country level are presented in Figure 10 and Table 9; they have been calculated through Eq. (9). The most relevant facts are listed below: • Qatar has the highest percentage of ζs such that R Interestingly, we find that in places where the pandemic has 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 of different markets. To include financial distress in a proper way, the first  -having experienced an easy/hard lockdown. The lower is the rank, the higher is the optimism experienced in that week by the respective country.
wave of the pandemic has been taken into account. The analysis is carried out at the country level.
Thus, the word "coronavirus" has been opportunely translated, 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 [16] and hypothesise that anxiety is manifested via the intensity of the searches run on Google related to the virus.
The proposed indicators allow to capture changes in moods along the time -for the case of the A j 's in (2) and (3) -and permit also classification of markets and countries under a more global perspective on the overall considered period -see the H j 's in (6) and (7) and the R j 's in (8) and (9). Moreover, the A j 's compare the values of Google searches and prices, while the H j 's and the    The study allows having a panoramic view of the evolution of the pandemic in its first wave, its effects on the behaviour of people and its impact on financial markets. Furthermore, the country-level approach gives insights on similarities and discrepancies of the different populations in respect of the link between the anxiety for COVID- 19 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  1  ARE ARE LUX SGP MLT LUX ESP MLT CYP KOR ISL  MLT ISL  MNE ISL  ISL  GRC MNE GRC GRC MNE MNE MNE GRC  2  JPN QAT HRV NOR ESP ESP AUT KAZ GRC GRC KOR GRC MYS MLT MYS GRC MLT GRC ISL  ISL  CYP GRC CYP ISL  3  EST JPN BGR HRV NOR ITA  LUX RUS MNE OMN GRC KOR CHN MYS ISR  FIN ISL  MLT LVA MLT GRC ISL  GRC MLT  4  EST DEU ESP ITA  AUT KAZ GRC MLT MNE NOR MNE MNE ISR  FIN ISR  LVA ISL  FIN LVA ISL  MLT ISL  KOR  5  LTU SVN FRA HRV HRV HUN SAU KOR SAU SAU NOR KOR GRC CHN NLD MYS LVA MYS FIN MLT NOR MLT NOR  6  SWE HKG PRT TUR NOR BEL NOR TUR RUS KWT LVA ISR  FIN GRC MYS NLD MYS CYP NLD NOR LVA NOR