Research Article — Fear of Contagion and Consumption Expectations (Rose, Ordine, De Luca, 2024)

Summary

A research article by Giuseppe Rose and Patrizia Ordine of the University of Calabria and Desirè De Luca of the University of Bologna, titled Fear of Contagion and Consumption Expectations: Evidence Using Government Communications of Case Fatality Rates during the Covid-19 Pandemic. The abstract states that the study uses Italian microdata from the Bank of Italy's Special Survey to examine fear of Covid-19 and consumption expectations during the Covid-19 pandemic. It states that information about mortality in cases of contagion differentiated between groups of different ages, and that a fuzzy regression discontinuity design is applied. It reports that results demonstrate that the impact of the pandemic on consumption expectations operates markedly through households' fear of contagion.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Full text

Fear of Contagion and Consumption Expectations:
Evidence Using Government Communications of
Case Fatality Rates during the Covid-19 Pandemic
Giuseppe Rose



 University of Calabria
Patrizia Ordine
 University of Calabria
Desirè De Luca
 University of Bologna



Research Article

Keywords: Consumption Expectations, Fear of Contagion, Covid-19 Pandemic, Case Fatality Rate, Fuzzy
Regression Discontinuity Design, In- strumental Variables.

Posted Date: May 2nd, 2024

DOI: https://doi.org/10.21203/rs.3.rs-4317601/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Additional Declarations: No competing interests reported.
Fear of Contagion and Consumption Expectations:

 Evidence Using Government Communications of

              Case Fatality Rates during the Covid-19

                                         Pandemic1


                Giuseppe Rosea2 , Patrizia Ordineb , Desirè De Lucac
       a
           Dept. of Economics, Statistics and Finance, University of Calabria, Italy

       b
           Dept. of Business Administration and Law, University of Calabria, Italy

                 c
                     Dept. of Statistical Sciences, University of Bologna, Italy




   1
     The authors declare they have no con‡ict of interests. Data are available upon
request.
   2
     Corresponding author. Email: giuseppe.rose@unical.it. Phone: +39-0984-
492446. Address: Department of Economics, Statistics and Finance, Via P. Bucci
Cubo 1C, 87036 Rende (CS), Italy.
                                 Abstract


Using Italian microdata from the Bank of Italy’s Special Survey, we study

the causal e¤ects of fear of Covid-19 on consumption expectations during

the Covid-19 pandemic. Identi…cation is achieved by exploiting the fact

that information disseminated by o¢cial press releases about mortality in

cases of contagion di¤erentiated between groups of di¤erent ages. Since the

50-59 and 60-69 (years of age) groups manifested a marked jump in the

mortality rate (2 deaths per hundred cases and 1 death per ten cases respec-

tively), we …rstly evaluate whether the fear of contagion and consumption

expectations are discontinuous at the threshold and then we apply a fuzzy

regression discontinuity design to evaluate the causal e¤ect of fear of con-

tagion on consumption expectations. Results demonstrate that the impact

of the pandemic on consumption expectations operates markedly through

households’ fear of contagion.

   Keywords: Consumption Expectations; Fear of Contagion; Covid-19

Pandemic; Case Fatality Rate, Fuzzy Regression Discontinuity Design; In-

strumental Variables.

   JEL Codes: D91, D84, D10.
1    Introduction

The need to …nd a response to economic crises brought about by the spread

of viruses is fueling the debate about the identi…cation of appropriate policy

tools that are able to limit recession without jeopardizing public health. In-

deed, the measures adopted to contain the recent Covid-19 pandemic turned

out to be at odds with those that promoted economic activity and aimed

at supporting employment and income. In recent research, Eichenbaum et

al. (2021) discuss the issue and argue that there is an inevitable trade-o¤

between the severity of the short-run recession caused by an epidemic and

its health consequences, while Chetty et al. (2020) argue that …scal policy

cannot restore full employment when an initial shock to consumer spend-

ing arises from health concerns. Furthermore, understanding what policy

measure is best to sustain consumption, income and employment during a

pandemic is also important in order to mitigate the consequences of massive

household savings made during the crisis which - once the risk of conta-

gion is over – may act as a sudden surge in aggregate demand and give rise

to an increase in the in‡ation rate (Cline, 2023). Indeed, during the recent

Covid-19 pandemic, consumers’ behavior changed and the propensity to save

reached a twenty-year high, while consumption decreased much more than

gross disposable income. Figure 1 depicts this occurrence in EU countries

                                      1
and the UK.


                          Figure 1 about here


   This sudden increase in the saving rate in response to the Covid-19

pandemic is related to government restrictions, such as lockdowns, and pre-

cautionary behavior which arised because of income uncertainty and higher

risk of unemployment. However, health concerns in themselves may also

have operated through the fear of contagion and generated a reduction in

consumption since people could have voluntarily reduced their shopping fre-

quency in order to lessen the infection risk. Moreover, the outbreak of a

pandemic might induce fear, anxiety, distress and other psychological dis-

orders, so leading to changes in consumption behavior. Investigating de-

terminants of spending decisions at the individual/household level during

epidemics plays a key role in setting appropriate policies to curb recessive

e¤ects.

   In this paper, we contribute to the comprehension of this issue by provid-

ing an empirical investigation to estimate the causal e¤ect of fear of conta-

gion on consumption expectations. This is important since - in the presence

of such a causal relationship - policy measures targeted to sustain household

income may turn out to have a very low impact on consumption during

a pandemic and then may be ine¤ective in stimulating aggregate demand,

                                     2
GDP and employment, and they may simply contribute to an increase in pri-

vate saving. Instead, in this scenario, policy instruments should be aimed at

restoring consumer con…dence and reducing the fear of contagion - assisted

by strengthening the National Health Care System and providing e¤ective

and timely health services - so that the contraction of aggregate demand

can be limited in a more e¤ective manner. However, it must be borne in

mind that identifying the causal e¤ect of a fear of contagion on consumer

behavior is not an easy task since unobserved individual components may

simultaneously a¤ect the two variables. Indeed, unobserved habits mak-

ing individuals limit their consumption may simultaneously determine their

fear of contagion so that parameters derived from using a simple regression

model are likely to be biased.

      Our empirical strategy is designed to tackle endogeneity issues. It is

grounded on Italian data providing information about i) consumption ex-

pectations and ii) questions designed to assess individuals’ fear of contagion.

Our identi…cation strategy exploits the fact that information about the prob-

ability of mortality in the case of contagion - de…ned as case fatality rate

(henceforth, CFR) - provided by the media to Italian citizens, was commu-

nicated together with a separating into age-groups.1 Since the groups 50-59

  1
    CFR indicates the proportion of individuals diagnosed with a disease who die from
that disease.


                                         3
and 60-69 were characterized by a sharp jump in mortality rates (2.7% to

10.8%), we can use this age-threshold by separating these two groups so as

to apply a fuzzy regression discontinuity design to assess the causal e¤ect of

fear on consumption expectations.

   Recent economic research looks at the determinants of consumers’ be-

havior during the pandemic, …nding important insights. Most of this re-

search evaluates how the pandemic e¤ected household spending by using

transactional data (Baker et al., 2020; Chronopoulos et al., 2020; Chen et

al., 2020; Andersen et al., 2020a) or retrospective surveys on consumption

(Coibion et al., 2020). Generally, a basic di¢culty with data measuring

actual consumption is that observed choices may be consistent with many

combinations of expectations and preferences and, especially in the pres-

ence of imposed lockdowns, actual choices may be forced by impediments

imposed by law.

   Our study adds to the existing research by evaluating consumption ex-

pectations rather than actual consumption choices. By means of data on

consumption expectations, it is possible to examine unconstrained potential

behavior by avoiding confounding problems relating to the occurrence of

government restrictions such as lockdowns. Indeed, despite the central role

economic analysis allocates to agents’ expectations in their decision mak-


                                      4
ing, the standard practice of economists has been to infer decision processes

from data on observed choices, often by making unveri…able assumptions

on expectations. Recently, a growing body of research has used subjective

expectations data relating to spending, income, employment, retirement, in-

‡ation, stock value, marriage and many other factors in order to understand

economic facts (de Bruin et al., 2023; de La Vande et al., 2023; Armantier

et al., 2022). This practice is generated by an awareness of the potential

value of expectations data for the purpose of predicting choice behavior and

evaluating how expectations are formed.

   The expectations data used in our study are derived from the second

wave of the Bank of Italy’s Special Survey of Italian Households (SSIH),

which is a survey that was made during the pandemic period covering about

2,000 households. The survey is designed to gather timely information on

the economic situation in households in Italy during the Covid-19 crisis,

asking questions about the way households perceived the economic conse-

quences of the pandemic, including data on consumption and saving deci-

sions. Using these data, we provide empirical evidence about the causal

relationship linking fear of contagion and expected consumption by using

a quasi-experimental setup based on a fuzzy regression discontinuity design

(FRDD) which addresses the potential endogeneity arising when one or more


                                     5
explanatory variables are correlated with unobservables.2 We exploit the

fact that, according to data from Covid-19 integrated surveillance reports

coordinated by the Istituto Superiore di Sanità, the CFR in Italy increased

steeply in individuals aged 60-69. To be precise, CFR during the …rst pan-

demic wave - and hence in the months just before the survey was submitted

- was 2,7% on average for those aged 50-59 and 10,8% for those in the 60-69

age group, meaning a perception of about 3 deaths per hundred infected

people for the …rst group as opposed to 1 death per ten infected people for

the second. Crucially, the statistics on the severity of the consequences of

the disease were conveyed daily by the mass media while considering these

age groups. Therefore, it seems plausible to expect the perceived fear of

Covid-19 infection to show an increase at the threshold separating these two

groups. Our FRDD relies on the fact that a higher level of perceived fear is

de…ned as a discontinuous function of the respondent’s age, considering that

we observe similar individuals at the threshold who have a di¤erent percep-

tion of infection risks since they have been informed by o¢cial sources that

they have a di¤erent resistance to the virus. Using this identi…cation strat-

egy - corroborated by a series of robustness exercises, as well as by evidence

supporting our main assumptions -, we show that the fear of contagion has

  2
      See Angrist and Pischke (2009) and Lee and Lemieux (2010).



                                           6
a direct negative impact on expected consumption decisions of a magnitude

that is almost 10 times that detected by a simple regression model.

    Our results highlight the relevance that a government’s communication

strategy on public health issues may have in determining agents’ behavior

and economic outcomes. This issue needs to be attentively considered since

it provides an additional tool to be utilized in economic policy in order to

get out of a recession.

    The paper proceeds as follows. The next section contains a review of the

existing literature on the impact of Covid-19 on consumer behavior, as well

as a review of empirical studies using data on expectations. In section 3,

the data set is described and the variables used in the analysis are de…ned.

Section 4 presents the identi…cation strategy while section 5 summarizes

the main results. Section 6 presents some robustness checks and section 7

concludes.



2     Literature

2.1   Consumption and the Covid-19 pandemic

A timely exploration of the impact of Covid-19 on consumer behavior has

been made possible by information technology that enables the real-time col-



                                     7
lection of transaction-level data and allows researchers to analyze consumer

expectations and spending patterns as they occur, whereas o¢cial statistics

at the time were often produced with a signi…cant delay. Early empirical

studies were mainly concerned with the timing and the heterogeneity of the

response of di¤erent expenditure categories during the outbreak. Baker et

al. (2020) uses transactional data for the US and observes that spending

increased sharply at the start of the pandemic, but also that this initial in-

crease was followed by a decrease of about 25-30% in the middle of March

2020. Other studies explore consumer spending behavior outside the US by

using high-frequency data. Chronopoulos et al. (2020) …nd that consumer

spending in the UK remained relatively stable in the early stages of the

Covid-19 pandemic, while, at a later stage, when the government imposed

lockdown, it declined quite signi…cantly. Spending reduction varied across

product categories, so that non-essential items such as dining, travel and en-

tertainment were the most severely hit, while, in the initial phase, grocery

consumption increased as households stockpiled essential goods. This evi-

dence is consistent with that of other studies, such as Chen et al. (2020) for

China and Andersen et al. (2020a) for Denmark. Other authors explore the

stay-at-home e¤ects on consumer behavior. Coibion et al. (2020) analyze

how the di¤erential timing of local lockdowns causally a¤ected household


                                      8
spending and macroeconomic expectations, …nding that local economic con-

ditions and household expectations about the economy were signi…cantly

a¤ected by government restrictions. Alexander and Karger (2021) show

that the stay-at-home orders caused a large reduction in spending in sec-

tors that require mobility and in-person physical interaction. Chetty et al.

(2020) report that high-income households reduced spending mainly because

of concerns about the fear of infection rather than the fear of a reduction

in income or wealth. These authors show that roughly two-thirds of the re-

duction in spending came from reduced spending on goods or services that

require in-person contact and thereby carried a risk of Covid-19 infection. In

addition, they …nd that spending on services that require personal contact

remained depressed even during the gradual reopening process. In line with

these results, Andersen et al. (2020b) exploit a natural experiment to in-

vestigate the e¤ects of the virus by considering the di¤erent orders aimed at

containing it in Denmark and Sweden, which were similarly exposed to the

pandemic, but implemented di¤erent restrictions on social and economic ac-

tivity. The authors …nd that the vast majority of the fall in economic activity

during the pandemic crisis can be attributed to the perceived risk of disease

rather than to government restrictions. The causal e¤ect of government pol-

icy on the economy during the initial spread of Covid-19 is also examined


                                      9
by Goolsbee and Syverson (2021). Their …ndings suggest that the largest

decrease in commercial activity was due to individuals’ voluntary decisions

to disengage from commerce rather than to government-imposed restrictions

on activity, so re‡ecting people’s concerns that commerce may expose them

to the virus. Eichenbaum et al. (2021) highlight how the pandemic had

both a supply and a demand e¤ect since people cut back on consumption

and work in order to reduce the chances of being infected. These e¤ects in-

teracted with each other, so exacerbating the size of the recession caused by

the epidemic. These …ndings are consistent with the results of Correia et al.

(2020), who studied the economic e¤ects of the largest in‡uenza pandemic

in U.S. history, the 1918 Flu Pandemic. They …nd that, in a pandemic,

economic activity slows down in the presence of non-pharmaceutical inter-

ventions such as social distancing, since households reduce consumption and

labor supply in order to decrease their risk of becoming infected. Fear of

contagion leads households to be prudent and keeps people away from ac-

tivities that expose them to a higher risk of being infected, so becoming the

main reducer of economic activity.

   Immordino et al. (2021) build on this literature by providing new evi-

dence for Italy, based on data from an ad hoc designed household Survey on

Covid-19 and Consumption (SCC) and questions about perceived risk of con-


                                     10
tracting Covid-19 in various hypothetical situations. Relying on information

on consumer concerns about the risk of contagion during in-person shopping

and services requiring in-person contacts, they …nd a positive correlation be-

tween drops in consumption and the fear of contagion. These conclusions

are partially shared by Guglielminetti and Rondinelli (2021) who, by using

data from the Special Survey of Italian Households (SSIH), …nd a strong

impact of the fear of contagion on consumption choices, independently of

government restrictions. However, these papers provide a descriptive analy-

sis of the phenomenon and do not address causality. Recently, Immordino et

al. (2023) look at the determinants of expected consumption in the period

right after the Covid-19 pandemic. The authors …nd that expected house-

hold income and expected GDP growth are strongly related to consumption

expectations while health-related variables were not a major drivers of con-

sumption expectations in 2022.

   We add to this literature by presenting a solid identi…cation strategy

to estimate the direct e¤ect that fear of contagion has on expected con-

sumption. This is achieved by implementing a FRDD which relies on an

exogenous threshold in the fear of contagion generated by the way in which

o¢cial Covid-19 statistics on CFR were transmitted daily by o¢cial media

to Italian households. As we will show in the following sections, despite the


                                     11
fact that individuals around the age of sixty were exposed to the same CFR,

the fear of contagion shows a sharp rise among people immediately above

the age of sixty. We will use this arguably exogenous variation in fear to

estimate its impact on consumption expectations.


2.2      Household surveys and probabilistic questions

Using the taxonomy of expectations data derived from Manski’s (2004) sem-

inal article on the measurement of expectations, the present study uses data

on expectations regarding future choices, i.e., data that include both state-

ments of what choices agents expect to make in the future and choices that

agents would make if faced with the circumstances that are speci…ed by the

survey instrument.3

       The SSIH provides data on unconditional choice expectations derived by

asking individuals about their expectations of making certain decisions (i.e.

consumption) in the future without o¤ering any details of the environment

the individuals will be facing when this actual decision is made (other than

the time period in which the decision will be made).4 According to the

literature, data on expected future choices can be used as a substitute for

   3
    See also Manski (2018).
   4
    For example, the survey asks: “Consider expenditure on food, clothing and footwear,
home goods and services. How will your household change total expenditure on these
items in the next 3 months?”.



                                          12
data on actual choices when unavailable.5 Early evidence of this approach

is given by Van der Klaauw (2012), who illustrates the value of self-reported

choice expectations by using information on respondents’ expected future

occupations in the estimation of a structurally dynamic model, showing

that these estimates are very close to those obtained using realized choice

data. In addition, data on choice expectations are often particularly valu-

able for identifying certain parameters that are di¢cult to observe without

expectations data (Arcidiacono et al. [2020]; Adda et al. [2022]). Finally,

unconditional choice expectations about the probabilities of working at older

ages and leaving a bequest have been used by of Van der Klaauw and Wolpin

(2008).

    In addition to unconditional choice expectations, the SSHI contains data

derived from conditional choice expectations. In particular, the survey asks

heads of households what they expect to do in hypothetical scenarios. This

leads to repeated observations for the same individual that allow the ex-

ploitation of within-person variations, which are often not available with

data on observed choices.6 Following this approach, Giustinelli and Shapiro

   5
     See Kosar and O’Dea (2023) for a detailed discussion on how expectations data can
be used in the estimation of structural microeconomic models.
   6
     As an example, the survey asks: “Compared to ordinary times, how frequently would
you do the following activities [... ] if the daily Covid-19 cases in your region were: fewer
than 10, between 10 and 100; between 100 and 1000; more than 1000”.




                                             13
(2019) measure person-speci…c e¤ects of health on labor supply by using

data from a survey which asks individuals to consider, on the basis of their

health status, the probability that they will still be in work after the age of

62. Conditional choice expectation questions are closely related to so-called

“Strategic Survey Questions”. These are questions designed to identify the

parameters of speci…c models. For instance, Ameriks et al. (2008) develop

two strategic surveys to study the reasons (bequest motive or public care

aversion) why annuities are rarely purchased, despite the fact that they can

provide valuable longevity insurance. In a similar manner, in 2020, they

combined observed wealth data and strategic survey to jointly estimate risk

aversion, aversion to publicly provided long term care and the strength of the

bequest motive in order to analyze why many households retain wealth late

into life. Armantier et al. (2022) design three strategic surveys to identify

the extent to which long-term in‡ation expectations are sensitive to shocks

to past or future in‡ation.



3    Data

The present study is based on data from the Special Survey on Italian House-

holds (SSIH), which has conducted by the Bank of Italy every three months

since April/May 2020. We use data from the second edition of the SSIH,


                                      14
which took place between August and September 2020 and covered about

2,300 individuals.7 Apart from basic socio-demographic variables (age, gen-

der, education and occupational status), the survey includes questions on

economic and labor market prospects, …nancial situations and savings, which

are important variables to control for when explaining consumption expecta-

tions.8 Tables 1 and 2 provide some descriptive statistics and the de…nition

of variables used in this study. We use sample weights to make the statistics

population-representative.


                                Table 1 about here

                                Table 2 about here


       As far as expected consumption is concerned, the survey reports answers

to the question: “Consider expenditure on food, clothing and footwear, home

goods and services. How will your household change total expenditure on

these items in the next 3 months?”, so recording whether consumption was

   7
     We use this speci…c wave for two reasons. The …rst one is that this survey was
implemented just two months after the …rst National lockdown (March-June 2020). This
means that individuals still had memories of the meaning of a given contagion rate, which
in our case is a very important aspect in …xing a credible measure of the fear of contagion.
The second reason is that the long lockdown of previous months made it likely that these
individuals had been actually exposed to information transmitted on National TV in the
shape of the Covid-19 bulletins conveying information about CFR by age-groups.
   8
     The main results of the survey and the methodological aspects are described in the
paper: "The main results of the second wave of the special survey of Italian households
in 2020" (Covid-19 Notes, November 2020) and in the box ’Italian households during the
epidemic: the Bank of Italy’s survey’, in Economic Bulletin, 4, 2020.



                                            15
expected to decrease, stay the same or increase (categorized as 0, 1 and

2 respectively). Answers to this question de…ne our dependent variable

formally de…ned as Exp_Cons_Change.

   As regards the construction of the fear of contagion indicator, we ex-

ploit a conditional expectations question asking the heads of households:

“Compared to ordinary times, how frequently would you do the following

activities if the daily Covid-19 cases in your region were: fewer than 10, be-

tween 10 and 100; between 100 and 1000; more than 1000”, and then a list

of 4 indicative activities followed. For each of these, answers are categorized

according to the following levels: I would stop doing it/rarely; less often;

with the same frequency; a little more often; I would do it more frequently.

These activities are: going out to buy food and essential goods; make pur-

chases in clothing stores; make visits to hotels/ restaurants/bars; go to the

salon/barber/cosmetician. Therefore, for each of these activities, we can

construct an indicator of fear of infection according to how it would vary as

a consequence of changes in the number of contagions per day in the region

of residence. We can see how this works by considering, as an example, the

activity of shopping for food and essential goods. In practice, fear of conta-

gion is represented by an ordered variable that increases together with the

level of fear and assumes a value from 0 to 4. The variable assumes a value


                                      16
of 0 (no fear) if the respondent would continue buying food and essential

goods with the same frequency in each assumed epidemiological scenario; 1

if respondent declares they would only eliminate or drastically reduce/avoid

this activity if the daily Covid-19 cases exceeded 1000; 2 if the respondent

declares they would eliminate or drastically reduce/avoid the activity if the

daily Covid-19 cases were more than 100; 3 if they declare they would elim-

inate or drastically reduce/avoid the activity if the daily Covid-19 cases

numbered more than 10; and 4 (maximum fear) if the respondent declares

they would eliminate or drastically reduce/avoid the activity if the Covid-19

daily cases were fewer than 10.

   All in all, we have 4 indicators of fear for each individual. However, we

note that since going out for beauty and personal care services is not that

common for men in their sixties, and since these men make up a large share

of our sample of heads of households (57% of the sample is constituted by

men of over 54 years of age), it is plausible that we do not actually capture

the fear of being infected by looking at the fear indicator that is built on

observation of this activity. In practice, since the activity was only rarely

performed in ordinary times, there is no room for its reduction, regardless of

the epidemiological scenario. For this reason, we discard this fear indicator

from our further analyses. In relation to this argument, we believe that


                                     17
the main activity that would e¤ectively capture fear of contagion is that of

shopping for food and essential goods, since this is likely to be performed

by all the households in our sample and so recording its variation as a

consequence of the severity of the epidemiological scenario is likely to capture

the fear of contagion. Therefore, in our analysis, we will consider this as

our main fear indicator, henceforth simply indicated as F ear1. Measures

of fear derived from changes in behavior relating to shopping for clothes

and furniture (F ear2) and to visits to bar/restaurants/hotels (F ear3) will

be used to assess the robustness of our results. We note that correlations

between these F ear2 and F ear3 variables is 0.91 while their correlation with

F ear1 is 0.70 and 0.73 respectively.



4    Endogeneity and Identi…cation Strategy

Let us consider the following linear model where the expected consumption

change (Exp_Cons_Change) of individual i is regressed against a set of

explanatory variables:




                                        18
Exp_Cons_Changei =            + Xi + Liquidity_Constraintsi +                   (1)

                            + Exp_M acro_Scenarioi + Exp_M icro_Scenarioi

                            + F ear1i + ui



   In eq. (1), the dependent variable takes value 0 for households expect-

ing a decrease in their total spending in the next three months, 1 if they

expect their total spending will remain the same and 2 if they expect it to

increase. X is a vector of respondent characteristics (age, gender, marital

status, family size, occupational status, education and geographical area of

residence), while Liquidity is a set of dummy variables that approximates

liquidity constraints based on the subjective evaluation households make of

their general economic condition in terms of their ability to make ends meet.

Following the literature on how expectations about macroeconomic variables

a¤ect household consumption plans (Coibion et al., 2020), we control for

personal macroeconomic expectation, i.e. their personal expectation of how

the general economic situation in Italy and the labor market will develop

(Exp_M acro_Scenario). We also include expectations about employment

status in the near future (Exp_M icro_Scenario), which is a variable that



                                     19
takes the value 1 if the self-reported probability of losing their job in the

next 12 months is higher than 50% for one household member and assumes

a value of 2 if the probability of losing their job is higher than 50% for

more than one household member. Finally, we introduce our key variable

F ear1, which accounts for the fear of contagion. As illustrated in the pre-

vious section, this variable is derived from the conditional expectations of

how activities relating to the provision of essential goods requiring social

interaction would vary according to the number of Covid-19 cases per day

in the region of residence. In eq. (1), indicators F ear2 and F ear3 will be

used instead of F ear1 for robustness.

   In order to assess to what extent the fear of contagion is relevant in

determining household spending decisions in such a model, we need to take

into account endogeneity and reverse causality problems brought about by

the simultaneous relationship between these two variables. The main issue

is the fact that even if they are particularly scared of the virus, households

with low levels of consumption have no room for further reduction in the

activities they perform to acquire essential goods so parameter     in eq. (1)

is likely to be biased toward zero. Furthermore, households which, for any

unobserved reason, have low levels of consumption might also think they are

at a low risk of infection because of their limited social interaction. Since


                                     20
these unobserved factors simultaneously shape consumption expectations

and fear, another source of bias arises.

   In order to tackle these endogeneity issues, we implement a Fuzzy Re-

gression Discontinuity Design, using an (arguably) exogenous variation in

the fear of contagion which regards belonging to a speci…c age group. As

is well known, the general idea underlying the regression discontinuity ap-

proach is that it is possible to assign all the units that have received a speci…c

treatment, the “treated”, to one side of a pre…xed threshold and all the other

units, those that have not received the treatment, the “control group”, to

the other side. The advantage of this analysis is that it makes it possible to

compare units with very similar values of a running variable, but di¤erent

levels of treatment in order to draw causal inference on the e¤ect of treat-

ment at the threshold. In the fuzzy design, the realized value of the forcing

variable does not result in the receiving of the treatment in a deterministic

way. This methodology is based on the assumption that a given value of the

forcing variable, falling above or below a exogenously determined threshold,

generates a discontinuity in the outcome of interest.

   The design implemented in this study exploits variation in the perception

of health risks in di¤erent age groups. It has been well documented that

age appears to be a crucial factor in Covid-19 health outcomes. Throughout


                                       21
Europe, fatalities caused by Covid-19 have been mostly concentrated in older

age groups. During the pandemic, National media outlets provided daily

reports and updates concerning the coronavirus situation, reporting on the

number of infected, quarantined, hospitalized and deaths, and emphasizing

that advanced age was a risk factor for Covid-19 mortality. One of the most

frequently referred to statistics was the CFR, which is a recognized measure

of severity among detected cases. We argue that systematic exposure to

news on the CFR during the Covid-19 pandemic may have produced an

intensi…cation of fear of contagion and anxiety on the basis of belonging

to a speci…c age group, transforming individuals’ behavior and increasing

preventive conduct.

   In Italy o¢cial statistics reported that the elderly population ran sig-

ni…cant higher health risks than other groups and daily o¢cial statements

informed the population of this risk. Remarkably, all mass media commu-

nication focused on stacked age groups separating individuals into ten-year

categories, i.e., 0-9, 10-19 and so on until the last group (all people older than

90). Our main point is that, in the data, while the CFR is always around

zero percent for all groups up until 50-59, a sharp increase occurs when con-

sidering the 60-69 age group, for which the CFR is suddenly around 10% on

average. In Figure 2, we report a scatter-plot of daily data on CFR by age


                                       22
group provided by the Italian Ministry of Health’s daily communications for

the period 20th March-26th June 2020. This information was presented on

national TV by the National Covid-19 Bulletin at 6 p.m. every day, as well

as by all the major newspapers.


                           Figure 2 about here


   According to this …gure, a fairly evident jump in terms of CFR occurs

around the age of 60. Although another jump is fairly evident at the age of

70, we do not exploit this speci…c discontinuity for two reasons. The …rst

one is that the impact of variation of CFR on fear of contagion is likely to

be relevant when fatality in the case of contagion jumps from almost 0% to

a positive value. Instead, when CFR jumps from 10% to 20%, fear is likely

to be less a¤ected since the CFR starting point is already high enough to

induce a notable fear of contagion, so that there is no room for an additional

increase of it. The second one is that, by focusing on the 70 age we would lose

several observations to derive meaningful results. Consequently we look at

the discontinuity in CFR arising around the age of 60. Our main (testable)

point is that, since individuals had been highly exposed to this information,

especially during the very long and severe lockdown that started in Italy

on March 9th, 2020 and ended June 1st, 2020, it had probably already

in‡uenced their fear of contagion. In this case, albeit individuals who were

                                      23
just above or below the age of 60 would have faced substantially the same

CFR if they had caught Covid-19 since they are almost identical in terms

of age, a discontinuity in the fear of contagion around this threshold should

be observed in the data. In other words, individuals whose ages di¤ered by

just a few months may have had di¤erent perceptions of their risk from the

coronavirus simply because they were classi…ed in di¤erent, a priori de…ned,

age groups. Thus, in order to study the causal e¤ect of fear of Covid-19

on consumption expectations, we exploit this quasi-experimental setup by

using a fuzzy regression discontinuity design and relying on the fact that a

higher level of perceived fear is de…ned as a discontinuous function of the

respondent’s age.

   At this stage, an important aspect that deserves attention is how we …x

the age-threshold in our empirical model. Indeed, although the chronologi-

cal age separating the two groups is 60, the respondent’s age in the survey

is derived from reference to year of birth. This means that some individuals

who had not reached 60 when exposed to the mass media communications

are reported as people of 60. Therefore, in order to identify those who had

already turned 60 in the March-June 2020 period, we should consider those

whose age reported in the survey is 61. Furthermore, we should consider

the self-identi…cation of individuals who, in the period of interest, were ex-


                                     24
actly 60 years old. Indeed, several studies have reported how humans in

later adulthood often avoid the negative consequences of their age-group

membership by psychologically distancing themselves from it and this is

particularly true for those who are located towards the lower limit of a spe-

ci…c age-group (among others see Weiss and Lang, 2009). In order to reduce

biases induced by the self-distancing problem, we need to consider as treated

those who, in our period of interest, were 61. In the light of these insights,

we …x the age-threshold at 62 in order to single out those who recognized

themselves as members of the 60-69 age-group at the time of their exposure

to the communications on CFR and therefore perceived the risk of a higher

CFR.

   That being said, it is important to clarify that we employ a fuzzy dis-

continuity approach since the respondent’s age will not perfectly predict

fear of Covid-19. Indeed, some respondents will not be afraid of contracting

the virus even if they are older (these individuals are labelled as “never-

takers”) while others might have higher levels of fear of contracting the

virus even though they are younger (known as “always-takers”). A fuzzy

discontinuity design explicitly allows for the presence of “never-takers” and

“always-takers” by utilizing the threshold to predict a jump in the fear of

contracting the virus at the cut-o¤ (Imbens and Lemieux, 2008). We can


                                     25
use such discontinuities to produce instrumental variable estimators of the

e¤ect of the treatment on expected consumption.

    Finally, it is important to stress that an additional advantage of our

identi…cation strategy is that it allows us to get rid of problems arising from

the presence of government restrictions, such as lockdowns, which constrain

individual behavior. Indeed, even if we use expectations on consumption, it

may be that an individuals’ forecast takes into account expected lockdowns.

However, as long as we compare similar units at the threshold, this e¤ect

should be irrelevant for our inference.



5     Results

5.1   Naïve Methods: Linear Regression and Ordered Probit

We start our empirical investigation by estimating eq. (1) through the use

of a linear regression model (Table 3) and an Ordered Probit model (Table

4) in order to provide some evidence on the relationship between fear of

contagion and consumption expectations.


                            Table 3 about here

                            Table 4 about here




                                      26
    In each column of these tables, we include a di¤erent indicator of fear.

These preliminary estimates show a very strong correlation (always signif-

icant at 1%) between the spending expectations of Italian households and

the fear of contagion, regardless of the indicator used. Coe¢cients go in the

expected direction and show a negative impact of fear on the probability

that consumption will remain the same or increase in the next future. The

estimated coe¢cient is always around -0.050.9 This evidence is consistent

with that reported by Immordino et al. (2021). Looking at other regres-

sors, the expected expenditure has a U-shaped pro…le over an individual’s

life cycle and the probability of a drop in consumption is higher for younger

people. As expected, the self-employed and unemployed are more likely to

expect a fall in consumption in the next three months. In addition, being

subject to liquidity constraints - captured by the make-ends-meet variable -

increases the possibility of …nancially constrained households cutting future

expenditure more than wealthy households will. Consumption expectations

decrease for those foreseeing a worsening in their personal labor market

perspectives.

   9
     In Tables 3 and 4, as well as in all the other speci…cations considered in this paper,
we consider a 0-2 dependent variable for expected consumption as de…ned in Table 2. We
show that all results hold if i) we consider a 0-1 dependent variable where we collapse
answers stating that consumption will remain the same or will increase in the future; ii)
we drop observations declaring that consumption will increase in the near future.




                                            27
    As widely discussed in the previous section, we should be aware that

the fear of contagion in eq. (1) is potentially endogenous due to the simul-

taneous determination of fear of contagion and consumption expenditure

expectations. Households with a high level of fear may be among those

which already consume less so that they have no room for additional reduc-

tion in consumption. Therefore, not considering endogeneity issues would

likely lead to biased estimates of the underlying causal relationship.



6    Fuzzy Regression Discontinuity Design

We deal with endogeneity by using an arguably exogenous variation of the

independent variable of interest. Our FRDD is targeted to estimate the

e¤ect of fear of contagion on expected consumption when using the age of 62

as a forcing variable for both expected consumption and fear. Indeed, since

of the information about CFR transmitted to people during the pandemic

underlined that, in the case of contagion among people in the 50-59 age-

group, 3 out of 100 of the sick died because of adverse consequences relating

to the virus while the CFR jumped to 10 over 100 for those in the 60-69

group, the causal e¤ect of fear of contagion on consumption can be identi…ed

by instrumenting the former with the indicator of whether an individual’s

age (at the time of the survey) is 62. Consequently, the e¤ect on consumption


                                     28
expectations of a sharp increase in fear is given by the jump in consumption

expectations at the age threshold (the “Reduced Form” or “Intention-to-

Treat” e¤ect) divided by the jump in fear at the threshold (the “First-Stage”

e¤ect).


6.1   First stage

Panel (a) of Table 5 reports our …rst stage estimates. In column (1), the

outcome variable is our main fear indicator (F ear1) while column (2) and (3)

refer to F ear2 and F ear3 respectively. The cut-o¤ is set at the age of 62 and

a nonparametric method is implemented. This approach tries to produce

reliable estimates by focusing on small samples to the right and to the left of

the cut-o¤ point (bandwidths) through a Local Polynomial Regression that

represents a sort of Weighted Least Squares estimation where the weights

are larger the closer the observations are to the cut-o¤.


                            Table 5 about here


   In Panel (b) of the same table, we report the estimates obtained after

including individual characteristics X, as discussed in eq. (1). We use a

Triangular Kernel, a linear polynomial and the MSE data driven bandwidth

selection (mean squared error optimal bandwidth selector). The results of

the …rst stage actually point toward the presence of a signi…cant jump in the

                                      29
fear of contagion at the threshold of 62. This is true at a 1% signi…cance level

for our main indicator (F ear1) and at a 5% level for the other two. What

is more, the magnitude of the e¤ect is stable across the three models and is

not a¤ected by the inclusion of explanatory variables. This means that this

age threshold can be used as an instrument to predict the F ear1 variable

in order to obtain, in the second stage, unbiased estimates of the e¤ect of

fear on expected consumption modi…cations since, from this analysis, we can

conclude that instrumental-variable (IV) estimates do not su¤er from weak

instrument issues.


                           Figure 3 about here


   A graphical analysis of this …rst-stage is presented in Figure 3 and our

assumption concerning the impact of CFR communication and fear of conta-

gion also appears evident graphically. In all the panels in Figure 3, dots are

the sample average of fear within each bin against the respondent age (each

panel refers to a speci…c fear indicator). The cut-o¤ point is represented

by the vertical line at the respondent age of 62. Clearly, individuals on the

RHS of the cut-o¤ point record a signi…cantly higher level of fear. This is

particularly evident with our main fear indicator (F ear1).




                                      30
6.2   Main outcome

Table 6 presents the intention-to-treat e¤ect, i.e., the e¤ect on expected

consumption variations of being on one side or the other of the 62 y.o.a.

threshold. In column (1), we report the estimated e¤ect without covariates

while in column (2) we consider all the covariates included in eq. (1). The

e¤ect is always statistically signi…cant at a 5% level, meaning that there is

a reduction in expected consumption for those individuals older than 62.

Figure 4 provides clear evidence of a discontinuous change in consumption

expectations at the threshold.


                           Table 6 about here

                          Figure 4 about here


   Table 7 presents the corresponding instrumental-variable (IV) second-

stage estimates as given by the reduced-form or intention-to-treat (ITT)

estimate scaled by the …rst stage e¤ect. With a …rst stage e¤ect of about

0.65 in the case of F ear1, the IV estimates are about 54 percent larger than

the ITT results. In this case, the magnitude of the e¤ect is -0.381 without

covariates and -0.352 when controls are included. These numbers are about 9

times larger than those estimated with the simple regression models reported

in Table 3. This shows that the endogeneity issue is severe when evaluating


                                     31
how relevant the fear of contagion is to consumption expectations: fear of

being infected substantially reduces expectations of future spending.


                           Table 7 about here



7     Check the Validity of FRDD

7.1   No Manipulation and Continuity

In this Section, we check the validity of the FRDD. There are two key iden-

ti…cation assumptions in the context of FRDD: there is no “manipulation”

of the forcing variable and so the cut-o¤ provides exogenous variations in

the treatment; and “continuity”, meaning respondents just above and below

the cut-o¤ point should be similar so observable characteristics do not vary

discontinuously at the cut-o¤. In our case, manipulation should be under-

stood as a change in the density of the forcing variable around the threshold.

To verify that this is not the case, following to the McCrary (2008) proce-

dure in Figure 5, we plot a histogram of the density of the forcing variable

(respondent age) around the cut-o¤ and we test for the existence of any

density-discontinuity.


                           Figure 5 about here




                                     32
   The graph does not show any evidence of jumps. The formal McCrary’s

test, using local polynomial density estimation, detects a coe¢cient of 0.478,

which, with a p-value=0.633, is far from being signi…cant.

   As far as the continuity assumption is concerned, we examine the con-

tinuity of observed covariates to determine whether they are continuous at

the threshold of the respondent age of 62 years. Figure 6 summarizes all the

graphs and shows that all the observable characteristics are continuous at

the threshold, assuring the similarity of the two groups around the cut-o¤

and con…rming that the e¤ect estimated in our main analysis is not due to

any spurious correlation between age and other factors.


                           Figure 6 about here


7.2   FRDD with discrete running variable

In what follows, we take into account the fact that our running variable

(age in years) is discrete and that this poses some problems in regression

discontinuity design. We decided to address this concern by implement-

ing honest, e¢cient con…dence intervals in regression discontinuity designs

by using procedures from Armstrong and Kolesár (2020), Armstrong and

Kolesár (2018) and Kolesár and Rothe (2018). These authors point out that

the discreteness of the running variable may cause problems if the number of


                                     33
support points close to the cut-o¤ is so small that using a smaller bandwidth

to make the bias of the estimator negligible relative to its standard devia-

tion is not feasible. We implement the approach of bounding the second

derivative (BSD) by using the Armstrong and Kolesár (2020) rule of thumb

for a bound on the second derivative of the conditional mean function. Our

nonparametric estimates, using BSD inference for the di¤erent indicators

of fear of contagion as an outcome variable, are reported in Table 8. The

results are similar to those obtained in the previous section as far as the

F ear1 indicator is concerned. In this case, a discontinuity signi…cance of a

5% level at the age of 62 is reported. The other two indicators do not show

any discontinuity at a 5% level and only F ear2 turns out to be signi…cant

at a 10% level. These results do not undermine our identi…cation strategy

since, as discussed in Section 4, we actually expect to capture the fear of

contagion through variable F ear1, which is derived from answers relating

to conditional expected behavior when essential goods are considered. Fur-

thermore, as reported in Table 8, expected consumption changes show a

signi…cant discontinuity at the threshold even when honest standard errors

are considered and this crucially supports our identi…cation strategy.


                           Table 8 about here




                                     34
7.3   6.3 Relevance of the Forcing Variable

Our identi…cation strategy rests on the fact that the age threshold is ex-

ogenously determined by the perception of infection risk that is induced by

o¢cial statements providing information on the rate of mortality for speci…c

age groups. In order to further support this assumption, we use a di¤erent

data set which contains information for other countries which had similar

infection rates, but di¤erent communication strategies. During the Covid-

19 pandemic, a special survey questionnaire, Survey of Health, Aging and

Retirement in Europe (SHARE), was designed and administered in all 28

European Union countries to examine the impact of the pandemic on the

health and living situation of the population that was older than 50. By

using data for France from the second wave of this survey, we can provide

evidence on the e¤ect of Covid-19 CFR information strategies on a di¤erent

proxy of fear of contagion, the decision to have a vaccination or not.

   France adopted similar communication strategies to Italy with regard the

severity of the disease. In particular, the French health authorities dissem-

inated weekly and/or daily National epidemiological updates in which the

most important indicators (con…rmed cases, new hospitalizations, deaths)

for monitoring the Covid-19 outbreak were presented as age-group speci…c

data. We consider the second wave of the survey, June-August 2021, which


                                     35
contains information about vaccination decisions and so it enables us to

construct a proxy for fear of contagion. We exclude Italy from this valida-

tion exercise since was already compulsory there at the time of the survey

vaccination.10 We take advantage of the di¤erent age strati…cation used

to disseminate information on the severity of the disease by the National

Health Authority, “Santé Publique France”, in France.11 We implement a

Regression Discontinuity Design approach (RDD) to verify whether any dis-

continuity in the willingness to get the vaccine against Covid-19 arises at the

cut-o¤ according to the age strati…cation used in national communication of

the CFR. In the survey questionnaire, each respondent said whether he/she

has been vaccinated against Covid-19 and those who answered “No” received

another question on whether they have already scheduled an appointment,

wanted to get vaccinated, did not want to get vaccinated or were still unde-

cided. Starting from these questions, we make our outcome variable equal

to 1 if a respondent replied that she/he was vaccinated or she/he wanted to

get vaccinated (with or without a scheduled appointment) and 0 otherwise.

In total, we have about 1,700 respondents for France.

  10
     In fact, on 1st April 2021, Italy became the …rst country in Europe to make vaccination
against Covid-19 compulsory, the government having approved Decree Law No. 44 with
the intent to stop a third wave of the disease.
  11
     All information on age strata were extracted from the French National Health Au-
thority websites, in particular: Santé Publique France. Covid-19: point épidémiologique:
  https://www.santepubliquefrance.fr/recherche/#search=COVID%2019%20%20%20
  point%20epidemiologique&publications=donn%C3%A9es&regions=National&sort=date


                                           36
                            Table 9 about here


   Table 9 provides some descriptive statistics on the respondents. We use

sample weights to make the statistics population-representative. We apply

the RDD approach, setting a cut-o¤ according to the age clusters used in

France, where grouping in epidemiological updates started at the age of 15

using the following clusters: 15-44, 45-64, 65-74, 75+. Therefore, we set

the cut-o¤ at the age of 66 to be sure to include people who self-identi…ed

themselves with the 65-74 age group.


                           Figure 7 about here


   A graphical analysis is presented in Figure 7 showing the sample average

within each bin of the willingness to get the vaccine against respondent age.

The continuous line predicts the values of a 4-order polynomial, estimated

separately on each side of the cut-o¤ point, which is represented by the

vertical line at the age of 66. A clear jump arises in the relationship between

the outcomes (willingness to vaccinate against Covid-19) and the respondent

age variable in the proximity of the threshold.


                           Table 10 about here


   Table 10 reports regression discontinuity estimates by using local poly-

nomial regression with and without covariates, which, in this case, include

                                      37
gender, family size, the presence of any chronic disease, self-reported health

status and a dummy variable indicating whether the individual knew some-

one who has been infected by the virus. Results con…rm the sensitivity of

individual behavior to the communication strategy used for o¢cial state-

ments. There is a discontinuity in the probability of being vaccinated at the

age of 66 that is signi…cant at a 10% level when the bandwidth is set at 3.8

(MSE optimal bandwidth) and at a 5% level if the bandwidth is set at 5.

This evidence may be considered to support our identi…cation strategy.



8    Conclusions

This study uses novel micro-data from the Bank of Italy’s Special Survey of

Italian Households (SSIH) to examine the causal e¤ect of fear of Covid-19 on

consumption expectations during the pandemic. This analysis is extremely

relevant in order to set sound policy measures to recover after a recession

caused by infectious disease. Our …ndings show that the fear of contracting

a virus induces a change in consumer behavior and constrains individual

choices. Until now, the existing literature has just provided a descriptive

analysis of the interrelationships between these variables. The novelty of our

analysis lies in its providing neat evidence of the causal impact of the fear

of contagion on expected consumption through its use of an identi…cation


                                     38
strategy that overcomes endogeneity problems and drawbacks relating to

the presence of government restrictions. Moreover, the use of expectations

as opposed to actual spending makes it possible to obtain an evaluation of

unconstrained potential behavior, so managing problems tied to the fact that

observed choices may be consistent with many combinations of expectations

and preferences, especially in the presence of legally imposed lockdowns. In

the more recent literature, these considerations have generated an increasing

awareness of the potential value of expectations data for policy.

   Fear of contagion may depend on unobservable characteristics that lead

to endogeneity problems. Thus, in order to address these issues, an ex-

ogenous variation in the fear of contagion is used in order to identify the

coe¢cients of interest. We exploit the discontinuity in the fear of conta-

gion observed for people above a speci…c age threshold to implement a fuzzy

RDD. The age threshold is exogenously determined by daily o¢cial state-

ments that informed the Italian population that the CFR for those above

the age of 60 was nearly four times higher than that of those between 50

and 59. Our results, which are robust to a variety of estimation techniques

implemented to address speci…c caveats, provide evidence that the fear of

contagion has a relevant impact in determining consumption expectations

and is almost 10 times larger than that estimated in a simple regression


                                     39
model.

   Understanding to what extent consumption is constrained by health con-

cerns is extremely useful in identifying strategies which can reverse the

economic downturn. If consumer spending is constrained by health con-

cerns, governments should restore consumer con…dence by addressing health

care, while traditional macroeconomic tools that stimulate aggregate de-

mand might have a relatively reduced capacity to boost consumption and

promote economic recovery. Taken together, our …ndings suggest that the

evolution of expenditure is a¤ected by health concerns due to the infectious

nature of a pandemic and, hence, may not be …xable through conventional

macroeconomic tools. In this view, our estimates highlight that the Covid-

19 pandemic created new challenges for public policy, which should …nd

functional strategies to restore economic activity by considering that the

infectious nature of such events a¤ects economies through its psychological

impact on the population. During a pandemic, policy makers should atten-

tively consider strengthening the National Health Care System to provide

e¤ective and timely health services to all as instruments aimed at restoring

consumer con…dence and reducing the fear of contagion.

   Finally, this research highlights the relevance that o¢cial strategy for

communication of the consequences of adverse events may have in determin-


                                    40
ing public perception of risk and uncertainty, shaping individual behavior

and economic outcomes. This topic has though so far received very little

attention from researches.


   Con‡ict of interest statement: The authors declare they have no

con‡ict of interests.



References

  1. Adda, J., Dustmann, C., & Görlach, J. S. (2022). The dynamics of

      return migration, human capital accumulation, and wage assimilation.

      The Review of Economic Studies, 89(6), 2841-2871.


  2. Alexander, D., & Karger, E. (2021). Do stay-at-home orders cause

      people to stay at home? E¤ects of stay-at-home orders on consumer

      behavior. " Working Paper Series WP-2020-12, Federal Reserve Bank

      of Chicago.


  3. Ameriks, J., Caplin, A., Laufer, S., & Van Nieuwerburgh, S. (2008).

      The Joy of Giving or Assisted Living? Using Strategic Surveys To

      Separate Bequest and Precautionary Motives. Journal of Finance 66

      (13105), 519–561.


  4. Ameriks, J., Briggs, J., Caplin, A., Matthew, D, Shapiro, D., Tonetti,

                                    41
  C. (2020) “Long-Term-Care Utility and Late-in-Life Saving”. Journal

  of Political Economy. University of Chicago Press, vol. 128(6), 2375-

  2451.


5. Andersen, A. L., Hansen, E. T., Johannesen, N., & Sheridan, A.

  (2020a). Consumer responses to the COVID-19 crisis: Evidence from

  bank account transaction data. Available at SSRN 3609814.


6. Andersen, A. L., Hansen, E. T., Johannesen, N., & Sheridan, A.

  (2020b). Pandemic, shutdown and consumer spending: Lessons from

  Scandinavian policy responses to COVID-19. arXiv preprint arXiv:2005.04630.


7. Angrist, J. D., & Pischke, J. S. (2009). Mostly harmless econometrics.

  Princeton University press.


8. Arcidiacono, P., Hotz, V. J., Maurel, A., & Romano, T. (2020). Ex

  ante returns and occupational choice. Journal of Political Economy,

  128(12), 4475-4522.


9. Armantier, O., Sbordone, A., Topa, G., Van der Klaauw, W., &

  Williams, J. C. (2022). A new approach to assess in‡ation expec-

  tations anchoring using strategic surveys. Journal of Monetary Eco-

  nomics, 129, S82-S101.



                                 42
10. Armstrong, T.B., & Kolesár, M. (2018). Optimal inference in a class

   of regression models. Econometrica, 86(2), 655-683.


11. Armstrong, T. B., & Kolesár, M. (2020). Simple and honest con…dence

   intervals in nonparametric regression. Quantitative Economics, 11(1),

   1-39.


12. Baker, S. R., Farrokhnia, R. A., Meyer, S., Pagel, M., & Yannelis,

   C. (2020). How does household spending respond to an epidemic?

   Consumption during the 2020 COVID-19 pandemic. The Review of

   Asset Pricing Studies, 10(4), 834-862.


13. Armantier, O., Sbordone, A., Topa, G., Van der Klaauw, W. & Williams,

   J.C. (2022). A new approach to assess in‡ation expectations anchoring

   using strategic survey. Journal of Monetary Economics, 129, S82-S101.


14. Baker, S. R., Farrokhnia, R. A., Meyer, S., Pagel, M., & Yannelis,

   C. (2020). Income, liquidity, and the consumption response to the

   2020 economic stimulus payments (No. w27097). National Bureau of

   Economic Research.


15. Börsch-Supan, A., M. Brandt, C. Hunkler, T. Kneip, J. Korbmacher,

   F. Malter, B. Schaan, S. Stuck, S. Zuber (2013).      Data Resource



                                  43
    Pro…le: The Survey of Health, Ageing and Retirement in Europe

    (SHARE). International Journal of Epidemiology. DOI: 10.1093/ije/dyt088.


16. Börsch-Supan, A. (2022). Survey of Health, Ageing and Retirement

    in Europe (SHARE) Wave 9. COVID-19 Survey 2. Release version:

    8.0.0. SHARE-ERIC. Data set. DOI: 10.6103/SHARE.w9ca.800


17. Calonico, S., Cattaneo, M. D., Farrell, M. H., & Titiunik, R. (2017).

    rdrobust: Software for regression-discontinuity designs. The Stata

    Journal, 17(2), 372-404.


18. Chen, H., Qian, W., & Wen, Q. (2021). The impact of the COVID-19

    pandemic on consumption: Learning from high-frequency transaction

    data. In AEA Papers and Proceedings (Vol. 111, pp. 307-11).


19. Chetty, R., Friedman, J. N., Hendren, N., Stepner, M., & The Oppor-

    tunity Insights Team. (2020). How did COVID-19 and stabilization

    policies a¤ect spending and employment? A new real-time economic

    tracker based on private sector data (pp. 1-109). Cambridge, MA:

    National Bureau of Economic Research.


20. Christelis, D., Georgarakos, D., Jappelli, T., & Kenny, G. (2020).

    The Covid-19 crisis and consumption: survey evidence from six EU

    countries.

                                  44
21. Chronopoulos, D. K., Lukas, M., & Wilson, J. O. (2020). Consumer

   spending responses to the COVID-19 pandemic: An assessment of

   Great Britain. Available at SSRN 3586723.


22. Cline, W. (2023). Fighting the Pandemic In‡ation Surge of 2021-2022.

   Economics International Inc. working paper series n. 23-1.


23. Coibion, O., Gorodnichenko, Y., & Weber, M. (2020). The cost of the

   covid-19 crisis: Lockdowns, macroeconomic expectations, and con-

   sumer spending (No. w27141). National Bureau of Economic Re-

   search.


24. Coibion, O., Georgarakos, D., Gorodnichenko, Y., Kenny, G., & We-

   ber, M. (2021). The e¤ect of macroeconomic uncertainty on household

   spending (No. w28625). National Bureau of Economic Research.


25. Correia, S., Luck, S., & Verner, E. (2020). Fight the pandemic, save

   the economy: Lessons from the 1918 ‡u. Federal Reserve Bank of New

   York, 5.


26. Cox, N., Ganong, P., Noel, P., Vavra, J., Wong, A., Farrell, D., &

   Deadman, E. (2020). Initial impacts of the pandemic on consumer

   behavior: Evidence from linked income, spending, and savings data.

   Brookings Papers on Economic Activity, 2020(2), 35-82.

                                  45
27. de Bruin, W., Chin, A., Dominitz, J. & van der Klaauw, W. (2023)

    Chapter 1 - Household surveys and probabilistic questions, Editor(s):

    Bachmann, R., Topa, G. van der Klaauw, W., Handbook of Economic

    Expectations, Academic Press, 3-31.


28. Delavande, A., van der Klaauw, W., Winter, J. & Zafar, B. (2022). In-

    troduction to the Journal of Econometrics Annals Issue on “Subjective

    Expectations and Probabilities in Economics”. Journal of Economet-

    rics, 231(1), 1-2.


29. Dominitz, J. (2001). Estimation of income expectations models using

    expectations and realization data. Journal of Econometrics, 102(2),

    165-195.


30. Dominitz, J., & Manski, C. F. (1997). Using expectations data to

    study subjective income expectations. Journal of the American Sta-

    tistical Association, 92(439), 855-867.


31. Eichenbaum, M. S., Rebelo, S., & Trabandt, M. (2021). The macro-

    economics of epidemics. The Review of Financial Studies, 34(11),

    5149-5187.


32. Giustinelli, P., & Shapiro, M. D. (2019). Seate: Subjective ex ante

    treatment e¤ect of health on retirement (No. w26087). National Bu-

                                   46
    reau of Economic Research.


33. Goolsbee, A., & Syverson, C. (2021). Fear, lockdown, and diversion:

    Comparing drivers of pandemic economic decline 2020. Journal of

    Public Economics, 193, 104311.


34. Grundy, E., & Read, S. (2012). Social contacts and receipt of help

    among older people in England: are there bene…ts of having more

    children?. Journals of Gerontology Series B: Psychological Sciences

    and Social Sciences, 67(6), 742-754.


35. Guglielminetti, E. & Rondinelli, C. (2021). Consumption and saving

    patterns in Italy during Covid-19.Bank of Italy Occasional Papers n.

    620.


36. Hahn, J., Todd, P., & Van der Klaauw, W. (2001). Identi…cation and

    estimation of treatment e¤ects with a regression-discontinuity design.

    Econometrica, 69(1), 201-209.


37. Hall, R. E. (1978). Stochastic implications of the life cycle-permanent

    income hypothesis: theory and evidence. Journal of Political Economy,

    86(6), 971-987.


38. Imbens, G. W., & Angrist, J. D. (1994). Identi…cation and estimation



                                    47
    of local average treatment e¤ects. Econometrica 62 467–476.


39. Immordino, G., Jappelli, T., Oliviero, T., & Zazzaro, A. (2021). Fear

    of COVID-19 Contagion and Consumption: Evidence from a Survey

    of Italian Households. Health Economics, 31(3), 496-507.


40. Immordino, G., Jappelli, T., & Oliviero, T. (2023). Consumption and

    income expectations during Covid-19. Review of Economics of the

    Household, 22(1), 95-116.


41. Kolesár, M., & Rothe, C. (2018). Inference in regression discontinuity

    designs with a discrete running variable. American Economic Review,

    108(8), 2277-2304.


42. Koşar, G., & O’Dea, C. (2023). Expectations data in structural mi-

    croeconomic models. In Handbook of Economic Expectations (pp.

    647-675). Academic Press.


43. Lee, D. S., & Lemieux, T. (2010). Regression discontinuity designs in

    economics. Journal of economic literature, 48(2), 281-355.


44. Lin, W., & Wooldridge, J. M. (2019). Testing and correcting for endo-

    geneity in nonlinear unobserved e¤ects models. In Panel Data Econo-

    metrics (pp. 21-43). Academic Press.



                                   48
45. Manski, C. F. (2004). Measuring expectations. Econometrica, 72(5),

   1329-1376.


46. Manski, C. F. (2018). Survey measurement of probabilistic macroeco-

   nomic expectations: progress and promise. NBER Macroeconomics

   Annual, 32(1), 411-471.


47. Nino, M., Harris, C., Drawve, G., & Fitzpatrick, K. M. (2021). Race

   and ethnicity, gender, and age on perceived threats and fear of COVID-

   19: Evidence from two national data sources. SSM-population health,

   13, 100717.


48. Scherpenzeel, A., K. Axt, M. Bergmann, S. Douhou, A. Oepen, G.

   Sand, K. Schuller, S. Stuck, M. Wagner, A. Börsch-Supan (2020). Col-

   lecting Survey Data among the 50+ Population during the COVID-19

   Outbreak: The Survey of Health, Ageing and Retirement in Europe

   (SHARE). Survey Research Methods 14 (2).


49. Van der Klaauw, W. (2012). On the use of expectations data in es-

   timating structural dynamic choice models. Journal of Labor Eco-

   nomics, 30(3), 521-554.


50. Van der Klaauw, W., & Wolpin, K. I. (2008). Social security and the

   retirement and savings behavior of low-income households. Journal of

                                 49
   Econometrics, 145(1-2), 21-42.


51. Weiss, D., & Lang, F. R. (2009). Thinking about my generation:

   Adaptive e¤ects of a dual age identity in later adulthood. Psychology

   and Aging, 24(3), 729–734.


52. Weiss, D., & Lang, F. R. (2012).     The two faces of age identity.

   GeroPsych: The Journal of Gerontopsychology and Geriatric Psychi-

   atry, 25(1), 5–14.




                                    50
                                                                                    !
                                             Figure 1: Gross Households Saving Rate in EU and UK: 2007-2022
                                                                 (Seasonally and calendar adjusted data)

                                    30,00


                                    25,00




   Saving!rate!(percentage!point)
                                    20,00


                                    15,00


                                    10,00


                                     5,00


                                     0,00




                                       European!Union!-!27!countries    Germany         France   Italy     Spain   United!Kingdom
                                                                                                                                    !

Source: Eurostat. The gross saving rate of households is defined as gross saving divided by gross disposable income,
with the latter being adjusted for the change in the net equity of households in pension funds reserves. Gross saving is the
part of the gross disposable income which is not spent as final consumption expenditure. Indicator described is calculated
on the basis of quarterly sector accounts data by institutional sectors. Household sector comprises all households,
household firms and Non-Profit Institutions Serving Households. Data are expressed in percentage, in seasonal and
calendar adjusted form. The United Kingdom left the European Union on 31 January 2020, and quarterly data from 2021
for British households are not available.
 Figure 2: Official daily communications of CFR by age groups in Italy over the period March 2020-
                                             June 2020




Note: The CFR indicates the case fatality rate, i.e., the proportion of people diagnosed with a certain disease,
who end up dying of it. Data Source: Istituto Superioriore per la Sanità.
Figure 3: Graphical analysis of the relationship between Fear indicators and the respondents age

The vertical line at age 62 denotes the cut-off. The vertical axis reports the indicator of Fear (Fear1
                        in panel (a), Fear2 in panel (b) and Fear3 in panel (c)




                                                                                              !

                                          Panel (a): Fear1
      Figure 3: continued




                            !
    Panel (b): Fear2




                            !

    Panel!(c): Fear3!!
!
    Figure 4: Graphical analysis of the relationship between the Fear’s indicators and the
                                       respondents age
!




                                                                                      !
Figure 5: RD validity check I: Density of the forcing variable (Respondent’s age)




                                                                                    !
                                        !
Figure 6: RD validity check II: continuity of observed covariates




                                                                    !
                       Figure!6:!continued!
!

!
          Figure!6:!continued!




!!!!!!!
Figure 7: Graphical analysis of the relationship between the willingness to vaccinate against
                            Covid-19 and the respondents ‘age




                                                                                     !
                            The vertical line at age 66 denotes the cut-off!
                         Table 1: Household heads Summary Statistics


                                                          Mean         Std. Dev.   Obs.
Age                                                       57.742        15.151     2346
  18-34                                                   0.068          0.252     2346
  35-44                                                   0.158          0.364     2346
  45-54                                                   0.208          0.406     2346
  55-64                                                   0.185          0.388     2346
  Equal/more than 65 years                                0.381          0.486     2346

Male                                                      0.685         0.465      2346
Female                                                    0.315         0.465      2346

Married                                                   0.519         0.499      2077

Family size                                               2.383         1.288      2346

N. of income earners                                      1.346         0.795      2346

N. of components over’s55                                 0.888         0.852      2346

Education:
None/Primary School Certificate                           0.192         0.394      2346

Lower Secondary School Certificate                        0.373         0.484      2346

Upper Secondary School Certificate                        0.295         0.456      2346

University Degree                                         0.141         0.348      2346

Occupational status:
Fix-Term employee                                         0.042         0.201      2346
Self-employed                                             0.099         0.298      2346
Unemployed                                                0.156         0.363      2346
Retirees                                                  0.345         0.475      2346

Resident in the North                                     0.491         0.500      2346
Resident in the Centre                                    0.201         0.401      2346
Resident in the South                                     0.307         0.462      2346
 Note. Statistics are computed using sample weights
!
                                                     Table 2: Summary statistics and definition of variables

                                                                                 DESCRIPTION                                          Mean    Std. Dev.   Obs.

    Making ends meet (very difficult)            Dummy variable equal to 1 if households report very difficult in making ends meet;   0.230    0.421      2346
                                                 reference category is very easily.
    Making ends meet (difficulty)                Dummy variable equal to 1 if households report difficulty in making ends meet;       0.338    0.473      2346
                                                 reference category is very easily.
    Making ends meet (easily)                    Dummy variable equal to 1 if households report easily in making ends meet;           0.295    0.456      2346
                                                 Reference category is very easily.
    Making ends meet (very easily)               Dummy variable equal to 1 if households report very easily in making ends meet.      0.137    0.343      2346


    Exp. on economic situation (worsen)          Dummy variable equal to 1 if households expect the economic situation to worsen      0.485    0.499      2346
                                                 in the next 12 months, and 0 otherwise.


    Exp. on economic situation (stay the same)   Dummy variable equal to 1 if households expect the economic situation to stay the    0.219    0.414      2346
                                                 same in the next 12 months, and 0 otherwise.


    Exp. on economic situation (improve)         Dummy variable equal to 1 if households expect the economic situation to improve     0.202    0.402      2346
                                                 in the next 12 months, and 0 otherwise


    Exp. on labor market (worsen)                Dummy variable equal to 1 if households expect the labor market to worsen in the     0.525    0.499      2346
                                                 next 12 months, and 0 otherwise.

    Exp. on labor market (stay the same)         Dummy variable equal to 1 if households expect the labor market to stay the same     0.214    0.410      2346
                                                 in the next 12 months, and 0 otherwise.

    Exp. on labor market (improve)               Dummy variable equal to 1 if households expect the labor market to improve in the    0.190    0.392      2346
                                                 next 12 months, and 0 otherwise.

!                                    !
                                                                Table 2: continued.

                                                                DESCRIPTION                                           Mean    St. Dev.   Obs.
Income Risk                    Continuous variable based on the self-reported probability of losing the job in the    0.052    0.208     2169
                               next 12 months for employed household members, in relation to the number of
                               income earners

Expected consumption changes   Answer to the question “Consider the expenditure for food, clothing and footwear,
                               home goods and services. How will your household change total expenditure for
                               these items in the next 3 months?” Multi-level dummy dependent variable (0=will
                               decrease; 1=will stay the same; 2=will increase).


Will decrease                                                                                                         0.318    0.466     2346
Will stay the same                                                                                                    0.631    0.483     2346
Will increase                                                                                                         0.051    0.220     2346

Fear                           Answer to the question “Compared to ordinary times, how frequently would do the
                               following activities if the daily Covid-19 cases in your region were <10, between
                               10 and 100; between 100 and 1000; more than 1000”
                               1 = I would stop doing it/rarely; 2 = less often; 3 = with the same frequency; 4 = a
                               little more often; 5 = I would do it more frequently.
                               The variable assumes value 0 (no fear) if respondent would continue the activity
                               with the same frequency in each assumed epidemiological scenario; 1 if respondent
                               declares to eliminate or drastically reduce/avoid this activity only when the Covid-
                               19 daily cases are more than 1000; 2 if respondent declares to eliminate or
                               drastically reduce/avoid the activity when the Covid-19 daily cases are more than
                               100; 3 if he/she declares to eliminate or drastically reduce/avoid the activity when
                               the Covid-19 daily cases are more than 10; and 4 (maximum fear) if respondent
                               declares to eliminate or drastically reduce/avoid the activity when the Covid-19
                               daily cases are less than 10.
Fear1                          The considered activity is: Shopping foods and essential goods.                        2.337    1.608     2119
Fear2                          The considered activity is: Make purchase in shops for clothing and furniture.         2.851    1.491     2183
Fear3                          The considered activity is: Visit hotel, restaurants and bars.                         2.971    1.466     2192
                                Table 3: Determinants of expected consumption changes: Linear Regression Model

    VARIABLES                                          (1)                          (2)                             (3)
                                           EXPECTED CONSUMPION           EXPECTED CONSUMPION            EXPECTED CONSUMPION
                                                 CHANGES                       CHANGES                        CHANGES
    Age                                            -0.050**                     -0.050***                       -0.049**
                                                    (0.020)                      (0.019)                         (0.019)
    Age squared                                   0.001***                      0.001***                       0.001***
                                                    (0.000)                      (0.000)                         (0.000)
    Male                                             0.165                        0.170*                          0.180*
                                                    (0.101)                      (0.099)                         (0.099)
    Married                                          -0.037                       -0.036                          -0.014
                                                    (0.095)                      (0.094)                         (0.093)
    Family size                                      0.048                         0.044                          0.057
                                                    (0.037)                      (0.035)                         (0.035)
    N. of income earners                             -0.026                       -0.021                          -0.055
                                                    (0.049)                      (0.048)                         (0.049)
    N. of over’s 55                                 -0.107*                      -0.098*                         -0.102*
                                                    (0.058)                      (0.057)                         (0.058)
    Fix-Term employee                                -0.226                       -0.200                          -0.243
                                                    (0.158)                      (0.162)                         (0.154)
    Self Employed                                 -0.375***                     -0.350***                      -0.388***
                                                    (0.110)                      (0.113)                         (0.112)
    Unemployed                                    -0.512***                     -0.577***                      -0.588***
                                                    (0.187)                      (0.184)                         (0.182)
    Making ends meet (very difficult)             -0.480***                     -0.542***                      -0.515***
                                                    (0.121)                      (0.120)                         (0.119)
    Making ends meet (difficult)                  -0.298***                     -0.310***                      -0.277***
                                                    (0.101)                      (0.098)                         (0.097)
    Making ends meet (easily)                        -0.104                       -0.099                          -0.057
                                                    (0.095)                      (0.091)                         (0.092)
!                     !
                                                             Table 3: continued.!

VARIABLES                                          (1)                               (2)                                (3)
                                        EXPECTED CONSUMPION               EXPECTED CONSUMPION                 EXPECTED CONSUMPION
                                              CHANGES                           CHANGES                             CHANGES
Exp. on the economic sit. (worsen)               -0.074                           -0.063                              -0.075
                                                (0.123)                          (0.120)                             (0.122)
Exp. on the economic sit. (stay the same)        -0.022                           -0.055                              -0.020
                                                (0.116)                          (0.113)                             (0.115)
Exp. on the labor market (worsen)               -0.221*                          -0.224*                             -0.210*
                                                (0.125)                          (0.121)                             (0.124)
Exp. on the labor market (stay the same)         0.0927                           0.115                               0.078
                                                (0.121)                          (0.117)                             (0.120)
Income risk                                      -0.275                          -0.337*                             -0.321*
                                                (0.177)                          (0.190)                             (0.185)
Fear1                                          -0.045***
                                                (0.008)
Fear2                                                                               -0.048***
                                                                                     (0.022)
Fear3                                                                                                                     -0.041***
                                                                                                                            (0.022)
Observations                                          1,640                           1,683                                  1,697
Notes: The dependent variable takes value 0 in case of an expected reduction in consumption, 1 in case of no expected change in consumption,
2 in case of an expected increase in consumption. Marginal effects reported. Standard errors in parenthesis. Significance: *** p<0.01, **
p<0.05, * p<0.1.
                                   Table 4: Determinants of expected consumption changes: Ordered Probit Model

    VARIABLES                                          (1)                           (2)                              (3)
                                           EXPECTED CONSUMPION            EXPECTED CONSUMPION             EXPECTED CONSUMPION
                                                 CHANGES                        CHANGES                         CHANGES
    Age                                            -0.048**                      -0.048***                        -0.051**
                                                    (0.020)                       (0.019)                          (0.019)
    Age squared                                    0.001***                      0.001***                        0.001***
                                                    (0.000)                       (0.000)                          (0.000)
    Male                                              0.185                        0.180*                           0.180*
                                                    (0.101)                       (0.098)                          (0.099)
    Married                                          -0.037                        -0.036                           -0.014
                                                    (0.098)                       (0.099)                          (0.094)
    Family size                                       0.047                         0.043                           0.058
                                                    (0.038)                       (0.035)                          (0.033)
    N. of income earners                             -0.025                        -0.022                           -0.056
                                                    (0.048)                       (0.049)                          (0.049)
    N. of over’s 55                                 -0.108*                       -0.099*                          -0.107*
                                                    (0.058)                       (0.057)                          (0.058)
    Fix-Term employee                                -0.226                        -0.200                           -0.243
                                                    (0.155)                       (0.166)                          (0.155)
    Self Employed                                 -0.377***                      -0.351***                       -0.389***
                                                    (0.111)                       (0.113)                          (0.112)
    Unemployed                                    -0.512***                      -0.577***                       -0.588***
                                                    (0.187)                       (0.184)                          (0.182)
    Making ends meet (very difficult)             -0.481***                      -0.522***                       -0.515***
                                                    (0.121)                       (0.120)                          (0.119)
    Making ends meet (difficult)                  -0.298***                      -0.310***                       -0.277***
                                                    (0.101)                       (0.098)                          (0.097)
!                     !
                                                                   Table 4: continued.

    VARIABLES                                              (1)                               (2)                                 (3)
                                                EXPECTED CONSUMPION                EXPECTED CONSUMPION                 EXPECTED CONSUMPION
                                                      CHANGES                            CHANGES                             CHANGES
    Making ends meet (easily)                            -0.104                            -0.099                              -0.057
                                                         (0.095)                          (0.091)                             (0.092)
    Exp. on the economic sit. (worsen)                   -0.074                            -0.063                              -0.075
                                                         (0.123)                          (0.120)                             (0.122)
    Exp. on the economic sit. (stay the same)            -0.022                            -0.055                              -0.020
                                                         (0.116)                          (0.113)                             (0.115)
    Exp. on the labor market (worsen)                   -0.223*                           -0.226*                             -0.213*
                                                         (0.125)                          (0.121)                             (0.124)
    Exp. on the labor market (stay the same)             0.0927                             0.115                              0.078
                                                         (0.121)                          (0.117)                             (0.120)
    Income risk                                          -0.275                           -0.337*                             -0.321*
                                                         (0.177)                          (0.190)                             (0.185)
    Fear1                                              -0.050***
                                                         (0.022)
    Fear2                                                                                     -0.049***
                                                                                                (0.022)
    Fear3                                                                                                                        -0.051***
                                                                                                                                   (0.022)
    Constant cut1                                          -2.600***                          -2.636***                          -2.557***
                                                             (0.599)                            (0.568)                            (0.578)
    Constant cut2                                            -0.352                             -0.362                             -0.284
                                                             (0.590)                            (0.557)                            (0.568)
    Observations                                              1,640                              1,683                              1,697
    Notes: The dependent variable takes value 0 in case of an expected reduction in consumption, 1 in case of expectation of no change in consumption,
    2 in case of an expected increase in consumption. Marginal effects reported. Standard errors in parenthesis. Significance: *** p<0.01, ** p<0.05, *
    p<0.1.
!                    !
                        Table 5: RD Non-Parametric Estimates of the discontinuity of Fear of Covid-19 at the age of 62

                                                                                 PANEL A
                                                                               (no covariates)
                                                           (1)                                      (2)                                  (3)
                                                          Fear1                                    Fear2                                Fear3
First-stage estimates                                    0.647***                                 0.759***                             0.409**
                                                          (0.160)                                  (0.171)                             (0.167)

Observations                                               1887                                     1942                                 1952
Bandwidth Type                                            MSE                                      MSE                                  MSE
Kernel                                                  Triangular                               Triangular                           Triangular
Eff. obs- Left of c                                        307                                      186                                   318
Eff. obs - Right of c                                      382                                      279                                   390
Bandwidth size                                            8.614                                    5.074                                8.017
Order polyn.                                                1                                        1                                     1
                                                                                 PANEL B
                                                                              (with covariates)
                                                           (1)                                      (2)                                  (3)
                                                          Fear1                                    Fear2                                Fear3
First-stage estimates                                    0.689***                                 0.844***                             0.470***
                                                          (0.166)                                  (0.243)                              (0.130)

Observations                                               1640                                     1683                                 1799
Bw Type                                                   MSE                                      MSE                                  MSE
Kernel                                                  Triangular                               Triangular                           Triangular
Eff. obs- Left of c                                        201                                      131                                   206
Eff. obs - Right of c                                      275                                      210                                   282
Bandwidth                                                 6.750                                    7.305                                6.112
Order polyn.                                                1                                        1                                     1
Note. Panel A and Panel B report the first-stage estimates, i.e., estimates of the discontinuity of the fear indicators arising at the age of 62, without
and with the inclusion of covariates respectively. The outcome variable in each column is a specific fear’s indicator. The running variable is Age of
respondent. In parentheses standard errors for clusters in age groups. Estimates are evaluated using the Stata program rdrobust. MSE indicates one
common mean squared error optimal bandwidth selector for the RD treatment effect estimator. Significance: *** p<0.01, ** p<0.05, * p<0.1
        Table 6: RD Non-Parametric Estimates of the discontinuity of Expected Consumption at the age of 62

                                                  (1)                                         (2)
                                     Expected consumption change                 Expected consumption change
                                            (no covariates)                            (with covariates)

Intention-to-treat effect                       -0.247***                                   -0.243***
                                                 (0.092)                                     (0.110)

Observations                                       2093                                        1818
Bandwidth Type                                    MSE                                         MSE
Kernel                                          Triangular                                  Triangular
Eff. obs- Left of c                                 169                                         102
Eff. obs - Right of c                               259                                         171
Bandwidth size                                    4.007                                       3.775
Order polyn.                                         1                                           1
Note. Estimates of the discontinuity of “Expected Consumption Changes” arising at the age of 62, without and with the
inclusion of covariates (column 1 and 2 respectively). The running variable is Age of respondent. In parentheses standard
errors adjusted for clusters in age groups. The estimates are evaluated using the Stata program rdrobust. MSE indicates one
common mean squared error optimal bandwidth selector for the RD treatment effect estimator. Significance: *** p<0.01, **
p<0.05, * p<0.1
    Table 7: Fuzzy-RD Non-Parametric Estimates of the discontinuity of Expected Consumption at the age of 62

                                                                 PANEL A
                                                               (no covariates)
                                                        (1)                         (2)                    (3)
                                                       Fear1                       Fear2                  Fear3


IV Estimates                                         -0.381***                   -0.325***              -0.609***
                                                       (0.060)                    (0.064)                (0.241)

                                                                  PANEL B
                                                               (with covariates)
                                                        (1)                         (2)                    (3)
                                                       Fear1                       Fear2                  Fear3


IV Estimates                                         -0.352***                   -0.287***              -0.510***
                                                       (0.067)                    (0.054)                (0.041)

Note. Panel A and Panel B report the IV second-stage estimates of the discontinuity of expected consumption arising at the
age of 62, without and with the inclusion of covariates respectively. The outcome variable is expected consumption change
instrumented with a specific fear’s indicator (each column refers to a fear’s indicator). The running variable is Age of
respondent. In parentheses standard errors adjusted for clusters in age groups. The estimates are evaluated using the Stata
program rdrobust. Significance: *** p<0.01, ** p<0.05, * p<0.1
              Table 8: Estimates of discontinuities in Fear and in Expected consumption changes at the age 62 with BSD inference

                                                        (1)                      (2)                       (3)                       (4)
                                                       Fear1                    Fear2                     Fear3             Expected Consumption
                                                                                                                                  Changes



Estimates                                             0.710**                   0.654*                    0.371                     -0.236**
BSD Standard Errors                                    0.356                     0.351                    0.311                       0.111
M                                                      0.042                     0.051                    0.038                       0.029

Note. The table reports the nonparametric RD estimates with Bounded second derivative (BSD) standard errors as discussed in Kolesár and Rothe (2018);
the Implied bandwidth is the one that minimizes the length of the resulting CI for a given choice of M, which is a bound on second derivative of the
conditional mean function using Armstrong and Kolesár (2020) rule of thumb. No covariates included.
                                Table 9: Summary Statistics: France


                                            Mean             Std. Dev   Obs.

Age (years)                                 67.893            8.279     1,726

Male                                         0.429            0.494     1,726

Female                                       0.571            0.494     1,726

Family size                                  1.940            0.821     1,726

Chronical Disease
Heart Problem                                0.122            0.328     1,715
Diabetes                                     0.124            0.329     1,715
Chronic Lung Disease                         0.064            0.244     1,715
Cancer or Malignant Tumour                   0.047            0.211     1,715
Hip Fracture                                 0.008            0.090     1,715
Others                                       0.359            0.479     1,715

Self-reported Health status
Poor                                         0.079            0.269     1,725
Good                                         0.470            0.499     1,725
Excellent                                    0.050            0.218     1,725

Know someone with Covid-19                   0.335            0.472     1,707

Willingness to be vaccinated against         0.894            0.307     1,723
Covid-19

Note. Statistics are computed using sample weights
                                Table 10: RD estimates using local polynomial regression

                                                      (1)             (2)             (3)                               (4)
                                                Willingness to Willingness to Willingness to                      Willingness to
                                                get the vaccine get the vaccine get the vaccine                   get the vaccine
                                                    France          France          France                            France
                                                 No covariates No covariates With covariates                      With Covariates
                                                  Bandwidth:      Bandwidth:      Bandwidth:                        Bandwidth:
                                                      MSE           Manual            MSE                             Manual

    RD_Estimate                                      0.135*               0.122**                0.128*                0.110**
                                                     (0.076)              (0.053)               (0.0748)               (0.052)

    Kernel                                          Uniform               Uniform               Uniform                Uniform
    Covariates                                        No                    No                    Yes                    Yes
    Observations                                     1762                  1762                  1732                   1732
    Eff. obs- Left of c                              179                   310                    178                    305
    Eff. obs - Right of c                            330                   497                    328                    590
    Bandwidth                                         3.8                    5                   3.573                    5
    Order polyn.                                       1                     1                     1                      1
    Note. Table reports the RD estimates using local polynomial regression without and with covariates for France. The outcome
    variable is the willingness to get the vaccine and the running variable is the respondent’s age. RD estimates using local polynomial
    regression without covariates in column (1) and (2) and with covariates in column (3) and (4). The estimates are implemented
    using the Stata program rdrobust. In column (1) and (3) MSE indicates one common mean squared error optimal bandwidth
    selector for the RD treatment effect estimator. In column (2) and (4) the bandwidth has been set equal to 5. Significance: ***
    p<0.01, ** p<0.05, * p<0.1

!


File and source

File

fear-of-contagion-and-consumption-expectations-evidence-using-government-communications-of-case-fatality-rates-during-th.pdf

Size

1,413,756 bytes

SHA-256

137cfb8158254d52f1f58f00e7307d20adb5aac9f92c3e20608a3bddfcfffe76

Our copy

fear-of-contagion-and-consumption-expectations-evidence-using-government-communications-of-case-fatality-rates-during-th.pdf

Original

doi.org