Anxiety about the pandemic and trust in financial markets

The world has experienced the rapid and dramatic widespread of COVID-19 (Li et al. 2020; 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 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 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 and Bozos 2019, and Chizema 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.

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 contagion—while 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. 2000, 2003; Cohen-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.

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.

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 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.

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 \(j=1, \dots , J\). 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 \(k=1, \dots , K\).

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 \({\textbf{p}}_k^{j}=(p_k^{j}(1), \dots , p_k^{j}(T))\). Analogously, the sample of the Google searches for country j is \({\textbf{w}}^{j}=(w^{j}(1), \dots , w^{j}(T))\).

Notice that the range of variation of the components of \({\textbf{p}}_k^{j}\) and \({\textbf{w}}^{j}\) is different. Indeed, \({\textbf{p}}_k^{j}\) has non-negative components without a pre-defined ceiling, while the components of \({\textbf{w}}^{j}\) are integer numbers ranging in [0, 100], and there exists \({\bar{t}}\) such that \(w^{j}({\bar{t}})=100\). Time \({\bar{t}}\) 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 \({\textbf{w}}^{j}\) is not necessarily null. Indeed, null search means the absence of interest for the considered word in the country j—i.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 normalisation—implemented 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 \({\textbf{p}}_k^{j}\) for each j and k through a simple normalisation procedure. We denote the normalised series of the prices by \(\bar{{\textbf{p}}}_k^{j}\).

First, we identify \({\bar{t}} \in \{1, \dots , T\}\) such that \(p_k^{j}({\bar{t}})=\max \{p_k^{j}(t):t=1, \dots , T\}\). Then, we set \({\bar{p}}_k^{j}({\bar{t}})=100\). Null price is associated with zero value for the normalized series, so that we set \({\bar{p}}_k^{j}(t)=0\) when \({p}_k^{j}(t)=0\). Evidently, one can have \({p}_k^{j}(t)>0\) for each \(t=1, \dots , T\), so that one has \({\bar{p}}_k^{j}(t)>0\) for each t.

The entire series can be derived as followswhere \([\bullet ]\) is the integer part of the real number \(\bullet \).

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.

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.