Portfolio Decision Analysis For Pandemic Sentiment Assessment Based On Finance and Web Queries

Annals of Operations Research
https://doi.org/10.1007/s10479-024-05966-x
ORIGINAL RESEARCH
Portfolio decision analysis for pandemic sentiment

assessment based on finance and web queries
Roy Cerqueti1,2 · Francesco Cesarone3 · Valerio Ficcadenti4
Received: 28 June 2023 / Accepted: 21 March 2024

© The Author(s) 2024
Abstract
COVID-19 has spread worldwide, affecting people’s health and the socio-economic envi-
ronment. Such a pandemic is responsible for people’s deteriorated mood, pessimism, and
lack of trust in the future. This paper presents a portfolio decision analysis framework for
policymakers aiming at recovering the population from psychological distress. Specifically,
we explore the relative relevance of a country to the overall “mood of the world” in light of
pursuing predefined targets through optimization criteria. Toward this aim, we design a sta-
tistical indicator for measuring the mood by considering the financial markets’ outcomes and
the people’s online searches about COVID-19. Then, we adapt existing portfolio selection
models to evaluate the role of an extensive collection of countries and stock markets based on
different criteria. More precisely, such criteria are established assuming “rational” goals of a
policymaker, namely to aspire to a general and stable optimism and avoid waves of opposite
moods or excess pessimism. Empirical experiments validate the theoretical proposal. The
employed dataset contains 39 countries selected on the basis of data reliability and relevance
in the context of COVID-19. Data on daily Google Trends searches of the term “coronavirus”
(and its translations) and closing prices of relevant domestic stock indexes are considered
for 2020 to develop the statistical mood indicator. Results offer different insights based on
the selected optimization criteria. The practical implications of the proposed models have
been illustrated through arguments based on a National Recovery and Resilience Plan-type
normative framework.
The authors sincerely thank Prof. Anna Maria D’Arcangelis for helpful discussions.
B Francesco Cesarone
francesco.cesarone@uniroma3.it
Roy Cerqueti
roy.cerqueti@uniroma1.it
Valerio Ficcadenti
ficcadv2@lsbu.ac.uk
1 Department of Social and Economic Sciences, Sapienza University of Rome, Piazzale Aldo Moro,
5, 00185 Rome, Italy
2 GRANEM Université d’Angers SFR Confluences, 49000 Angers, France
3 Department of Business Studies, Roma Tre University, Via Silvio D’Amico 77, 00145 Rome, Italy
4 School of Business, London South Bank University, 103 Borough Road, London SE10AA, UK
123
Keywords Portfolio optimization · Statistical indicators · COVID-19 · Mood · Financial

stock index · Google trends
1 Introduction
The current period is characterised by one of the most significant pandemics in the modern
era—the so-called COVID-19. This disease, caused by a coronavirus, was officially identified
for the first time in China towards the end of 2019 (Li et al., 2020; Zhu et al., 2020). The world
is grappling with the dramatic effects of COVID-19 on health, evidenced by an increasing
number of infections and numerous fatalities. Additionally, the non-pharmaceutical interven-
tions to curb the spread of the pandemic share a common underlying principle: the reduction
of social interactions (see, e.g., Dave et al., 2021, Lewnard & Lo, 2020). In this context,
COVID-19 has also precipitated economic and financial crises, leading to widespread unem-
ployment (see, e.g., Kong & Prinz, 2020, Susskind & Vines, 2020, Milani, 2021, and the
monograph by Gans, 2020).
Generally speaking, the global pandemic situation has had deleterious effects on people’s
behaviour, manifesting in widespread anxiety and a diminishing trust in a positive future
outlook (see, e.g., Lee et al., 2020, Mann et al., 2020). In this regard, Asmundson and Taylor
(2020) coined the term coronaphobia.
This paper presents an operational research perspective to explore global citizens’
responses to the pandemic. The conceptualisation of this response is based on two imme-
diately interconnected aspects: the anxiety induced by the pandemic and the confidence in
the future performance of financial markets. We henceforth refer to the term “mood” to
describe the sentiment of a country’s citizens. As we shall see, mood is measured through a
combination of the aforementioned aspects.
Specifically, we integrate and complement the financial aspect of pandemic-induced anx-
iety using data from Google Trends related to COVID-19, as supported by the literature (see
Sect. 2). The mood of a country’s citizens is examined by jointly considering the prices of
one or more stock indices linked to it and the respective Google Trends indicator for the
term “coronavirus” within that country. When necessary, the term “coronavirus” has been
appropriately translated. The proposed synthetic mood indicator increases (decreases) when
prices rise (fall) and the Google search index diminishes (escalates).
The period under review spans 1 year, from January 1st, 2020, to December 31st, 2020, and
we consider daily data. Notably, this period was predominantly marked by the implementation
of non-pharmaceutical measures to combat COVID-19. By analysing this timeframe, we
exclude the influence of vaccines introduced at the end of 2020, as it would require a separate
analysis in the context of population mood swings. We select relevant stock indices for each
country based on Bloomberg’s relevance score R and the Human Development Index (HDI,
2019). For more details, see Sect. 5.1.
The process of selecting stock indices results in one index per country, with the exceptions
of China and Finland, which have two each. Therefore, with the aforementioned exceptions
in mind, we will refer to the index, market, or country interchangeably in the following
sections.
The global mood indicator is defined as the weighted mean of the moods of each stock
index, where a weight represents the relative importance of a stock index/country to the
world’s mood. From a modelling perspective, these weights can also be interpreted as the
shares of a portfolio. Consequently, one can suitably select the portfolio’s weights to pursue
123
a specific target mood for the world or, alternatively, a specific target for its fluctuation—for
example, maximising the expected world’s mood while minimising its variance.
More specifically, the aim of this paper is twofold. Firstly, we propose a mood indicator
based on both financial data and Google Trends concerning COVID-19. Secondly, we deter-
mine each country’s contribution to the global mood when specific targets are set—thereby
guiding policymakers in developing strategies to mitigate the psychological distress caused
by COVID-19. This issue can be seen as a project selection process to achieve a predefined
target.
In this paper, we employ Portfolio Decision Analysis (PDA) to investigate the global
reaction to COVID-19. We consider and compare different portfolio models, each tailored
to specific criteria for evaluating the relevance of indices in optimally describing the world’s
mood in relation to specific targets.
In the first approach, a high and stable mood is considered positive, representing a condition
of general unperturbed optimism. The target is to elevate the level of the global mood while
curtailing its fluctuations. This approach is supported by empirical evidence suggesting that
an increase in countries’ mood tends to bolster their economic, financial, and social prosperity.
Furthermore, minimal fluctuations in the global mood are believed to enhance the stability
of financial markets by mitigating economic imbalances caused by the so-called “animal
spirits” (see De Grauwe, 2011). In this context, we propose a well-known Risk-Gain model
based on symmetric risk measures, namely the Mean-Variance (MV) model (Markowitz,
1952, 1959).
In the second approach, the focus shifts from minimising mood fluctuations to avoiding
low mood levels associated with an overly pessimistic environment. In this regard, we observe
that waves of negative mood and pessimism might contribute to deteriorated economic per-
formances (see, e.g., Jouini & Napp, 2011). Thus, we explore a different Risk-Gain model,
the Mean-CVaRD (Conditional Value-at-Risk Deviation) portfolio problem (Rockafellar &
Uryasev, 2000; Sarykalin et al., 2008; Filippi et al., 2020).
Under a third perspective, we utilise the Risk Parity (RP) approach for conceptualising
the optimal portfolio model (see Maillard et al., 2010, Roncalli, 2014, Cesarone & Tardella,
2017, Cesarone & Colucci, 2018, Bellini et al., 2021). This strategy focuses on the concept of
equal risk contribution and aims to ensure that all indices associated with countries contribute
equally to mood fluctuations. The rationale behind this approach is to prevent any single
country from predominantly influencing high and low mood waves, thereby fostering a sense
of shared belonging among the countries linked to the stock indices. This could promote
solidarity and facilitate a swifter psychological recovery from the crisis.
Lastly, we propose a risk diversification model, introduced by Choueifaty and Coignard
(2008), which consists of maximising the relative distance between the volatility of the
world’s mood and that associated with the worst scenario. Here, the policymaker’s objective
transcends merely maximising the mood level or reducing its fluctuations. This approach
aligns with the notion that mood extremes might be associated with economic distress—
see, for instance, Lowe and Ziedonis (2006) for the economic analysis of overoptimism
and unjustified self-confidence, and Jouini and Napp (2011) for the impact of excessive
pessimism.
Several noteworthy results emerge from our empirical analysis, offering valuable insights
for policymakers.
The MV optimal portfolios with low target levels for the expected mood allocate more
relevance to indices associated with a set of countries of the European continent, namely
Greece, Turkey, Norway, and Denmark. This reflects the model’s tendency to mitigate pes-
simism and optimism waves in the world’s mood. Conversely, stock indices related to Hong
123
Kong, Qatar, and Singapore assume significant roles in the optimal portfolios when targeting
a high world mood level.
The Mean-CVaRD optimal portfolios, focusing on low expected mood targets, assign
greater importance to indices related to Saudi Arabia, South Korea, Norway, and Malaysia,
aiming to temper fluctuations towards a low-level mood of the world.
The aforementioned portfolio models are comparable in terms of the expected mood target.
In both cases, there is greater diversification when the expected mood target is low compared
to when it is high. However, in the MV case we have a penalization of high fluctuations of
the mood, disregarding their sign. Differently, the Mean-CVaRD model specifically penalises
countries having only high deviations from the expected mood from the negative side. So, the
MV model curbs countries exhibiting high levels of pessimism and optimism—essentially
moderating extreme mood behaviours—whereas the Mean-CVaRD model focuses solely on
high levels of pessimism. This outcome is more evident when the expected mood target is
low, in agreement with standard features of the Risk-Gain analysis á la Markowitz.
The RP portfolio, by construction, includes all countries/stock indices in the composition
of the world’s mood. As a result, the scores associated with each index within a country tend
to be relatively uniform, precluding any single country from assuming a dominant role and
promoting a sense of unity among the countries.
For the Most Diversified (MD) portfolio, we observe a high relevance for indices linked to
Denmark, Oman, Greece, and Chile. This allocation aims to minimise fluctuations caused by
pessimism and optimism waves and to distance them as much as possible from the worst-case
scenario (where the imbalances in financial markets and the countries’ moods are perfectly
correlated). We note that, despite the appealing theoretical aspects of the Gain-Risk models,
their results are quite sensitive to variations in the preselected expected gain of the optimal
portfolio. In contrast, the RP and MD strategies exhibit less sensitivity to estimation errors
of the model inputs (see Cesarone et al., 2020, and references therein).
The methodology we adopt for modelling pandemic-related anxiety aligns with Cerqueti
and Ficcadenti (2023), where the authors develop mood indicators based on a combination
of Google searches for the term “coronavirus” and the financial market performances at a
country level. However, the referenced paper focuses solely on illustrating the country-based
situation of anxiety, without delving into portfolio decision problems. Therefore, our paper
has a more strategic objective, guiding policymakers towards different mood targets under
various portfolio model characterisations. In this respect, our paper has practical implications.
Indeed, we can suggest a reliable identification of a supra-national entity acting as a portfolio
optimiser, giving tangible meaning to the optimal portfolio shares. Specifically, we draw a
parallel with the recent global action to recover from the pandemic, the so-called National
Recovery and Resilience Plan (NRRP). In the NRRP, the European Union allocated funds
to individual countries for developing actions and policies to support the socio-economic
environment post-pandemic shock. Among these interventions, healthcare played a crucial
role. However, some countries lacked a focus on the psychological well-being of citizens.
This underscores the importance of considering psychological recovery as a critical target
(see, e.g., Ussai et al., 2022). Our work fits within a framework akin to the NRRP, but with
a specific focus on psychological recovery from COVID-19. In this context, the European
Union can be viewed as a policymaker deeply invested in determining the frustration levels
of EU state citizens. To this end, one could introduce a mood indicator—like the one we
propose in this paper—to represent the general mood state of a given country’s inhabitants.
The mood indicators of individual countries are then aggregated to form a global (European)
mood—analogous to the world mood in our context. The European Union would then estab-
lish actions to achieve specific targets related to this global mood by optimally allocating
123
resources to individual countries—as already executed for the NRRP. These policies are for-
malised through the establishment of portfolio problems in various contexts—as explained
in this paper. The optimal portfolio shares give the percentage of the capital to be allocated
to the individual countries to achieve the proposed goal. In our model, the shares capture the
relevance of the countries (in terms of indices and searches) in achieving specific targets.
These shares can also be suitably scaled to account for the differing population sizes of the
countries. As with the NRRP, the European Union provides guidelines for explicit actions by
institutions, universities, and research centres. The inclusion of non-European countries such
as China—whose relevance in the context of COVID-19 is undeniable, see Sect. 5.1—can
be interpreted as the establishment of strategic interactions between the European Union and
other countries, envisaging an extended version of the NRRP specifically targeting psycho-
logical recovery from the pandemic.
The remainder of the paper is organised as follows. Section 2 reviews relevant literature
for our study. Section 3 defines the statistical indicator for measuring mood, introducing its
main properties and characteristics. In Sect. 4 we develop the employed concept of portfolio
and present the four portfolio selection models considered. Section 5 is dedicated to empirical
experiments, presenting the dataset used and critically analysing the study’s results. The final
section offers some concluding remarks.
2 Literature review
There is a wide literature showing that financial markets are vulnerable to catastrophic and
unexpected events of non-financial nature like aviation disasters (see, e.g., Kaplanski & Levy,
2010), terrorist attacks (see, e.g., Goel et al., 2017), or, recently, COVID-19 (see, e.g., Štefan
Lyócsa et al., 2020). Importantly, there is evidence that the occurrence of disasters tends to
generate approximately instantaneous dramatic collapses in prices (see, e.g., Barro, 2006,
Gabaix, 2012, Gourio, 2012). In the particular context of the pandemic, Goodell (2020) high-
lights the existence of an explicit parallelism between COVID-19 and terrorism, disasters of
ecological nature and nuclear conflicts. Undoubtedly, intuitive psychological reasons explain
the link between disasters and the approximately instantaneous fall in financial indices, being
disasters associated with anxiety, panic, and consequent loss of confidence in the financial
markets. In Da et al. (2014), the authors found that online searches of locutions attributable to
the sense of fear are instantaneously related to financial returns (some aspects of the timing
in the relationship have been recently confirmed in Nikkinen and Peltomäki, 2020), and in
Da et al. (2011) a proposal for detecting investors’ attention in a timely manner is made, and
it is based on Google searches.
We are in line with this strand of literature, and we adopt the view that the collapse of the
stock index prices might effectively contribute to measuring the effect of the pandemic on
the mood of the citizens of the related country. We then align with recent literature showing
that COVID-19 has led to anxiety and pessimistic expectations about economic performance.
Binder (2020) provides a survey study showing that the anxiety about the pandemic mirrors
the loss of confidence in positive future outcomes of macroeconomic variables. Fetzer et
al. (2020) extend the perspective offered by Binder (2020) by including time dependence
and exploring the causal effect of COVID-19 on the growth of anxiety of economic nature.
Interestingly, when reporting the results of Binder’s survey, she wrote, “Nearly all participants
follow news about coronavirus, 50% somewhat closely and 43% very closely”. Concerning
the expectations related to the pandemic, she reports: “concerns about economic effects are
123
most prevalent, with 52% somewhat concerned and 38% highly concerned. Consumers who
follow news about the coronavirus more closely tend to be more concerned about the effects of
the virus”. Furthermore, “respondents who own stocks or follow news about the stock market
seem more attentive to coronavirus news, more concerned […]”. Under the same perspective,
Fetzer et al. (2020) focus on sentiment information by highlighting the interrelation between
Google Trends data and the measurement of how people react to the pandemic. Google Trends
data are also used by Nikolopoulos et al. (2021) for implementing forecasting exercises on
the “excess and intermittent demand” that leads to “panic buying” (see Tsao et al., 2019) of
“different products and services including groceries, electronics, automotive and fashion” for
timely adjustments of the logistics in supply chains management. Garfin et al. (2020) clearly
state that online searches and exposure to media are strongly connected to the fear of such a
pandemic disease. More specifically, a large (small) amount of coronavirus-related searches
on the web is associated with a high level of pessimism (optimism).
We now discuss relevant contributions related to PDA being the employed methodological
framework chosen for this paper.
Taking portfolio selection models whose optimal shares represent a measure of the rel-
evance of the associated alternatives is also the ground of the well-known and widely
used Composite Indicator of Systemic Stress (CISS). It is a portfolio-based systemic risk
assessment device developed by the European Central Bank (ECB, see the research paper
Hollo et al., 2012). The authors consider five stress measures—based on five categories—
in the context of systemic risk; then, they identify the relative weights of such measures
for detecting the relevance of the related stress categories—the higher the weight, the more
relevant the category for minimising systemic risk.
In general, the need to make optimal decisions under uncertain conditions—eventually
subject to specific constraints—is one of the main drivers of OR-based studies. Recently, in
Salo et al. (2011), the idea of taking decisions within a set of available alternatives has been
translated into mathematical methods and models, defining the Portfolio Decision Analysis
(PDA). Such a theoretical framework is the approach adopted in this paper for facing our
research objectives. Liesiö et al. (2020b) contains a recent review of developments and
prospects of PDA, and it is the paper from which we grasp motivations for adopting a
PDA approach in our study. Indeed, quoting Liesiö et al. (2020b), one reads “Typical PDA
analyses provide recommendations for the selection of projects or the allocation of resources”.
Furthermore, our work falls in the “two-thirds of articles in the sample of the methodologically
oriented PDA papers reporting realistic applications” (again from Liesiö et al., 2020b, where
148 articles related to PDA are reviewed).
In general, PDA is helpful for formalising rational decision-making processes and iden-
tifying straightforward optimal solutions. PDA helps in obtaining a view and consequent
management insights throughout the analysis of the resulting efficient sets of solutions. In
this respect, such a methodological framework supports decisions among multiple scenarios
by evaluating the individual contributions to pre-identified objectives.
PDA has been widely employed in several applied contexts. Baker et al. (2020)’s authors
take the government as the policymaker and deal with the analysis of research and devel-
opment expenses for designing the energy technology in the context of climate change. In
the same vein, the authors in Liesiö et al. (2020a) and Toppila et al. (2011) present a deci-
sion model for allocating resources in portfolios of decision-making units and research and
development activities. Differently, Grushka-Cockayne et al. (2008) explore a problem of
air traffic flow optimisation by clearly illustrating the various elements of the reference PDA
context. In Barbati et al. (2020), the authors identify the optimal combination of facilities
by considering their multi-attribute nature, with a particular reference to their locations and
123
activating times. Mild et al. (2015) adopt a PDA approach for the selection of the mainte-
nance projects of the bridges and apply their study to the data of the Finnish Transportation
Agency. Kangaspunta et al. (2012) evaluate the cost-efficiencies of weapon systems portfo-
lios, while Chowdhury and Quaddus (2015) select “the most satisfactory efficient portfolio
of supply chain resilience strategies” for the case of three Bangladesh companies. More in
general, research involving PDA applications can be found in the IT sector (Gemici-Ozkan
et al., 2010), energy management (Cranmer et al., 2018), health care field (Mastorakis &
Siskos, 2016; Phillips & e Costa, C. A. B., 2007) and many others. In Barbati et al. (2018),
the authors advance a portfolio decision problem by also considering the qualitative satis-
faction level of the policymaker’s choices as the objective function to be maximised. We are
close to the perspective of Barbati et al. (2018), in that we here consider the mood—which
is undoubtedly a qualitative variable—as the objective function. As stated above, we offer
a quantitative translation of mood to be used in the optimisation process. Castellano et al.
(2021) also define the shares of a portfolio as relative relevance scores assigned to units for
pursuing a predefined target. They use countries as units, and the target is the systemic risk
minimisation. However, to the best of our knowledge, our paper is the first to address the
optimal selection of units for analysing COVID-19 and its consequences in the context of
sentiment formation.
3 Statistical indicator for the mood
This section provides formal definitions of reliable measures of the mood of citizens in various
countries when dealing with pandemics and financial markets. As previously discussed,
our scientific basis is grounded in the evidence—well-established in recent authoritative
literature—that a simultaneous increase in Google searches and a decrease in stock index
prices capture a sense of pessimism, while a concurrent decline in Google searches and a
rise in prices indicate an optimistic mood.
To address the optimal portfolio problem, we have designed a statistical indicator that
captures the interplay between anxiety about the pandemic and expectations regarding the
future performance of financial markets—in brief, the mood.
3.1 Preliminaries and notation
Consider J countries. In a country j, there are K ( j) stock indices, so that the generic stock
index of country j is denoted by k( j) = 1, . . . , K ( j)—with the majority of their components
linked to the country in question. The total number of considered stock indices is denoted
by

J
K = K ( j).
j=1
By construction, K ≥ J .
We consider a common observation period of T days for both the stock indices’ prices
and the Google searches data. Specifically, we excluded Google search results that do not
correspond with financial data for the same date. When the market is closed in a country,
there are no values for price; thus, the corresponding Google searches cannot be used in
constructing the mood indicator.
123
For country j, the daily prices of stock index k( j) are denoted by pk( j) = ( pk( j) (1), . . . ,
pk( j) (T )), while the volume of Google searches for the term “coronavirus” (and its transla-
tions) in country j is represented by w j = (w j (1), . . . , w j (T )).
Note that prices are nonnegative quantities, and Google searches—sourced from Google
Trends, as detailed in Sect. 5.1—range from 0 to 100. A value of w j (t) = 0 indicates no
Google searches reported at time t in country j. Hence, null values of ws are not always
observed. Indeed, there may be a nonnull amount of searches every “day”—by taking “days”
as the time units—so a value of 0 is rarely achieved. Conversely, a value of 100 certainly
appears in correspondence with the day having the highest level of searches. However, when
querying a large volume of data from Google Trends, careful attention is required due to
Google’s scaling system for the search volume index. During data download from Google
Trends, the geographical area of interest is selected, leading to normalization within that
geographical area. This means that each query to download the search volume of a word
in a country is normalized between 0 and 100, allowing for comparisons among countries.
For 1-year data requests, as in our case, the most granular accessible data is weekly (a
common limitation noted by scholars, see Maggi & Uberti, 2018, 2021). In this scenario, the
index value of 100 corresponds to the week with the maximum searches over the queried
period. Conversely, for a 1-month request, daily data is normalized for that month. Therefore,
a value of 100 obtained from a monthly query and a value of 100 derived from yearly
data normalization are not directly comparable. To address this issue, we implemented the
following procedure to derive daily observations from weekly data (in line with methods
used in other studies, such as Vollmer et al., 2021). Initially, for each country, monthly search
volumes are downloaded, providing daily volumes for each month—as Google permits daily
data downloads for 1-month periods. Subsequently, weekly data are downloaded for the entire
period (01/01/2020–31/12/2020, see Sect. 5.1), yielding 52 observations, one per week. The
weekly data are then multiplied by the corresponding daily values obtained in the first phase,
and the resultant figure is divided by 100.
For comparison purposes, we also normalize the series of observed daily prices to the
range of [0, 100], converting pk( j) (t) ∈ [0, +∞) to a corresponding p̄k( j) (t) ∈ [0, 100]. To
achieve this, we identify t¯ ∈ {1, . . . , T } such that pk( j) (t¯) = max{ pk( j) (t) : t = 1, . . . , T }
and set

pk( j) (t)
p̄k( j) (t) = 100 × , ∀ t = 1, . . . , T , (1)
pk( j) (t¯)
where [•] denotes the integer part of number •. This Formula (1) sets the p̄k( j) ’s value to 100
for the highest price observed in the period {1, . . . , T }; p̄k( j) (t) = 0 when pk( j) (t) = 0.
The daily variations of the normalized daily prices p̄’s and the Google searches w’s are
denoted as follows:
p̄k( j) (s) = p̄k( j) (s + 1) − p̄k( j) (s), ∀ s = 1, . . . , T − 1, (2)
and
w j (s) = w j (s + 1) − w j (s), ∀ s = 1, . . . , T − 1, (3)
3.2 Definition and properties of the mood indicator
We propose a time-dependent mood indicator based on the comparison of variations in daily

prices and Google searches, as given in Eqs. (2) and (3). First, we define the indicator; then,
we discuss its properties and relevant observations.
123
Consider t ∈ {1, . . . , T − 1}, j = 1, . . . , J and k( j) = 1, . . . , K ( j). We define

1
m t,k( j) = · α( p̄k( j) (t)) · p̄k( j) (t) − α(w j (t)) · w j (t) + 2 × 104 , (4)
4 × 10 4
where, given H = p̄k( j) , w j , we have

H (t), if H (t) ≥ 0;
α(H (t)) = (5)
100 − H (t), if H (t) < 0.
The indicator in Formula (4) jointly evaluates the stock index price p̄k( j) and the extent of
Google searches w j . The former serves as a proxy for confidence in the performance of index
k( j); the latter reflects the level of pandemic-related anxiety experienced by the citizens of
country j.
The daily prices in the p̄k( j) result from investments in k( j). Empirical evidence suggests
that transactions based on index k( j) are predominantly made by citizens of country j—the
host country of k( j). Nevertheless, the involvement of foreign investors cannot be discounted.
Thus, investing in k( j) also signifies trust in the economic status and prospects of country j.
Therefore, from a general perspective, a thoughtful combination of the prices of k( j)
and the Google searches in country j enables us to interpret the mood of country j. More
specifically—as mentioned in the Introduction—we explore the link between financial mood
and trust in the performance of stock indices in country j, and the anxiety surrounding the
pandemic. Pessimism is indicated when trust in financial performance is low, and anxiety
is high; conversely, optimism is signalled by high economic confidence and low pandemic-
related anxiety.
For analysing the financial aspect of country j’s mood, we rely on the stock indices
k( j) = 1, . . . , K ( j); thus, as detailed later, the mood of country j is also represented through
aggregated confidence in the performance of stocks within the indices related to country j
(see Formula (7) in Sect. 4).
By (4), we find m t,k( j) ∈ [0, 1]. Indeed, the construction of (4) is informed by the variation
range of the Google Trends data, which is [0, 100]. We have retained this range and accord-
ingly set the range of variation for the financial component to [0, 100]. The components of
the indicator are quadratic so that 104 acts as a scale factor for normalisation, setting the
range to [0, 1].
This indicator comprises terms of two different types: the ’s represent daily variations,
while the function α acts as a variation weight. To elucidate the meaning of m t,k( j) , let’s
delve into the details.
One has p̄k( j) (t) > 0 when the daily price of the stock index labelled k( j) has increased
from day t to t + 1. This signifies rising investor confidence in its future performance,
reflecting optimistic behaviour. In the same way, w j (t) < 0 is a signal of optimistic
behaviour; indeed, this case is associated with a decreasing interest in the pandemic as
reflected in Google searches, suggesting a reduction in related anxiety. In the former scenario,
optimism is amplified by a high level of price p̄k( j) (t). Specifically, for a given fixed level of
p̄k( j) (t) > 0, the optimism is more pronounced when the initial price p̄k( j) (t) is higher. The
rationale behind this condition lies in the evidence that when the increasing confidence in the
stock market moves from a good performance, optimism is persistent, and it is more radicated.
By following the same argument, the optimism described in the latter case is amplified when
w j (t) is small, i.e., when 100 − w j (t) is large. In the same way, we have that p̄k( j) (t) < 0
and w j (t) > 0 describe situations of pessimism. Such moods are exacerbated when price
p̄k( j) (t) is low and the Google searches w j (t) are large. Indeed, these cases are associated
with the persistence of pessimism.
123
The indicator m t,k( j) is a normalised measure reflecting the level of optimism in country
j for index k( j); m t,k( j) increases as the level of optimism grows. In particular, such an
indicator approaches one when, ceteris paribus, p̄k( j) (t) and p̄k( j) (t) approach 100, w j (t)
approaches -100 and w j (t) approaches zero. Conversely, m t,k( j) gravitates towards zero
when, ceteris paribus, p̄k( j) (t) goes to -100, p̄k( j) (t) approaches zero while w j (t) and
w j (t) approach 100. The former scenario is indicative of high optimism, while the latter is
symptomatic of strong pessimism.
4 Optimal portfolio models
This section outlines the considered optimal allocation models. First, we provide the definition
of the portfolios. Then, we present the portfolio selection strategies analysed.
4.1 Definition of the portfolio
The starting point of the analysis is a concept of time-dependent (non-financial) portfolio,

whose definition is formally given as follows.
Definition 4.1 Consider t = 1, . . . , T − 1. The K -ple of real numbers π(t) = (π1 (t), . . . ,
K
π K (t)) is said to be a portfolio at time t when i=1 πi (t) = 1 and πi (t) ≥ 0, for each
i = 1, . . . , K . The π’s are the shares of the portfolio. Vector π = (π(1), . . . , π(T − 1)) is
a portfolio trajectory—or, simply, a portfolio.
The mathematical definition of portfolio π in Definition 4.1 is the same as the standard
financial one in Markowitz (1952, 1959), when short-selling is not allowed. However, the
interpretation of the shares of the portfolio is radically different here. By employing the
notation introduced in the previous sections, we say that πk( j) (t) is the share of the portfolio
associated with the j-th country—with specific reference to the k( j)-th index—at time t,
for each j = 1, . . . , J and k( j) = 1, . . . , K ( j). Such a share provides a measure of the
relevance of country j and, specifically, index k( j) for pursuing the targets formalised in the
proposed portfolio models. After stating the optimisation models, we will provide several
details on the meaning of the portfolio below.
The proposed portfolio models are based on the evaluation of the aggregated moods of the
considered countries—briefly, the mood of the world—on the basis of the time-dependent
statistical indicator introduced in the Formula (4).
Definition 4.2 Fix a time t = 1 . . . , T − 1 and a portfolio π(t) = (π1 (t), . . . , π K (t)). The
mood of the world at time t associated to the portfolio π(t) is

J K ( j)
M(t, π) = πk( j) (t)m t,k( j) . (6)
j=1 k( j)=1
Definition 4.2 clarifies the meaning of the shares of the portfolio π(t). Indeed, the term
πk( j) (t) represents the relative contribution of the index k( j) in country j to the mood of
the world. Indeed, Formula (6) claims that the mood of the world is a weighted mean of
the moods based on the consider stock indexes, and the shares of the portfolio represent the
weights.
123
Furthermore, given j = 1, . . . , J , one can identify the relative contribution of country j

in π(t) by aggregating the shares of the portfolio related to the indexes linked to country j,
so that one has

K ( j)
π j (t) = πk( j) (t). (7)
k( j)=1
Interestingly, vector π(t; j) = (π 1 (t), . . . , π J (t)) is still a portfolio according to Definition

4.1, whose shares represent the relative relevance of the countries for the mood of the world.
Such a portfolio has intuitively relevant informative content. However, the term π j (t) in For-
mula (7) is an aggregated quantity; therefore, one does not have the possibility of identifying
from it its granular components—i.e., the relative contribution of an individual index k( j)
linked to country j.
4.2 Portfolio selection models
COVID-19 has a vast impact on countries’ moods (and therefore on the world’s mood),
affecting social, political and economic aspects of the citizens’ lives. An analysis of these
impacts is essential to understand and manage the effects of the world’s mood on society
and the economy. Indeed, waves of generalised optimism and pessimism—i.e., the behaviour
of the world’s mood—might be driven by the anxiety of the citizens in specific countries.
Thus, the action for leading the world’s mood towards a target has to provide the relative
contributions to be assigned to stock indexes and countries, i.e., the optimal portfolio. We
aim to identify the weights of a given set of stock indexes and countries so that the selected
world’s mood portfolio is the best one according to specific criteria. To accomplish this objec-
tive, we exploit several concepts of portfolio optimisation. As a consequence, these selected
portfolios could give the policymaker an indication of which stock indexes and countries are
more relevant to achieve prefixed goals in terms of global mood. For this purpose, we intro-
duce the optimal portfolio models used for this analysis. More precisely, we investigate two
classes of models for selecting a portfolio: the classical Risk-Gain analysis á la Markowitz
and risk diversification approaches. The first class consists of maximizing the expected value
of the world’s mood while simultaneously minimizing its fluctuations. Such fluctuations are
evaluated by means of two deviation risk measures (see Rockafellar et al., 2006), volatility
and Conditional Value-at-Risk Deviation. In fact, in portfolio selection, typically, we can
distinguish a first step of the Risk-Gain analysis where the efficient portfolios, namely the
Pareto-optimal solutions, are identified. Then, among the Pareto-optimal portfolios, we can
adopt preference criteria, for instance, by requiring specific target levels in terms of portfolio
risk or gain, which, therefore, are eventually introduced in the second step of the Risk-Gain
analysis (see, e.g., Cesarone, 2020). In this specific case, we know a-priori that when the
required level of the expected value of the world’s mood increases, the number of countries
involved in the Pareto-optimal solutions typically tends to decrease. For the second class, we
consider two relatively recent risk-focused portfolio selection approaches. The Risk Parity
(RP) strategy aims to achieve a balanced portfolio in terms of mood fluctuation. This risk
diversification approach is based on the general notion of equal risk contribution from each
asset. It can be shown that the fluctuation of the RP portfolio is bounded between the fluctu-
ation of the minimum risk portfolio and that of the Equally Weighted portfolio (Cesarone &
Tardella, 2017; Cesarone et al., 2020; Cesarone & Colucci, 2018). The Maximum Diversifi-
cation ratio approach consists in maximizing the relative distance between the portfolio risk
123
in the worst-case scenario, namely when risk is additive, and the generic portfolio risk. Note
that risk is additive when the risk drivers are highly dependent (see, Cesarone et al., 2023,
Ararat et al., 2024, Bellini et al., 2021).
We assume that the country moods are discrete random variables distributed in a discrete
state space, and that there are T states of nature, each with probability qt with t = 1, . . . , T .
We use a look-back approach where the outcomes of the discrete random variables correspond
to past realised data. More precisely, the choice of the relative contributions of stock indexes
and countries to the world’s mood is made using T historical scenarios, each with probability
qt = 1/T , if there are no ties (see, e.g., Carleo et al., 2017, Cesarone, 2020, and references
therein). In so doing, we start from Formulas (4) and (6) and consider the related random
variables m k( j) and M(π), respectively.
The optimal allocation models’ formalizations follow the arguments of the Introduction,
related to optimal criteria based on what is expected by the value of the mood and on its
fluctuations. A location parameter captures the former quantity, i.e., the expected value of
the mood of the world μ(π) = E[M(π)], whilst the latter one is given by a dispersion
parameter (e.g., the variance of M(π)), that yields an indication of the amplitude of the
oscillations around the location.
4.2.1 Pareto-optimal risk-gain portfolios
In this section, we describe two portfolio selection models focused on the Risk-Gain analysis,
namely the study of the best trade-off between the maximisation of the gain—i.e., the expected
value—and the minimisation of the risk—i.e., the fluctuations of the mood. In particular, we
consider two well-known Risk-Gain models based on both symmetric and asymmetric risk
measures. In the former case, we seek the stability of the mood, while the latter is associated
with avoiding overpessimism.
The symmetric framework is the classical MV model (Markowitz, 1952, 1959). With it,
we aim to determine the relative contribution of each index k( j) linked to country j, πk( j) , to
the total mood of the world that minimises the whole portfolio mood instability represented
by its variance while binding the expected mood of the portfolio to attain at least a fixed
target level of the mood. Thus, denoting by σ 2 (π) the variance of the world’s mood, the MV
model can be expressed by the following optimisation problem
(i)
J K ( j)
J K
min σ 2 (π) = σh(i)k( j) πh(i) πk( j)
π i=1 h(i)=1 j=1 k( j)=1
s.t.
( j)
J K
μ(π) = πk( j) μk( j) = η (8)
j=1 k( j)=1
( j)
J K
πk( j) = 1
j=1 k( j)=1
πk( j) ≥ 0 k( j) = 1, . . . , K ( j) and j = 1, . . . , J
where μk( j) represents the expected mood of the index k( j) in country j, and σh(i)k( j) is
the covariance between moods of the index h(i) (of country i) and k( j) (of country j)
with i, j = 1, . . . , J . Furthermore, η ∈ [ηmin , ηmax ] is the required level of the portfolio
expected mood, where ηmin denotes the value of the portfolio expected mood μ(π) at an
optimal solution of the problem obtained by deleting the first constraint in Formula (8), whilst
ηmax = max{. . . , μk( j) , . . . , μ K ( j) , . . .} with j = 1, . . . , J (see, e.g., Cesarone et al., 2013,
Cesarone, 2020).
123
The second model is based on an asymmetric risk measure, the Conditional Value-at-Risk
(CVaR) (Rockafellar & Uryasev, 2000) at a specified confidence level ε > 0, C V a Rε , i.e., in
financial terms, the mean of losses in the worst 100ε% of the cases (Acerbi & Tasche, 2002),
where losses are defined as negative outcomes. Therefore, it makes sense to minimize C V a Rε
only if it is positive (Sarykalin et al., 2008). Since our quantity of interest is the world’s mood
portfolio M(π), that is a nonnegative random variable—more precisely, M(π) ∈ [0, 1]—we
use CVaR Deviation (CVaRD) as a risk measure, namely C V a R Dε (π) = C V a Rε (μ(π) −
M(π)). A formal definition of CVaRD is
ε
1
C V a R Dε (π) = Q μ(π )−M(π ) (α)dα , (9)
ε
0
where Q μ(π )−M(π ) (α) is the α-quantile function of the deviation of the portfolio mood M(π)
from its mean μ(π).
Similar to the MV model, the Mean-CVaRD model can be expressed by the following
optimization problem:
min C V a R Dε (π)
π
s.t.
( j)
J K
μ(π) = πk( j) μk( j) = η
j=1 k( j)=1 (10)
( j)
J K
πk( j) = 1
j=1 k( j)=1
πk( j) ≥ 0 k( j) = 1, . . . , K ( j) and j = 1, . . . , J
Unlike Model (8) where both positive and negative deviations of the world’s mood portfolio
from its mean are penalized, Model (10) aims to identify those portfolio weights that minimize
the mean of the world’s mood portfolio in the worst 100ε% of the cases, namely the cases
of greatest “pessimism” in the world.
From a computational point of view, Model (10) can be reformulated as a linear pro-
gramming problem using the same approach of Rockafellar and Uryasev (2000). Thus, we
have
1
T
min ζ + dt
(π,ζ,d) εT
t=1
s.t.
( j)
J K
dt ≥ (μk( j) − m t,k( j) )πk( j) − ζ t = 1, . . . , T
j=1 k( j)=1
dt ≥ 0 t = 1, . . . , T (11)
( j)
J K
μ(π) = πk( j) μk( j) = η
j=1 k( j)=1
( j)
J K
πk( j) = 1
j=1 k( j)=1
πk( j) ≥ 0 k( j) = 1, . . . , K ( j) and j = 1, . . . , J
ζ ∈R
This reformulation is obtained by considering T auxiliary variables dt defined as the
J K ( j) J K ( j)
deviation of j=1 k( j)=1 (μk( j) − m t,k( j) )πk( j) from ζ when j=1 k( j)=1 (μk( j) −
123
m t,k( j) )πk( j) > ζ and 0 otherwise, and by adding the following constraints: dt ≥ 0,
J K ( j)
dt ≥ j=1 k( j)=1 (μk( j) − m t,k( j) )πk( j) − ζ ∀t. Note that when solving this optimisa-
tion problem, the optimal value of the variable ζ coincides with V a Rε (π ), where π is the
optimal solution of Problem (11).
4.2.2 Risk diversification strategies
In this section, we present two relatively recent portfolio selection approaches focused on
risk diversification. More precisely, we are looking for portfolios that can be interpreted as
indicative scores associated with each stock index in a country and that allow us to obtain
the goals described through the two risk-focused portfolio selection strategies described
below. One of the main methodological reasons why we examine models based on risk
diversification strategies is that, as shown by the extensive empirical analysis of Cesarone et
al. (2020), generally, they are less sensitive to estimation error of the inputs to the model.
The Risk Parity (RP) strategy, developed by Maillard et al. (2010), in the case where
J K (i) J K ( j)
the volatility measures the risk σ (π) = i=1 h(i)=1 j=1 k( j)=1 σh(i)k( j) πh(i) πk( j) ,
requires that each asset equally contributes to the total risk of the portfolio. In this way, one
pursues the target of avoiding the existence of a country with a leading role in generating
waves of optimism and pessimism, hence fostering a shared sense of belonging. The standard
approach used for decomposing the portfolio volatility is the Euler allocation, namely σ (π) =
J K ( j)
j=1 k( j)=1 RC k( j) (π), where
J K (i)
∂σ (π) 1
RCk( j) (π) = πk( j) = σh(i)k( j) πk( j) πh(i)
∂πk( j) σ (π)
i=1 h(i)=1
is the risk contribution of the k( j)th index in country j. The RP approach requires equality
of all total risk contributions
K (l)
J K (l)
J
RC h(i) (π ) = RCk( j) (π ) ⇔ σh(i)k(l) πh(i) πk(l) = σk( j)k(l) πk( j) πk(l) ∀h(i), k( j).
l=1 k(l)=1 l=1 k(l)=1
Then, a possible method for finding an RP portfolio is to solve the following system of linear
and quadratic equations and inequalities
⎧ (l)
⎪
⎪ J K
⎪
⎪ σh(i)k(l) πh(i) πk(l) = λ h(i) = 1, . . . , K (i) and i = 1, . . . , J
⎪
⎨ l=1 k(l)=1

J K ( j) (12)
⎪
⎪ πk( j) = 1
⎪
⎪
⎪
⎩ j=1 k( j)=1
πk( j) ≥ 0 k( j) = 1, . . . , K ( j) and j = 1, . . . , J
that has a unique solution when the covariance matrix is positive definite (Cesarone et al.,
2020). Therefore, applying the RP strategy to the world’s mood portfolio leads to the search
of the weights of all indexes in such a way that their contributions to the total volatility of
the world’s mood are equally distributed among all indexes.
123
An alternative strategy based on risk allocation has been introduced by Choueifaty and
Coignard (2008) and consists in maximising the so-called diversification ratio
J K ( j)
j=1 k( j)=1 πk( j) σk( j)
D R(π) = J K (i) J K ( j) , (13)
i=1 h(i)=1 j=1 k( j)=1 σh(i)k( j) πh(i) πk( j)
where σk( j) is the mood volatility of k( j). Note that D R(π) represents the ratio between
the portfolio volatility in the worst case—i.e., where the indexes are all perfectly positively
correlated with each other—and the generic portfolio volatility for any correlation structure of
the market. Clearly, employing the convexity property of volatility, we have that D R(π) ≥ 1.
As shown by Choueifaty et al. (2013), the Most Diversified (MD) portfolio, namely the
optimal portfolio that maximizes the diversification ratio (13), can be found by solving the
following (convex) quadratic programming problem
(i)
J K ( j)
J K
min σh(i)k( j) yh(i) yk( j)
y i=1 h(i)=1 j=1 k( j)=1
s.t.
( j)
(14)

J K
yk( j) σk( j) = 1
j=1 k( j)=1
yk( j) ≥ 0 k( j) = 1, . . . , K ( j) and j = 1, . . . , J
∗
yk(
The normalized portfolio weights are πk(
MD =
j) J
j)
K (i) ∗
with k( j) = 1, . . . , K ( j)
i=1 h(i)=1 yh(i)
and j = 1, . . . , J , where y∗
is the optimal solution of Problem (14). Hence, the optimal
portfolio choice π M D identifies the relative contribution of each index to the world’s mood,
so that the distance between the volatility of the world’s mood and the volatility relative to
the worst scenario (i.e., the case where all moods are perfectly correlated) is maximum.
5 Empirical experiments
This section proposes the empirical validation of the theoretical setting described above. We
first present the data used and then report the results.
5.1 Data
This section discusses the data collected and organized for implementing the optimal alloca-
tion models proposed in this study. Our experiment focuses on countries with a development
level that ensures access to online sources and financial markets. We have chosen a set of coun-
tries based on the Human Development Index (HDI) used by the United Nations Development
Programme (UNDP)’s Human Development Report Office. This index includes indicators
of life quality, education, and standard of living. We selected countries categorised as “very
high human developed countries” in Table 1 of UNDP (2019), based on 2018 data, namely
those with an HDI greater than 0.8. Additionally, China, with an HDI of 0.75, is included
due to its central role in the pandemic’s spread, bringing the total to 63 countries.
In each selected country, the term “coronavirus” is translated into the most spoken lan-
guage using Google Translate. These translations are listed in Table 1. Online searches for
123
Table 1 The table reports the country name, translation from English of “coronavirus”, in the most used
language in the respective country. A statistical summary of the volumes of searches (from Google Trends)
along the period of reference
Country Terms µ σ Skew Kurt µ/σ
Andorra Corona virus 1.789 8.699 6.988 57.651 0.206

Argentina coronavirus 12.769 15.724 2.985 10.410 0.812
Australia coronavirus 13.080 16.993 2.583 6.574 0.770
Austria Coronavirus 8.962 13.481 3.859 17.115 0.665
Bahamas coronavirus 7.794 14.121 3.133 11.283 0.552
Bahrain 4.957 9.470 4.647 33.664 0.523
Barbados coronavirus 9.378 16.913 2.705 7.234 0.554
Belarus korona virus 5.910 14.094 3.568 13.643 0.419
Belgium coronavirus 12.036 15.742 2.669 7.336 0.765
Brunei Virus korona 1.082 9.713 9.456 90.811 0.111
Bulgaria koronavirus 14.403 14.626 2.329 7.197 0.985
Canada coronavirus 11.507 15.649 2.684 7.167 0.735
Chile coronavirus 8.961 13.358 3.411 14.141 0.671
China 5.815 12.446 3.113 11.741 0.467
Croatia koronavirus 17.143 16.323 1.718 3.959 1.050
Cyprus κoρων oïóσ 1.910 9.601 6.297 43.759 0.199
Czech Republic koronavirus 10.777 14.553 3.800 16.216 0.741
Denmark coronavirus 8.694 14.878 3.626 14.068 0.584
Estonia koroona viirus 2.443 9.918 6.181 46.038 0.246
Finland koronaviirus 1.889 8.788 7.327 64.551 0.215
France Corona virus 5.949 13.429 4.043 18.379 0.443
Germany Coronavirus 11.088 14.463 2.719 7.708 0.767
Greece κoρωνoïóσ 2.342 7.974 9.592 101.357 0.294
Hong Kong 13.900 16.678 2.091 5.531 0.833
Hungary koronavírus 14.084 16.162 2.499 6.742 0.871
Iceland kórónaveira 2.183 13.242 6.082 36.013 0.165
Ireland coronavirus 13.695 16.672 2.547 6.975 0.821
Israel 8.341 16.722 2.818 8.615 0.499
Italy coronavirus 13.731 16.933 2.882 9.239 0.811
Japan 16.255 18.273 2.154 4.644 0.890
Kazakhstan koronavirus 15.540 20.284 1.813 2.662 0.766
Kuwait 6.102 10.521 3.805 21.492 0.580
Latvia koronavı̄russ 4.778 12.392 4.206 20.845 0.386
Liechtenstein Coronavirus 5.380 12.513 3.817 17.307 0.430
Lithuania koronavirusas 11.839 17.571 2.632 6.801 0.674
Luxembourg Corona virus 4.563 12.499 4.416 22.394 0.365
Malaysia virus korona 3.664 10.643 4.898 28.955 0.344
Malta koronavirus 0.818 7.302 9.821 104.445 0.112
Montenegro virus korona 1.279 7.956 8.016 77.539 0.161
Netherlands coronavirus 10.071 16.422 2.886 8.424 0.613
123
Table 1 continued
Country Terms µ σ Skew Kurt µ/σ
New Zealand coronavirus 10.707 17.530 2.935 8.504 0.611

Norway koronavirus 3.542 11.088 5.951 40.026 0.319
Oman 7.119 10.359 3.358 19.464 0.687
Palau coronavirus 1.821 6.520 4.758 26.743 0.279
Poland Korona wirus 5.002 11.522 5.440 35.266 0.434
Portugal coronavírus 7.176 11.911 4.367 22.652 0.602
Qatar 7.539 12.359 2.802 11.217 0.610
Romania coronavirus 11.143 15.777 2.953 9.574 0.706
Russia koronavirus 10.164 13.017 3.618 17.458 0.781
Saudi Arabia 6.659 9.910 3.768 23.378 0.672
Seychelles coronavirus 5.865 12.682 3.442 15.934 0.462
Singapore 6.360 12.693 3.086 12.676 0.501
Slovakia koronavírus 12.509 16.508 3.268 11.438 0.758
Slovenia Corona virus 4.323 11.763 5.043 30.816 0.368
South Korea 3.850 8.926 5.788 39.068 0.431
Spain coronavirus 11.478 14.580 3.351 13.594 0.787
Sweden coronavirus 9.004 14.758 3.067 10.214 0.610
Switzerland Coronavirus 11.462 14.167 3.054 11.081 0.809
Turkey koronavirüs 20.009 16.292 1.687 4.490 1.228
United Arab Emirates 8.329 11.265 3.004 14.179 0.739
United Kingdom coronavirus 13.036 16.150 2.759 8.223 0.807
United States coronavirus 11.081 16.652 2.883 8.644 0.665
Uruguay coronavirus 9.727 12.777 3.435 13.926 0.761
these translations within each country are sourced from Google Trends, resulting in 63 time
series from 01/01/2020 to 31/12/2020.
By applying the method described in Sect. 3.1, we compile a time series of daily data
for each country, tracking the daily search volume for the term “coronavirus” in various
languages, over the specified period.
Table 1 and Fig. 1 illustrate the search patterns in each country.
Regarding financial data, we identify at least one stock index for each country listed in
Table 1.
Utilising Bloomberg’s “SECF” function, these indexes are identified and ranked using
Bloomberg’s proprietary relevance indicator, “R”, which ranges from 0 to 4, increasing with
relevance. We select stock indexes with the highest relevance (R = 4) and exclude countries
with indexes having R < 4. As a result, Andorra, Bahamas, Bahrain, Barbados, Belarus,
Brunei, Bulgaria, Croatia, Cyprus, Estonia, Iceland, Kazakhstan, Kuwait, Latvia, Liechten-
stein, Lithuania, Luxembourg, Malta, Montenegro, Palau, Seychelles, Slovakia, Slovenia,
and Uruguay are excluded. Notably, China and Finland each have two stock indexes with
R = 4, referred to as China 1, China 2, and Finland 1, Finland 2. The final dataset comprises
39 countries and 41 stock indexes, detailed in the first two columns of Table 2.
123
Fig. 1 Google Trends volumes of “coronavirus” translations in the most spoken languages of the respective
countries. The white areas at the beginning of each row in the heat map indicate the onset of interest in
COVID-19 in each country
The closing levels of these indexes, sourced from Bloomberg for the period 01/01/2020–
31/12/2020, are normalized as per Formula (1). Table 2 presents a statistical summary of
these normalized closing prices.
To calculate the mood indicator as per Formula (4), we exclude data corresponding to non-
trading days from each country/stock index’s time series. Due to the asynchronous nature
of global trading (e.g., Chinese indexes not traded during the Chinese New Year), the mood
index is not available for every day between 01/01/2020 and 31/12/2020. To address this, we
omit days where mood data is unavailable, resulting in a matrix A containing daily moods
across 106 days and 41 stock indexes. Table 3 provides a statistical summary of the mood
index for each stock index.
123
Table 2 Country names, stock indexes, Bloomberg’s tickers, and a statistical summary of the normalized
indexes’ closing prices (according to Formula 1) for the reference period
Country Bloomberg ticker μ σ Skew Kurt μ/σ
Argentina MERVAL Index 76,463 13,669 −0.587 −0.084 5594

Australia AS51 Index 85,193 7943 −0.071 −0.413 10,726
Austria ATX Index 74,682 11,725 0.950 −0.160 6369
Belgium BEL20 Index 81,295 8503 0.209 −0.228 9561
Canada SPTSX Index 89,259 7555 −1.014 0.894 11,814
Chile IPSA Index 79,883 7923 0.329 0.567 10,082
China SHSZ300 Index 84,108 8670 −0.063 −1.510 9701
China SHCOMP Index 90,077 6689 −0.214 −1.498 13,466
Czech Republic PX Index 81,585 8303 0.597 0.072 9826
Denmark OMXC25 Index 81,969 9273 −0.354 −0.540 8839
Finland HEX Index 87,660 8698 −0.996 0.314 10,078
Finland HEX25 Index 88,715 8726 −1.070 0.449 10,167
France CAC Index 83,091 8871 0.234 −0.518 9367
Germany DAX Index 89,477 8901 −1.250 0.710 10,052
Greece ASE Index 72,587 12,084 1.044 0.010 6007
Hong Kong HSI Index 87,079 5165 0.459 −0.258 16,859
Hungary BUX Index 80,301 9012 0.707 −0.638 8910
Ireland ISEQ Index 84,251 9851 −0.474 −0.430 8553
Israel TA-35 Index 82,338 7571 0.940 0.130 10,875
Italy FTSEMIB Index 78,801 9306 0.285 −0.498 8468
Japan TPX Index 87,787 6814 −0.552 −0.057 12,884
Malaysia FBMKLCI Index 89,865 5466 −0.817 0.301 16,440
Netherlands AEX Index 88,983 7361 −0.795 0.755 12,089
New Zealand NZSE50FG Index 87,170 6544 −0.743 0.660 13,320
Norway OSEBX Index 86,819 7817 −0.622 −0.051 11,106
Oman MSM30 Index 87,346 5170 1.215 0.189 16,896
Poland WIG20 Index 81,385 8600 0.264 −0.197 9464
Portugal PSI20 Index 82,335 7804 0.701 −0.237 10,550
Qatar DSM Index 89,622 6312 −0.273 −0.943 14,200
Romania BET Index 87,044 6869 −0.039 −0.292 12,672
Russia RTSI$ Index 76,471 9935 0.424 0.097 7697
Saudi Arabia SASEIDX Index 88,224 8391 −0.461 −0.862 10,515
Singapore STI Index 82,554 7715 1.038 −0.119 10,701
South Korea KOSPI Index 77,261 9562 −0.030 0.093 8080
Spain IBEX Index 75,460 10,186 1057 −0.090 7408
Sweden OMX Index 89,239 7584 −0.886 0.055 11,767
Switzerland SMI Index 90,058 4881 −1.059 1.919 18,451
Turkey XU100 Index 76,827 8700 0.158 0.202 8830
United Arab Emirates DFMGI Index 78,363 10,495 0.424 −0.670 7467
United Kingdom UKX Index 81,779 7879 0.922 0.266 10,380
United States SPX Index 85,671 8482 −0.702 0.189 10,100
123
Table 3 Country names, stock indexes, Bloomberg’s tickers, and a statistical summary of the mood index
calculated using Formula (4)
Country Bloomberg ticker μ σ Skew Kurt μ/σ
Argentina MERVAL Index 0.5004 0.0135 − 5.4665 44.5997 37.0107

Australia AS51 Index 0.5017 0.0090 − 1.9556 28.9805 56.0020
Austria ATX Index 0.5016 0.0058 − 1.4123 13.3691 85.8751
Belgium BEL20 Index 0.5014 0.0119 − 7.3179 69.4856 41.9649
Canada SPTSX Index 0.5006 0.0105 − 5.5138 48.1026 47.7653
Chile IPSA Index 0.4995 0.0228 − 9.2079 90.9126 21.9058
China SHSZ300 Index 0.5042 0.0339 − 0.8187 17.8851 14.8817
China SHCOMP Index 0.5043 0.0339 − 0.8015 17.8303 14.8800
Czech Republic PX Index 0.5016 0.0074 − 5.9728 52.9433 67.6568
Denmark OMXC25 Index 0.5014 0.0135 − 0.3399 34.4556 37.1294
Finland HEX Index 0.5020 0.0133 0.5475 30.5426 37.7080
Finland HEX25 Index 0.5021 0.0133 0.5619 30.3767 37.6361
France CAC Index 0.5010 0.0084 − 5.3866 44.9941 59.3906
Germany DAX Index 0.5020 0.0069 0.5139 15.8341 72.4869
Greece ASE Index 0.5019 0.0083 1.5563 35.6850 60.5857
Hong Kong HSI Index 0.5073 0.0255 − 0.4115 6.9257 19.9027
Hungary BUX Index 0.5006 0.0138 − 7.3394 66.6646 36.3924
Ireland ISEQ Index 0.5016 0.0084 − 2.3469 19.8482 59.3838
Israel TA− 35 Index 0.5014 0.0203 − 1.9593 14.3032 24.6684
Italy FTSEMIB Index 0.5021 0.0065 1.4757 10.9006 77.5981
Japan TPX Index 0.5017 0.0166 − 4.2209 44.9289 30.2284
Malaysia FBMKLCI Index 0.5033 0.0118 3.7398 19.7559 42.7125
Netherlands AEX Index 0.5017 0.0098 − 4.7745 47.9132 51.0409
New Zealand NZSE50FG Index 0.5016 0.0121 − 3.0266 35.4076 41.3005
Norway OSEBX Index 0.5021 0.0067 8.2774 78.8757 75.3277
Oman MSM30 Index 0.5045 0.0153 5.8698 47.2990 32.9650
Poland WIG20 Index 0.5017 0.0135 − 2.2056 42.5489 37.2614
Portugal PSI20 Index 0.5030 0.0098 2.3569 14.3185 51.5228
Qatar DSM Index 0.5057 0.0147 2.5489 13.2770 34.3981
Romania BET Index 0.5019 0.0086 − 3.3010 38.1974 58.0270
Russia RTSI$ Index 0.5026 0.0066 3.3604 22.7373 76.7256
Saudi Arabia SASEIDX Index 0.5028 0.0060 4.9223 32.5224 83.4407
Singapore STI Index 0.5049 0.0186 1.1760 5.8426 27.1833
South Korea KOSPI Index 0.5025 0.0083 6.2256 50.5340 60.3632
Spain IBEX Index 0.5013 0.0091 − 5.7556 52.7410 55.2326
Sweden OMX Index 0.5017 0.0104 1.0803 21.5532 48.2034
Switzerland SMI Index 0.5018 0.0066 − 1.1947 18.9062 76.2267
Turkey XU100 Index 0.5014 0.0151 − 2.7749 14.3304 33.2976
United Arab Emirates DFMGI Index 0.5036 0.0074 3.2388 16.2601 67.6691
United Kingdom UKX Index 0.5004 0.0122 − 5.4345 45.8719 40.9221
United States SPX Index 0.5002 0.0117 − 5.1739 34.6032 42.6016
123
Fig. 2 MV efficient frontier in terms of relative portfolio weights with respect to the normalized target level
of the expected world’s mood
5.2 Results and discussion
In Fig. 2, we display the composition profile of the MV efficient frontier, showing the variation
in the portfolio composition as the target level of the expected mood η shifts from ηmin to
ηmax . For enhanced readability, the x-axis of Fig. 2 denotes the normalized target level
η−ηmin
η̃ = ηmax −ηmin ∈ [0, 1]. Specifically, we present the Pareto-optimal solutions of Problem (8),
illustrating the relative contribution of each index to the world’s mood. These configurations
optimally balance maximizing the expected mood of the world and minimizing its variance.
According to the mathematical properties of the MV model, the diversification of stock
indexes generating the lowest volatility of the world’s mood is high. While approaching
the highest targets of the expected world’s mood portfolios become more concentrated,
particularly in the Hong Kong and Qatar indexes. Increasing the world’s expected mood
level in Model (8) tends to prioritize countries with higher expected mood values in the
efficient solutions, that happens because of the model design.
To further elucidate the behaviour of optimal MV portfolios in terms of the world’s mood,
Fig. 3 depicts the time evolution of the world’s mood for three distinct portfolios: the Mini-
mum Variance (MinV) portfolio, the Pareto-optimal portfolio with maximum expected mood
(MaxExpMood, that corresponds to a portfolio with only one country, in this case, Hong
Kong), and the Equally Weighted (EW) portfolio, which distributes the contribution to the
global mood equally among all indexes. Here, πk( E W = 1 for k( j) = 1, . . . , K ( j) and
j) n
j = 1, . . . , n. The MinV portfolio, chosen to minimize volatility, contrasts starkly with the
MaxExpMood portfolio, which, while offering the highest expected mood, exhibits consid-
erable instability over time.
Figure 4 uses a heatmap to visualize each country’s relative contribution to the total mood
of the world for the MV Pareto-optimal portfolios. The y-axis lists the countries, while the
x-axis displays 100 equally-spaced target levels of the world’s expected mood between ηmin
and ηmax . The colour spectrum from light yellow to dark red represents low to high impact on
the world’s mood, respectively. White blocks indicate minimal impact of a country’s financial
123
Fig. 3 Daily time evolution of the world’s mood for the MinV, EW, and MaxExpMood portfolios during the
pandemic year
markets and COVID-19 anxiety on the global mood. Policymakers can use this visualisation
to discern which countries significantly influence the global mood. Again, we observe that to
achieve less fluctuation in world mood values, more countries should be considered relevant
in Pareto-optimal solutions.
In Fig. 5, we present the profile of the Pareto-optimal portfolios obtained by Problem (11),
η−ηmin
varying the normalized target level of the mood η̃ = ηmax −ηmin ∈ [0, 1]. In this empirical
experiment, the confidence level is set at ε = 0.10. At a fixed η̃, the portfolio weights minimize
the mood values in the most pessimistic scenarios without penalizing the optimistic ones.
Notably, as observed by Cesarone et al. (2013); Mansini et al. (2007) for equity portfolios,
the portfolio with minimum CVaRD tends to be less diversified compared to the minimum
variance portfolio.
Figure 6 illustrates the time evolution of the world’s mood for the Minimum CVaRD
(MinCVaRD) portfolio, the MaxExpMood portfolio, and the EW portfolio. The results mirror
those in Fig. 3, with the risk-minimization portfolio leading to a relatively stable mood around
the fair value of 0.5.
Figure 7 offers a heatmap visualization of each country’s relative contribution to the total
mood of the world for the Mean-CVaRD Pareto-optimal portfolios (similar to Fig. 5 but with
a different visualization technique). In line with the representation in Fig. 4, on the y-axis
we report the countries, and on the x-axis, 100 equally-spaced target levels of the expected
mood of the world, assuming the same meaning for the colour of the blocks. Different from
the MV model, the Mean-CVaRD optimal portfolios with low target levels of the expected
mood assigns more relevance to Saudi Arabia, South Corea, Norway and Malaysia to reduce
pessimistic fluctuations of the world’s mood. Model (11) suggests that fewer countries are
relevant in Pareto-optimal solutions to contain downside fluctuations of world mood values
compared to those obtained by solving Model (8).
In Fig. 8 we compare the relevance of countries in the world’s mood as determined by
the RP model (12) and the Most Diversified Portfolio (MDP, namely the normalized optimal
solution of Model (14)), along with the MinV and MinCVaRD portfolios. By construction,
the RP portfolio includes all countries in composing the world’s mood. Thus, the resulting
123
Fig. 4 Heatmap visualization of the MV Pareto-optimal contributions of all countries to the global mood
123
Fig. 5 Mean-CVaRD efficient frontier showing relative portfolio weights with respect to the normalized target
level of the expected world’s mood
Fig. 6 Daily time evolution of the world’s mood for the MinCVaRD, EW, and MaxExpMood portfolios during
the pandemic year
scores (i.e., the weight in percentage shown in Fig. 8) associated with each market in a
country tend to have uniform weights across markets, preventing any single country from
dominating. In MDP, countries like Denmark, Oman, Greece, and Chile are most relevant.
This allocation ensures that the fluctuations caused by waves of pessimism and optimism are
as small as possible, and also as far as possible from the worst case where the turbolences of
the stock markets and the mood of the countries are perfectly correlated. We point out that
123
Fig. 7 Heatmap visualization of the Mean-CVaRD Pareto-optimal contributions of all countries to the global mood
123
123
Fig. 8 Heatmap visualization showing relative weights (percentage) for the MinV, RP, Most Diversified (MD), and MinCVaRD portfolios
Table 4 Descriptive statistics of the daily mood of the world for the EW, the MinV, the MinCVaRD, the
MaxExpMood, the RP, and the MDP strategies
μ σ Skew Kur Min Max
EW 0.50217 0.0048 − 2.21 22.7 0.471 0.522

MinV 0.50243 0.0025 0.36 9.9 0.491 0.514
MinCVaRD 0.50260 0.0035 3.21 16.4 0.497 0.524
MaxExpMood 0.50725 0.0256 − 0.41 9.5 0.384 0.601
RP 0.50244 0.0034 − 0.77 16.6 0.484 0.519
MDP 0.50248 0.0029 − 0.22 12.5 0.487 0.516
the MinCVaRD model tends to favour countries with high mood kurtosis levels (see Table 3)
induced by a noticeable number of days where the Google searches decrease, and the stock
indexes’ prices increase. Thus, such a selection seems to link with the model’s focus on
penalizing pessimistic fluctuations.
Finally, Table 4 highlights important features of the daily mood of the world when the
considered portfolio models are implemented. Such a table can also be interpreted along
with Figs. 3 and 6. Specifically, we notice that all the models are associated with an average
mood close to 0.5. The skewness oscillates between positive and negative values, to illustrate
different symmetry properties of the models. In all the cases the skewness is around zero,
except for EW—showing a remarkably negative value of the skewness—and MinCVaRD—
that presents a high level of positive skewness. Interestingly, MaxExpMood is the strategy
with the highest level of fluctuations, hence leading to extreme values of the mood as standard
ones. For this reason, this model does not present the highest level of kurtosis. Differently,
we notice that EW has quite stable behavior, with values around 0.5. However, the presence
of some specific deviations from the mean level leads to a high level of kurtosis.
6 Conclusive remarks
This paper proposes a portfolio decision analysis framework to analyse the world’s reaction
to COVID-19. We measure this reaction by combining the anxiety generated by the pandemic
with the stock markets’ perception of performance.
We applied and compared different portfolio selection models, each tailored to describe
the world’s mood optimally under the assumption of “rationality”, that is, a preference for
prosperity over decline. Our goals included pursuing a high and stable level of mood, but
also for avoiding extremes of overpessimism. We also explored a target aimed at fostering
solidarity, cooperation, and a shared sense of belonging. Supranational institutions, such as
the European Commission, the World Health Organization, or the World Bank, can utilise
our findings to analyse various scenarios and guide their focus, especially when the relevance
of certain countries is paramount. For instance, the European Union can employ our portfolio
decision analysis in the context of the National Recovery and Resilience Plan, to strategically
allocate funds to countries for developing actions and policies supporting the socio-economic
environment post-pandemic (refer to the Introduction for a thorough discussion).
The outcomes highlight regional differences when the adopted portfolio strategy varies.
For example, European countries play a leading role in the MinV context, while Asian
countries are predominant in the MinCVaRD portfolio.
123
Our approach disentangles the challenge of selecting countries based on their unique
mood realisations, facilitating a global analysis of the world’s reaction to COVID-19. It
enables analysts to focus on countries with assigned relevance according to the predetermined
target. The empirical experiment showcases the informative content of our proposal, revealing
different clusters of countries driven by weight distributions, depending on the chosen model
and target.
This study’s foundation lies in the selected data type for describing citizens’ mood and the
proposed mood indicator; these elements represent a clear limitation. However, the validity of
the models’ components is supported by existing studies (see Cerqueti & Ficcadenti, 2023).
New data-based conceptualizations of anxiety and novel indicators might lead to alterna-
tive portfolio decision models, offering results comparable to ours.
Several research ideas emerge from this study. Firstly, developing a set of portfolio mod-
els with an entropy-based objective function could describe the world’s mood in terms of
closeness to a uniform distribution. In so doing, one can analyse the assigned relevance
to the individual stock markets/countries for concentrating attention on a few paradigmatic
realities—e.g., by minimising the Shannon entropy of M(π), one can focus on the most criti-
cal situations, maintaining vigilance over the pandemic and mitigating COVID-19 spread due
to fear-driven defensive behaviour. Secondly, a data science approach could provide detailed
insights into citizens’ reactions at the country level. For example, conducting a rank-size
analysis might reveal inner structures within disaggregated data and allow for forecasting the
evolution of the world’s mood. Significantly, the outcomes of the portfolio models can be
interpreted through cluster analysis, optimising a similarity-based distance between coun-
tries. This approach can yield meaningful classifications of countries in terms of their reaction
to the pandemics. These challenging topics are reserved for future research.
Funding Open access funding provided by Università degli Studi Roma Tre within the CRUI-CARE Agree-
ment.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence,
and indicate if changes were made. The images or other third party material in this article are included in the
article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is
not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
References
Acerbi, C., & Tasche, D. (2002). On the coherence of expected shortfall. Journal of Banking & Finance, 26(7),
1487–1503.
Ararat, C., Cesarone, F., Pınar, M. C. & Ricci, J. M. (2024). MAD risk parity portfolios. Annals of Operations
Research (in Press).
Asmundson, G. J., & Taylor, S. (2020). Coronaphobia: Fear and the 2019-nCoV outbreak. Journal of Anxiety
Disorders, 70, 102196.
Baker, E., Bosetti, V., & Salo, A. (2020). Robust portfolio decision analysis: An application to the energy
research and development portfolio problem. European Journal of Operational Research, 284(3), 1107–
1120.
Barbati, M., Corrente, S., & Greco, S. (2020). A general space–time model for combinatorial optimization
problems (and not only). Omega, 96, 102067.
Barbati, M., Greco, S., Kadziński, M., & Słowiński, R. (2018). Optimization of multiple satisfaction levels in
portfolio decision analysis. Omega, 78, 192–204.
123
Barro, R. J. (2006). Rare disasters and asset markets in the twentieth century. The Quarterly Journal of
Economics, 121(3), 823–866.
Bellini, F., Cesarone, F., Colombo, C., & Tardella, F. (2021). Risk parity with expectiles. European Journal
of Operational Research, 291(3), 1149–1163.
Binder, C. (2020). Coronavirus fears and macroeconomic expectations. Review of Economics and Statistics,
102(4), 721–730.
Carleo, A., Cesarone, F., Gheno, A., & Ricci, J. M. (2017). Approximating exact expected utility via portfolio
efficient frontiers. Decisions in Economics and Finance, 40(1–2), 115–143.
Castellano, R., Cerqueti, R., Clemente, G. P., & Grassi, R. (2021). An optimization model for minimizing
systemic risk. Mathematics and Financial Economics, 15, 103–129.
Cerqueti, R., & Ficcadenti, V. (2023). Anxiety about the pandemic and trust in financial markets. The Annals
of Regional Science. https://doi.org/10.1007/s00168-023-01243-0
Cesarone, F. (2020). Computational finance: MATLAB® oriented modeling. London: Routledge.
Cesarone, F., & Colucci, S. (2018). Minimum risk versus capital and risk diversification strategies for portfolio
construction. Journal of the Operational Research Society, 69(2), 183–200.
Cesarone, F., Gheno, A., & Tardella, F. (2013). Learning and holding periods for portfolio selection models:
A sensitivity analysis. Applied Mathematical Sciences, 7(100), 4981–4999.
Cesarone, F., Giacometti, R., Martino, M. L. & Tardella, F. (2023). A return-diversification approach to
portfolio selection. arXiv:2312.09707.
Cesarone, F., Mango, F., Mottura, C. D., Ricci, J. M., & Tardella, F. (2020). On the stability of portfolio
selection models. Journal of Empirical Finance, 59, 210–234.
Cesarone, F., Scozzari, A., & Tardella, F. (2013). A new method for mean-variance portfolio optimization with
cardinality constraints. Annals of Operations Research, 205(1), 213–234.
Cesarone, F., Scozzari, A., & Tardella, F. (2020). An optimization-diversification approach to portfolio selec-
tion. Journal of Global Optimization, 76(2), 245–265.
Cesarone, F., & Tardella, F. (2017). Equal risk bounding is better than risk parity for portfolio selection. Journal
of Global Optimization, 68(2), 439–461.
Choueifaty, Y., & Coignard, Y. (2008). Toward maximum diversification. The Journal of Portfolio Management,
35(1), 40–51.
Choueifaty, Y., Froidure, T., & Reynier, J. (2013). Properties of the most diversified portfolio. Journal of
Investment Strategies, 2(2), 49–70.
Chowdhury, M. M. H., & Quaddus, M. A. (2015). A multiple objective optimization based QFD approach
for efficient resilient strategies to mitigate supply chain vulnerabilities: The case of garment industry of
Bangladesh. Omega, 57, 5–21.
Cranmer, A., Baker, E., Liesiö, J., & Salo, A. (2018). A portfolio model for siting offshore wind farms with
economic and environmental objectives. European Journal of Operational Research, 267(1), 304–314.
Da, Z., Engelberg, J., & Gao, P. (2011). In search of attention. The Journal of Finance, 66(5), 1461–1499.
Da, Z., Engelberg, J., & Gao, P. (2014). The sum of all FEARS investor sentiment and asset prices. The Review
of Financial Studies, 28(1), 1–32.
Dave, D., Friedson, A. I., Matsuzawa, K., & Sabia, J. J. (2021). When do shelter-in-place orders fight COVID-
19 best? Policy heterogeneity across states and adoption time. Economic Inquiry, 59(1), 29–52.
De Grauwe, P. (2011). Animal spirits and monetary policy. Economic Theory, 47(2–3), 423–457.
Fetzer, T., Hensel, L., Hermle, J., & Roth, C. (2020). Coronavirus perceptions and economic anxiety. Review
of Economics and Statistics. https://doi.org/10.1162/rest_a_00946
Filippi, C., Guastaroba, G., & Speranza, M. G. (2020). Conditional value-at-risk beyond finance: A survey.
International Transactions in Operational Research, 27(3), 1277–1319.
Gabaix, X. (2012). Variable rare disasters: An exactly solved framework for ten puzzles in macro-finance. The
Quarterly Journal of Economics, 127(2), 645–700.
Gans, J. (2020). Economics in the age of COVID-19. Cambridge: MIT Press.
Garfin, D. R., Silver, R. C., & Holman, E. A. (2020). The novel coronavirus (COVID-2019) outbreak: Ampli-
fication of public health consequences by media exposure. Health Psychology, 39(5), 355–357.
Gemici-Ozkan, B., Wu, S. D., Linderoth, J. T., & Moore, J. E. (2010). OR PRACTICE-R&D project portfolio
analysis for the semiconductor industry. Operations Research, 58(6), 1548–1563.
Goel, S., Cagle, S., & Shawky, H. (2017). How vulnerable are international financial markets to terrorism?
An empirical study based on terrorist incidents worldwide. Journal of Financial Stability, 33, 120–132.
Goodell, J. W. (2020). COVID-19 and finance: Agendas for future research. Finance Research Letters, 35,
101512.
Gourio, F. (2012). Disaster risk and business cycles. American Economic Review, 102(6), 2734–2766.
Grushka-Cockayne, Y., Reyck, B. D., & Degraeve, Z. (2008). An integrated decision-making approach for
improving European air traffic management. Management Science, 54(8), 1395–1409.
123
Hollo, D., Kremer, M. & Lo Duca, M. (2012). CISS-a composite indicator of systemic stress in the financial
system. In: ECB Working paper, 1426.
Jouini, E., & Napp, C. (2011). Unbiased disagreement in financial markets, waves of pessimism and the
risk-return trade-off. Review of Finance, 15(3), 575–601.
Kangaspunta, J., Liesiö, J., & Salo, A. (2012). Cost-efficiency analysis of weapon system portfolios. European
Journal of Operational Research, 223(1), 264–275.
Kaplanski, G., & Levy, H. (2010). Sentiment and stock prices: The case of aviation disasters. Journal of
Financial Economics, 95(2), 174–201.
Kong, E., & Prinz, D. (2020). Disentangling policy effects using proxy data: Which shutdown policies affected
unemployment during the COVID-19 pandemic? Journal of Public Economics, 189, 104257.
Lee, S. A., Jobe, M. C., Mathis, A. A., & Gibbons, J. A. (2020). Incremental validity of coronaphobia:
Coronavirus anxiety explains depression, generalized anxiety, and death anxiety. Journal of Anxiety
Disorders, 74, 102268.
Lewnard, J. A., & Lo, N. C. (2020). Scientific and ethical basis for social-distancing interventions against
COVID-19. The Lancet Infectious Diseases, 20(6), 631–633.
Li, Q., Guan, X., Wu, P., Wang, X., Zhou, L., Tong, Y., Ren, R., Leung, K. S., Lau, E. H., Wong, J. Y., et al.
(2020). Early transmission dynamics in Wuhan, China, of novel coronavirus-infected pneumonia. New
England Journal of Medicine, 382, 1199–1207.
Liesiö, J., Andelmin, J., & Salo, A. (2020). Efficient allocation of resources to a portfolio of decision making
units. European Journal of Operational Research, 286(2), 619–636.
Liesiö, J., Salo, A., Keisler, J. M., & Morton, A. (2020). Portfolio decision analysis: Recent developments and
future prospects. European Journal of Operational Research. https://doi.org/10.1016/j.ejor.2020.12.015
Lowe, R. A., & Ziedonis, A. A. (2006). Overoptimism and the performance of entrepreneurial firms. Man-
agement Science, 52(2), 173–186.
Maggi, M. & Uberti, P. (2018). Google searches for portfolio management: A risk and return analysis. In
Mathematical and statistical methods for actuarial sciences and finance: MAF 2018, pp. 461–465.
Springer International Publishing.
Maggi, M., & Uberti, P. (2021). Google search volumes for portfolio management: Performances and asset
concentration. Annals of Operations Research, 299(1–2), 163–175.
Maillard, S., Roncalli, T., & Teiletche, J. (2010). The properties of equally weighted risk contribution portfolios.
The Journal of Portfolio Management, 36(4), 60–70.
Mann, F. D., Krueger, R. F., & Vohs, K. D. (2020). Personal economic anxiety in response to COVID-19.
Personality and Individual Differences, 167, 110233.
Mansini, R., Ogryczak, W., & Speranza, M. (2007). Conditional value at risk and related linear programming
models for portfolio optimization. Annals of Operations Research, 152, 227–256.
Markowitz, H. (1952). Portfolio selection. The Journal of Finance, 7(1), 77–91.
Markowitz, H. (1959). Portfolio selection: Efficient diversification of investments. In: Cowles foundation for
research in economics at Yale University, Monograph 16. Wiley, New York.
Mastorakis, K., & Siskos, E. (2016). Value focused pharmaceutical strategy determination with multicriteria
decision analysis techniques. Omega, 59, 84–96.
Milani, F. (2021). COVID-19 outbreak, social response, and early economic effects: A global VAR analysis
of cross-country interdependencies. Journal of Population Economics, 34(1), 223–252.
Mild, P., Liesiö, J., & Salo, A. (2015). Selecting infrastructure maintenance projects with robust portfolio
modeling. Decision Support Systems, 77, 21–30.
Nikkinen, J., & Peltomäki, J. (2020). Crash fears and stock market effects: Evidence from web searches and
printed news articles. Journal of Behavioral Finance, 21(2), 117–127.
Nikolopoulos, K., Punia, S., Schäfers, A., Tsinopoulos, C., & Vasilakis, C. (2021). Forecasting and planning
during a pandemic: COVID-19 growth rates, supply chain disruptions, and governmental decisions.
European Journal of Operational Research, 290(1), 99–115.
Phillips, L. D., & Bana e Costa, C. A. (2007). Transparent prioritisation, budgeting and resource allocation
with multi-criteria decision analysis and decision conferencing. Annals of Operations Research, 154(1),
51–68.
Rockafellar, R., & Uryasev, S. (2000). Optimization of conditional value-at-risk. Journal of Risk, 2(3), 21–42.
Rockafellar, R., Uryasev, S., & Zabarankin, M. (2006). Generalized deviations in risk analysis. Finance and
Stochastics, 10(1), 51–74.
Roncalli, T. (2014). Introduction to risk parity and budgeting. Boca Raton: CRC Press.
Salo, A., Keisler, J. & Morton, A. (2011). Portfolio decision analysis: Improved methods for resource alloca-
tion, vol. 162. Springer Science & Business Media.
123
Sarykalin, S., Serraino, G., & Uryasev, S. (2008). Value-at-risk vs. conditional value-at-risk in risk management
and optimization. In State-of-the-Art decision-making tools in the information-intensive age, pp. 270–
294. Informs.
Susskind, D. & Vines, D. (2020). The economics of the COVID-19 pandemic: An assessment. Oxford Review
of Economic Policy, 36(Supplement_1), S1–S13.
Toppila, A., Liesiö, J. & Salo, A. (2011). A resource allocation model for R&D investments: A case study in
telecommunication standardization. In Portfolio decision analysis, pp. 241–258. Springer.
Tsao, Y.-C., Raj, P. V. R. P., & Yu, V. (2019). Product substitution in different weights and brands considering
customer segmentation and panic buying behavior. Industrial Marketing Management, 77, 209–220.
UNDP. (2019). Human development report 2019. United Nations. https://doi.org/10.18356/838f78fd-en
Ussai, S., Castelpietra, G., Mariani, I., Casale, A., Missoni, E., Pistis, M., Monasta, L., & Armocida, B. (2022).
What is next for public health after COVID-19 in Italy? Adopting a youth-centered care approach in mental
health services. International Journal of Environmental Research and Public Health, 19(22), 14937.
Vollmer, M. A., Glampson, B., Mellan, T., Mishra, S., Mercuri, L., Costello, C., Klaber, R., Cooke, G., Flaxman,
S. & Bhatt, S. (2021). A unified machine learning approach to time series forecasting applied to demand
at emergency departments. BMC Emergency Medicine, 21(9).
Lyócsa, Štefan, Baumöhl, E., Výrost, T., & Molnár, P. (2020). Fear of the coronavirus and the stock markets.
Finance Research Letters, 36, 101735.
Zhu, N., Zhang, D., Wang, W., Li, X., Yang, B., Song, J., Zhao, X., Huang, B., Shi, W., Lu, R., et al. (2020).
A novel coronavirus from patients with pneumonia in china, 2019. New England Journal of Medicine,
382, 727–733.
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
123

Portfolio Decision Analysis For Pandemic Sentiment Assessment Based On Finance and Web Queries

Uploaded by

Copyright:

Available Formats

Portfolio Decision Analysis For Pandemic Sentiment Assessment Based On Finance and Web Queries

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Portfolio Decision Analysis For Pandemic Sentiment Assessment Based On Finance and Web Queries

Uploaded by

Copyright:

Available Formats

Annals of Operations Research

Portfolio decision analysis for pandemic sentiment

Roy Cerqueti1,2 · Francesco Cesarone3 · Valerio Ficcadenti4

Received: 28 June 2023 / Accepted: 21 March 2024

Keywords Portfolio optimization · Statistical indicators · COVID-19 · Mood · Financial

3 Statistical indicator for the mood

3.1 Preliminaries and notation

3.2 Definition and properties of the mood indicator

We propose a time-dependent mood indicator based on the comparison of variations in daily

Consider t ∈ {1, . . . , T − 1}, j = 1, . . . , J and k( j) = 1, . . . , K ( j). We define

where, given H = p̄k( j) , w j , we have

4 Optimal portfolio models

4.1 Definition of the portfolio

The starting point of the analysis is a concept of time-dependent (non-financial) portfolio,

Furthermore, given j = 1, . . . , J , one can identify the relative contribution of country j

Interestingly, vector π(t; j) = (π 1 (t), . . . , π J (t)) is still a portfolio according to Definition

4.2 Portfolio selection models

4.2.1 Pareto-optimal risk-gain portfolios

4.2.2 Risk diversification strategies

Andorra Corona virus 1.789 8.699 6.988 57.651 0.206

New Zealand coronavirus 10.707 17.530 2.935 8.504 0.611

Argentina MERVAL Index 76,463 13,669 −0.587 −0.084 5594

Argentina MERVAL Index 0.5004 0.0135 − 5.4665 44.5997 37.0107

5.2 Results and discussion

EW 0.50217 0.0048 − 2.21 22.7 0.471 0.522

You might also like