COVID-19 and Unemployment Risk: Lessons for the Vaccination Campaign
FFrom stay-at-home to return-to-work policies: COVID-19mortality, mobility and furlough schemes in Italy
Valentina Pieroni , Angelo Facchini ∗ , Massimo Riccaboni IMT School for Advanced Studies, Lucca ∗ corresponding: [email protected] February 9, 2021
Abstract
Assessing the economic impact of COVID-19 pandemic and vaccination roll-out strate-gies is essential for a rapid recovery. In this paper we analyze the impact of mobilitycontraction on employee furlough and excess deaths in Italy. We provide a link betweenthe reduction of mobility and excess deaths, confirming that the first countrywide lock-down has been effective in curtailing the COVID-19 epidemics. Our analysis points outthat a mobility contraction of 10% leads to a mortality reduction of 5% whereas it leadsto an increase of 50% in Wage Guarantee Funds allowed hours. Based on our results, wepropose a prioritizing policy for the administration of COVID-19 vaccines in the mostadvanced stage of a vaccination campaign, when healthy active population is left to bevaccinated. keywords : COVID-19 mortality; Furlough schemes; Economic impact of lockdowns;Vaccination rollout
The spread of the novel coronavirus (SARS-CoV-2) worldwide and the subsequent enforce-ment of strict containment measures by several national governments has severely impactedthe world economy, which shrank by 4.3% in 2020 (Chetty et al., 2020, World Bank, 2021).On the supply side, social distancing measures placed workers under stay-at-home orders,shut down ‘non-essential’ activities and challenged supply chains. On the demand side thepandemic has reduced consumer spending virtually wiping out demand in entire economicsectors. A whole bunch of literature analyzes the actual effectiveness of the most restrictivepolicies, such as countrywide lockdown, in preventing the contagion by reducing mobilityflows and discouraging social interactions (Acemoglu et al., 2020, Favero et al., 2020, Yooand Managi, 2020). In particular, recent work has shown mobility reductions to be followedby a significant drop in the number of new COVID cases and the death toll (Farboodi et al.,2020, Glaeser et al., 2020, Warren and Skillman, 2020).1 a r X i v : . [ phy s i c s . s o c - ph ] F e b lthough a stream of literature has largely investigated the epidemiological and socio-economic consequences of lockdown measures, there is still a paucity of evidence aboutthe effect of the reduction of mobility on employment. To fill this gap, in this paperwe investigate the implications of the containment policies by considering the amount ofworking hours allowed by the Italian government to be covered by the Wage GuaranteeFund in the aftermath of the first countrywide lockdown in 2020.This aspect is relevant for two main reasons. On the one hand, the impact of theCovid-19 crisis on Italian workers is dramatic, with 444 thousand jobs lost in 2020 . Thisis on top of 3.6 million furlough workers . On the other hand, the estimated expenditurefor Covid-related Wage Guarantee Funds allowed hours is almost 20 billion Euros in 2020(Commission, 2021). This is by far the main Covid induced increase of public budgetexpenditure in Italy. This calls for an analysis of the socio-economic consequences of thevaccination roll-out strategy in Italy, to speed up the recovery and to limit unemployment.Vaccination strategic distribution plans generally follow the WHO guidelines (WHO,2004) and are also consistent with the scientific literature (Medlock and Galvani, 2009, Sahet al., 2018). With regards to Europe, recently entered in force several criteria for prioritizingpopulation according to multiple criteria related to age, work, and health vulnerability. TheItalian strategic plan, released in December 2020 (della Sanit`a, 2020), provides a detaileddefinition of priorities involving the first administration phases covering the first nine monthsof 2021 and about 50% of the Italian population. Regarding the last phase, specific criteriahave not been provided yet. On the same page, we found other EU countries. For instance,Germany identified six categories of prioritization: the first five categories have differenturgency according to age and health risk, and cover about 30 million people, whereas thesixth category includes the remaining population, covering about 45 million people. Austria and Switzerland adopted similar rules, with a developing plan going up to the the secondquarter of 2021. In France two final phases, namely 4 and 5, involve younger population(over 18 year old) without comorbidities, but details on allocation criteria have not beendisclosed yet . On the same page is the UK plan , that identifies a phase aimed at achievingcoverage for the entire population, and will start after vaccinating priority groups targetingthose who are at greater risk of exposure and those who provide essential public services.Ireland introduces some specifications regarding the population at lower risk , assuming thatpriority is given to the 18-34 age group because it includes people who have more social As from 2020 employment statistics issued by ISTAT (Italian National Institute of Statistics) in February2021. See As from INPS data updated on October 10, 2020. For further information see The vaccination plan is going to be updated following the limitation imposed to AstraZeneca vaccinefor over 55s. https://solidarites-sante.gouv.fr/ https://assets.publishing.service.gov.uk/ identifies somecategories of the population and provides criteria for assessing their higher or lower priority,and among the categories of medium-high priority are still workers in essential sectors (toensure the normal functioning of society) and people who are vulnerable because of theirsocio-economic conditions (e.g., those with precarious work, people in the lower incomebracket, etc). Indeed, drivers of socio-economic nature are found in the Irish and Spanishcases, that consider among the priority classes those who live in crowded neighborhoods orhousing (therefore at high risk of outbreak). There is therefore the need to define furthercriteria for the prioritization of vaccination for people not working in essential or strategicservices, that are substantially equivalent and generally represent a consistent share of thepopulation.To this end, here we introduce a criterion for the vaccine distribution to the share ofnon prioritized population, meaning the healthy and active population. We start providingadditional evidence on the effectiveness of restriction to mobility for the Italian case. Resultsshow that under the public health point of view, a ten-percent drop in mobility explainsa 5 percent drop in excess deaths in the following month. Furthermore, we analyze theimpact of mobility reduction on Wage Guarantee Fund (number of allowed working hours),as a proxy for the suspension of the economic activity due to Covid-19 and a proxy for theinduced public expenses. Results show that a 10% drop in human mobility corresponds toa 50% increase of the Wage Guarantee Fund (WGF) expressed in full time equivalent unitsduring the following month. Available data refer to a time window spanning from March toAugust 2020. We run a fixed-effects model on a monthly longitudinal dataset comprising 107Italian provinces (NUTS3 regions). For the best performance of the methods implementedwe also addressed potential endogeneity issues concerning our main variable of interest,mobility range, following an instrumental approach as in Glaeser et al. (2020).As a further result we observe higher mobility to be associated with a greater share ofessential working residents, hence we provide evidence supporting the inclusion of workersin essential sectors among the priority categories: if more people are allowed to move, sinceemployed in essential sectors or not eligible for remote working, we expect the risk of thecontagion to increase. The intervention is even more critical with respect to those essentialjobs which imply a high risk of exposure . Here the main aim of reducing morbidityand mortality comes together a socio-economic rationale, as the one of limiting economicdisruption. https://mscbs.gob.es As previously mentioned, the Irish provisional allocation plan identifies people working in essential jobsat a high risk of exposure among the priority categories, with the rational of minimizing harm while reducingeconomic disruption. A lower degree of priority is associated to workers in occupations which are essentialto the functioning of society (e.g. goods-producing industries) but where precautionary measures can beadopted without much difficulty. Also Spain provides the rationale to evaluate essential workers’ prioritylevel, by taking into account economic criteria and assessing the risk of exposure and of developing severemorbidity.
The analysis of the actual effectiveness of restrictive mobility policies to prevent COVID-19infections has been addressed in a body of scientific research spanning multiple disciplines.The consequences of such policies have been examined on an international scale, and arenowadays covered by a significant and rapidly expanding literature. Regarding mobilityrestriction policies in the U.S. (Glaeser et al., 2020) employ data on five U.S. cities toestimate the effectiveness of lockdowns and other restrictions in limiting the spread ofcoronavirus disease. The authors perform a panel and a cross sectional regression of thelogarithm of COVID-19 cases per capita on the percentage drop in mobility, employingthe two-periods lagged value of the explanatory variable in the panel setting. To addresspotential endogeneity issues concerning the main regressor of interest, mobility has beeninstrumented by the employment-weighted average share of essential workers and by theemployment-weighted average telecommuting share across industries at the zip code level.According to their main instrumental variable panel specification, when controlling for zipand week fixed effects, the authors find that a drop in mobility by 10 percent points leadsto a 30 percent decline in COVID-19 cases per capita. In an additional specification of thecross sectional model, they find a positive and significant association between the logarithmof per-capita deaths and mobility changes, which is robust to the inclusion of controls wheninstrumenting for mobility.Regarding Germany, Krenz and Strulik (2020) implement an instrumental variable strat-egy to investigate the association between COVID-19 diffusion and mobility containmentat a regional level (NUTS3 regions). As an instrument for mobility they employ a met-ric assessing the quality of the road infrastructure in German regions, namely the averagetravel time on roads towards the next major urban center is used as a proxy for remoteness.The authors argue that the impact of ”road infrastructure” on the spread of the diseasegoes through the effect it has on mobility flows. By regressing the logarithm of COVID-19 Data on essential industries from Minnesota and Delaware are used to this end. in an IV cross sectional setting, this study shows a neg-ative and significant association between a change in mobility and COVID-19 disease cases.According to the authors’ interpretation, German regions with a higher decline in mobilityon Easter Sunday are those which have accumulated the largest number of COVID-19 cases.Besides, the first stage of the IV model shows a positive relationship between mobility dropsand accessibility defined as ”travel time to the next urban center”, suggesting that mobilityflows declined most in those areas which are less remote (i.e. metropolitan areas).Moving to the Italian case, Borsati et al. (2020) provide evidence on the associationbetween public transports usage and the number of excess deaths, as transport modes havebeen addressed as a potential driver of the contagion in the ongoing debate. Using dataat local labour markets level the authors detect a non statistically significant correlationbetween the propensity to use public transports and excess deaths as recorded during thefirst six months of 2020. They find instead a positive and significant association betweenthe dependent variable and synthetic indices for internal and external commuting flows computed on 2011 national census data, and this result is still consistent in significance andsign when controlling for economic and demographic variables as well as for individual andtime fixed effects.Focusing on excess mortality, the work by Borri et al. (2020) explores the causal effectof lockdown policies in Italy on mortality by COVID-19 (again proxied by excess deaths)and mobility. Implementing a difference in differences model on an daily panel dataset, theauthors show that a higher intensity of the lockdown is associated to a significant decreasein the number of excess deaths with respect to the whole population, and this holds truein particular for older people (in the range 40-64 and beyond). A second finding is thatmunicipalities with a higher drop in the share of active people due to the lockdown arethose showing a stronger contraction in mobility.The analysis by Bonaccorsi et al. (2020) examines instead the socio-economic conse-quences of the Italian lockdown. By employing a network quantity, the node efficiency,to track changes in connectivity between municipalities 14 days after the lockdown withrespect to 14 days before the lockdown, the authors argue that richer municipalities interms of social indicators (index of material and social well-being) and fiscal capacity arethose showing a stronger contraction in mobility. At the same time, however, they observethat among those municipalities experiencing a higher drop in mobility the contraction ismuch higher for municipalities with a lower average income and higher levels of inequality(measured as the ration between mean and median income). Changes in mobility have been measured comparing mobility flows on Easter Sunday 2020 to an averageSunday in April 2019. Internal commuting for local labour market (LLM) i is computed as the ratio between the number ofpeople moving between municipalities within i and the population of i , while external commuting flowsaccounts for the number of people moving from i to other LLMs and the number of people moving to i fromother LLMs, again normalized on LLM i population. According to the definition given bu the authors, a municipality experiences a more intense lockdown ifthe reduction in the share of active population following the lockdown is above the median reduction acrossall municipalities located in the same province. Temporary shutdown of non essential economic activities as from DPCM March 22, 2020.
5n this expanding stream of literature, lockdown policies have been shown to explainchanges in epidemiological data often through their effects on mobility, but according tothe work by Goolsbee and Syverson (2021) on U.S. data, human mobility flows (especiallythose accounting for consumers’ visits to business locations and stores) are just partiallydriven by the enforcement of stay-at-home/shelter-in-place orders, as they may also arisefrom voluntary behavioral adjustments due to the fear of the pandemic.Following this short literature review, we notice that although the ongoing scientificresearch is largely dealing with the epidemiological and socio-economic impact of the lock-down even in terms of market labour flows (Casarico and Lattanzio, 2020), we still havelittle evidence about the effects of lockdown policies on measures which could be taken asproxies for public expenditure and economic activity contraction.
Data have been collected according to the three dimensions involved in the analysis: fur-lough schemes, mobility, and mortality. Furlough schemes are measured as Wage GuaranteeFunds hours that have been authorized by the Italian Government, as a wage integrationmeasure. Data are released by INPS (the Italian National Social Welfare Institution) andcover the period January-September 2020 (INPS, 2020). In addition, we considered theshare of working population and the number of workers according to the six digits ATECO(numerical classification of economic activities, the Italian version of European NACE).Data have been collected from ORBIS database . We computed for each province theshare of workers employed in those ATECO codes not suspended by the Italian govern-ment.Mobility data have been collected from the Facebook Data for Good program (Maaset al., 2019), whose reliability has been tested against the census commuting data collectedby the Italian statistical institute in 2011 (refer to appendix A for details). Finally, as rep-resentative of the epidemic spreading, we considered the excess mortality data at municipallevel collected by ISTAT (2020) expressed as the difference between the number of deathsrecorded in 2020 and the average number of deaths occurred between 2015 and 2019 inthe same period. As discussed in Buonanno et al. (2020), excess death toll is a reliableproxy of mortality by COVID-19. Such an assumption is needed to overcome the potentialissues related to the endogeneity of testing policies (especially during the first wave of theepidemics), hospital capacity and difference in death classification at the local level. Table1 shows that data span different time and spatial resolutions, ranging from monthly data ofWage Guarantee funds to 8-hourly data of Movement Range. With regards to the spatialaggregation variability of data, we observe a variability ranging from administrative regionsof Wage Guarantee funds to municipality level of excess mortality.The Facebook Data for Good program makes available different sets of data (Maaset al., 2019), covering both mobility flows between administrative regions and mobility We used firm-level employment data as from 2019 fiscal year reporting. the use of already existing wageguarantee schemes against the pandemic crisis to strengthen employment protection. In ajoint work from INPS and Bank of Italy (INPS and d’Italia, 2020), it is reported that inthe months of March and April 2020 around 50% of employers in the private sector havebeen allowed to use wage compensation schemes according to the new rules in force. Thiskind of intervention turns into lower labour costs for the firm but translates into a loss forthe employee: estimates by INPS-Bank of Italy (INPS and d’Italia, 2020) show a meanmonthly-gross income loss of around 27%. Moreover since wage subsidies are granted bythe government this leads to greater public expenditures.The growth in requests to be allowed to wage integration schemes by the employers canbe partially explained in light of an additional labour market measure issued in March,the firing freeze, that is a temporary suspension of firings. Starting from around the 12 th week of 2020 (which coincides more or less with the introduction of the firing freeze and theextension of wage integration schemes) firings dropped sharply with respect to their averagelevel in 2017-2019 (Casarico and Lattanzio, 2020); starting from week 9, a sharp decreasehas been detected in the number of hirings as well.National public policies have had a remarkable impact on labour market flows: accordingto recent estimates (Viviano, 2020), if measures like the extension of wage supplementationschemes together with firings freeze and financial supports for firms had never been issuedthere would have been 600 thousand more firings in 2020 because of the pandemic crisis.Figure 1 shows how intense the use of the Wage Guarantee Fund has been on averageover the last year. The figure shows the monthly average Wage Guarantee Fund (in termsof accumulated hours), the weekly average number of excess deaths and the weekly average Decree Law n. 18/2020 issued on March 17. th weekmobility drops significantly (w.r.t. the baseline) while almost simultaneously the numberof excess deaths shows a sharp increase reaching its peak during the 12 th week of theyear (around the last ten days of March). At the end of February the first containmentmeasures had been issued but just on a local scale, addressing those areas where newCOVID-19 cases had been recorded. However, a first contraction of mobility flows and agrowth in deaths can be detected. We also observe a peak in the average number of allowedworking hours to be covered with the Wage Guarantee Fund in April, with about one monthdelay with respect to the introduction of more restrictive measures, corresponding with thenational lockdown on March 12. This may be explained by the fact that the time whenthe employer is allowed to use the wage guarantee schemes do not correspond with theactual temporary suspension of the working activity INPS and d’Italia (2020).While excess deaths have been computed by comparing the number of deaths in 2020with average pre-pandemic death levels in the same time window, the amount of the WageGuarantee Fund has not. However, the intensity in the use of the Fund in the early monthsof 2020 before the contagion outbreak (January and February) can be taken as a referencepoint and the graph shows how early levels represent just a small fraction of the peak whichcan be observed around week 18.Maps in figure 7, Appendix B, instead, plot the monthly distribution of the WageGuarantee Fund allowed hours across Italian NUTS 3 regions: darker shades point outthose areas where furlough schemes (WGF hours) have been used with higher intensity ineach month.Before composing the panel, data have been checked for consistency and have beenaveraged/rescaled (whenever possible) in order to fit the weekly variation and the spatialaggregation of an administrative region. As a result, we obtained a longitudinal datasetcomprising monthly observations on a cross section of 107 Italian provinces. Prime Ministerial Decree February 23, 2020. Prime Ministerial Decree March 11, 2020
Note: the plot displays the trend over time of the monthly average amount of the Wage Guarantee Fund(average allowed working hours), the average number of excess deaths per week and in weekly averagemobility changes. All variables are expressed in normalized units: the Wage Guarantee Fund and excessdeaths have been normalized on their maximum while mobility range has been normalized on its minimum.Two-weeks moving average are reported for excess deaths and mobility range.
To explore the relationship between the dependent variables and the main explanatory vari-able, namely
Mobility Range , we employed a linear model for longitudinal data, accountingfor Italian provinces’ heterogeneity. The linear model is expressed as follows ln ( y ) it = βM ob.Range i ( t − + δLockdown t + pv i + ε it (1)where pv i denotes the individual-specific fixed effects, which allow to control for provinces’time invariant unobserved characteristics. Since we assume pv i to be potentially correlatedwith the observed regressor we implemented a fixed-effects model. The model also includesa dummy named Lockdown which takes value 1 in those months when the national lockdownwas in force - March, April and May - to control for potential time-related effects due tothe imposed restrictions. 9ere y it stands for Excess Deaths or Wage Guarantee Fund , since equation (1) has beenestimated using each of them as the dependent variable. In both cases the logarithm of theresponse variable has been regressed on a period lagged value of the explanatory variable
Mobility Range . With regards to the Wage Guarantee Fund especially, a delay could occurbetween the time in which firms may take advantage of the wage supplementation schemesintroduced against the crisis and the time in which it is officially authorized and recorded(INPS and d’Italia, 2020). A time window is likely to occur, as well, before we observea decline in the level of deaths at least partially driven by an adjustment of collectivebehaviours as a reaction to the spread of the virus (Borri et al., 2020).Model (1) has been then refined in order to overcome potential endogeneity issues con-cerning the main explanatory variable
Mobility Range .As already pointed out in previous scientific works, it is plausible to assume the en-dogeneity of a mobility measure with respect to variables which are strictly related to thespread of the disease like the number of COVID-19 cases or the count of deaths (Borriet al., 2020, Glaeser et al., 2020, Krenz and Strulik, 2020). A potential reversed causalityissue may affect the estimates, assuming that mobility flows are adjusted when people ob-serve an increase (or decrease) in the level of deaths potentially due to the contagion. Thesame argument can be extended with respect to the relationship between mobility and theamount of the Wage Guarantee Fund: a fall in commuting flows can explain an increase inthe Wage Guarantee Fund, since the enforcement of containment measures meant to dis-courage mobility traffic and limit social interactions could foster the use of wage guaranteeschemes by the employer even in attempt to reduce physical proximity in the workplace.However, a temporary suspensions of working activities could itself explain a further dropin commuting flows. This could be the most intuitive way to interpret the relationshipbetween mobility and furlough schemes but is not the only one, as mobility could impactthe Wage Guarantee Fund even through different channels. If less people move because ofcontainment rules or since they fear the contagion, we may observe a decline in the demandfor goods and services by final consumers. In turn, entrepreneurs may be led to a temporaryreduction of working time and to ask for wage compensation schemes in order to cope witha contraction in the demand .To overcome potential endogeneity-issues, we employed two instruments for MobilityRange , developing different specifications of our model. The first instrumental variable (IV),the betweenness centrality (Newman, 2010) of Italian NUTS3 regions, describes topologicalproperties of the mobility network built on Facebook data movement between administrativeregions . This measure is used as a proxy of the remoteness of Italian provinces, in the same Even as a potential effect of consumer substitution patterns (Goolsbee and Syverson, 2021). To measure the province centrality in the mentioned network we computed also the pagerank, then westudied the variation in nodal efficiency relying on mobility data as previously done by Bonaccorsi et al.(2020). We performed several trials employing each quantity alone and combined as instrumental variablesin the econometric model. We finally opted not to use more than one network quantity as an instrument(e.g. when the page rank is used as an excluded instrument together with the betweenness it appears tobe redundant), and we chose the betweenness to be used alone and combined with the share of essentialresidents. i ).The choice of the second instrument has been inspired by Glaeser et al. (2020):looking at the provisions of Prime Ministerial Decrees issued between March and May wecomputed for each Italian province the time-varying Share of Essential Residents (or
ShareEssentials ), that is the share of labour force which was allowed to move during the firstnational lockdown since employed in economic sectors designated as essential by the Italiangovernment. Finally, the share of authorized employees has been multiplied by province i to proxy the share of essential workers. Following the argumentin Glaeser et al. (2020) and Krenz and Strulik (2020), we assume that the centrality of aprovince in the mobility network and the share of people employed in essential industrieshave an impact on excess deaths just through mobility flows. A similar argument appliesfor Wage Guarantee Funds allowed hours.The first stage and main equations for the IV model are given by M ob.Range i ( t − = πIV i ( t − + γLockdown t + pv i + η it (2) ln ( y ) it = βM ob.Range i ( t − + δLockdown t + pv i + ε it (3)both stages control for individual-specific fixed effects and include the Lockdown dummy.We estimated three specifications of the model: the first one employs only betweennesscentrality as instrumental variable, the second one includes just
Share Essentials while thethird one uses both variables as instruments. To be in line with the instrumented variableboth IVs have been lagged by one month.With respect to the Wage Guarantee Fund we mostly rely on the IV specification employ-ing share essentials together with centrality, or using the betweenness alone as instrumentalvariable. Finally, we performed a GMM estimation of the coefficients. We refer to Glaeser et al. (2020) in the choice of the instrument for mobility but we computed themeasure according to a different formula. Our references are Dpcm March 11, Dpcm March 22, Dpcm April 1, Dpcm April 10, Dpcm April 26and Dpcm May 17, 2020. Source of employment data is ISTAT (Italian National Institute of Statistics) Labour Force Survey. Results
Estimates in table 3 have been obtained regressing the logarithm of excess deaths onthe main explanatory variable, namely mobility range, which keeps track of the change inmobility occurred in Italy during and after the first national lockdown.Results in model [A] have been estimated on the full panel comprising all 107 Italianprovinces and a time window spanning from March 2020 to October 2020. Column (1)reports fixed-effects estimates from model (1) while columns from (2) to (4) display theresults for the three IV specifications of the model. To overcome potential endogeneityissues concerning our main regressor, in column (2) we instrument for mobility range withprovince centrality in the mobility network, while specification in column (3) employs shareessentials as the external instrument, and model in column (4) uses both variables.As previously mentioned, each specification includes the Lockdown dummy variable,which takes value 1 in those months when the first national lockdown was in force in Italy:March, April and May.All columns report a positive and statistically significant association between a changein mobility and excess deaths in the full-length period.Model (1) shows that excess deaths increases by 0.35 percent at time t if mobilityincreases by one percent in the previous month . Point estimates from specification (1)should be interpreted carefully. Indeed, the magnitude of the effect grows as we instrumentour main regressor, as in specifications (2)-(4), suggesting a downward bias potentially dueto endogeneity issues. First stage F-statistics suggest that the instruments are actuallystrong.In order to point out a potential variation over time in the effect mobility on the depen-dent variable all estimates have been repeated by splitting the sample in two periods, onecomprising months from March to May (model [B]) when the national lockdown was in force( lockdown period), and the other including observations from June to October (model [C]).As expected, the increase in mobility led to a greater growth in deaths during the lockdownperiod (model [B]), when the first wave of the pandemic reached its peak in Italy, and thiseffect weakened in the following months (model [C]). All coefficients are significant exceptfor model [C](2), the one instrumenting for the centrality of provinces, when referring tothe post-lockdown period.Furthermore, results in section [B] display an increase in the coefficient once we instru-ment the main explanatory variable, while estimates in section [C] suggest the presence ofan upward bias in the baseline model [C](1). To cope with negative values we first rescaled variable excess deaths by adding the absolute value of itsminimum (i.e. 478 .
2) to each observation then we took the logarithm. About the interpretation of the coefficient: since we are dealing with semi-elasticities we say that aunit increase in mobility implies a ( β ∗
100 )% variation in the dependent variable. ’Mobility range’ is notexpressed in percentage points, meaning that a unit change in mobility means actually a 100% change (inorder to be expressed in percentage point it should be multiplied by 100). To get the effect of a one percentchange in mobility on excess mortality we should divide the coefficient by 100, that is . = 0 . Betweenness centrality of provinceswe observe a negative and significant coefficient instead, suggesting a higher contraction inmobility flows for provinces with a higher centrality in the national network.Let’s now turn our attention to the relationship between drops in mobility flows and theamount of the Wage Gurantee Fund as described in table 5.The dependent variable has been expressed in full time equivalent units for a moreintuitive interpretation. The metric, originally expressed in terms of allowed working hoursin a month, has been divided by the maximum number of working hours in a full-timemonthly schedule (excluding week-ends and bank holidays). If we assume that each workerhas been laid-off for an amount of hours close to the monthly full-time workload, the measureas above can be read as the number of full-time equivalent working employees who havebeen temporary suspended from work in a month. It seems reasonable actually, since inMarch and April 2020, each individual put under wage compensation schemes has beenlaid-off for an average amount of hours equal to 154, accounting for around 90% of thefull-time monthly schedule (INPS and d’Italia, 2020).As for excess mortality, all models have been estimated on the full sample comprisingtime units from March to August (model version [A]) and on two smaller samples includingthe same individual units observed in subsequent time windows, that is when the nationallockdown was in force (model [B] - March to May) and in the following months (model [C]- June to August).Results from the full sample regression show a negative and significant association be-tween
Mobility Range and the dependent variable, meaning that for every one percent dropin mobility at time ( t −
1) we observe a 4 .
35 percent increase in the full time equivalentWage Guarantee Fund in the following month (that is to say an increase in the number offull-time working employees under wage guarantee schemes, given a possible interpretationof a full time equivalent unit) according to specification (2) when instrumenting by thebetweenness centrality.This suggests that the enforcement of national policies meant to prevent the contagiondiscouraged mobility flows and fostered the use of wage compensation schemes providedby law to support workers. The relationship between mobility and furlough schemescould additionally be explained by a change in the demand for goods and services by finalconsumers.Estimates of β tend to decrease as we instrument our main regressor by just one orboth the selected IVs (columns (2)-(4)), even though results from column (4) should betaken carefully, since we have to reject the null hypothesis from the Sargan-Hansen test ofoveridentifying restrictions (Hansen J statistic=4 . . Again, recall that the explanatory variable Mobility Range is not expressed in percentage units andshould be multiplied by 100 to be so. Decree Law No. 18/2020 of 17 March 2020 − .
67, p-value= 0 . − .
87, p-value= 0 . − .
54, p-value= 0 . ln Excess Deaths it [A] Full Sample Regression (1) (2) (3) (4)FE IV IV IVMobility range i ( t − ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.046) (0.252) (0.066) (0.065)Lockdown 0.256 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.038) (0.077) (0.032) (0.032)Constant 6.232 ∗∗∗ (0.007)Observations 856 855 856 855Number Ids 107 107 107 107Individual FE Yes Yes Yes YesOverall R [B] Split sample regression (March to May) (1) (2) (3) (4)FE IV IV IVMobility range i ( t − ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.062) (0.338) (0.082) (0.081)Constant 6.501 ∗∗∗ (0.022)Observations 321 320 321 320Number Ids 107 107 107 107Individual FE Yes Yes Yes YesOverall R [C] Split sample regression (June to October) (1) (2) (3) (4)FE IV IV IVMobility range i ( t − ∗∗∗ ∗∗∗ ∗∗∗ (0.040) (0.282) (0.042) (0.040)Constant 6.225 ∗∗∗ (0.003)Observations 535 535 535 535Number Ids 107 107 107 107Individual FE Yes Yes Yes YesOverall R Robust standard errors in parentheses ∗ p < . ∗∗ p < . ∗∗∗ p < . [A] Full Sample First Stage IV (2) (3) (4)Mobility Mobility Mobilityrange i ( t − range i ( t − range i ( t − Betweenness i ( t − -1.192 ∗∗∗ -0.361 ∗∗∗ (0.173) (0.127)Lockdown -0.269 ∗∗∗ -0.098 ∗∗∗ -0.098 ∗∗∗ (0.014) (0.011) (0.011)Share essentials i ( t − ∗∗∗ ∗∗∗ (0.001) (0.001)Observations 855 856 855Number Ids 107 107 107Root MSE 0.166 0.109 0.109Individual FE Yes Yes Yes [B] Full Sample Reduced form IV (2) (3) (4)ln Excess ln Excess ln ExcessDeaths it Deaths it Deaths it Betweenness i ( t − -0.851 ∗∗∗ -0.418(0.302) (0.288)Lockdown 0.163 ∗∗∗ ∗∗∗ ∗∗∗ (0.018) (0.025) (0.025)Share essentials i ( t − ∗∗∗ ∗∗∗ (0.001) (0.001)Observations 855 856 855Number Ids 107 107 107Root MSE 0.208 0.198 0.198Individual FE Yes Yes Yes Robust standard errors in parentheses ∗ p < . ∗∗ p < . ∗∗∗ p < . ln Wage Guarantee Fund FTE it [A] Full Sample Regression (1) (2) (3) (4)FE IV IV IVMobility range i ( t − -6.232 ∗∗∗ -4.354 ∗∗∗ -5.736 ∗∗∗ -5.675 ∗∗∗ (0.208) (0.743) (0.262) (0.261)Lockdown -1.717 ∗∗∗ -1.265 ∗∗∗ -1.597 ∗∗∗ -1.590 ∗∗∗ (0.093) (0.219) (0.125) (0.126)Constant 8.226 ∗∗∗ (0.020)Observations 619 618 619 618Number Ids 104 104 104 104Individual FE Yes Yes Yes YesOverall R [B] Split sample regression (March to May) (1) (2) (3) (4)FE IV IV IVMobility range i ( t − -7.610 ∗∗∗ -5.188 ∗∗∗ -6.005 ∗∗∗ -5.979 ∗∗∗ (0.251) (1.013) (0.286) (0.287)Constant 6.018 ∗∗∗ (0.090)Observations 307 306 307 306Number Ids 104 104 104 104Individual FE Yes Yes Yes YesOverall R [C] Split sample regression (June to August) (1) (2) (3) (4)FE IV IV IVMobility range i ( t − -0.875 ∗∗∗ -1.317 -0.424 ∗ -0.469 ∗ (0.251) (0.854) (0.255) (0.251)Constant 8.849 ∗∗∗ (0.029)Observations 312 312 312 312Number Ids 104 104 104 104Individual FE Yes Yes Yes YesOverall R Robust standard errors in parentheses ∗ p < . ∗∗ p < . ∗∗∗ p < . [A] Full Sample First Stage IV (2) (3) (4)Mobility Mobility Mobilityrange i ( t − range i ( t − range i ( t − Betweenness i ( t − -1.257 ∗∗∗ -0.304 ∗∗ (0.193) (0.137)Lockdown -0.234 ∗∗∗ -0.066 ∗∗∗ -0.067 ∗∗∗ (0.015) (0.012) (0.012)Share essentials i ( t − ∗∗∗ ∗∗∗ (0.001) (0.001)Observations 618 619 618Number Ids 104 104 104Root MSE 0.187 0.118 0.118Individual FE Yes Yes Yes [B] Full Sample Reduced form IV (2) (3) (4)ln WGF FTE it ln WGF FTE it ln WGF FTE it Betweenness i ( t − ∗∗∗ -0.079(1.386) (1.322)Lockdown -0.244 ∗ -1.217 ∗∗∗ -1.219 ∗∗∗ (0.126) (0.149) (0.149)Share essentials i ( t − -0.092 ∗∗∗ -0.092 ∗∗∗ (0.007) (0.007)Observations 618 619 618Number Ids 104 104 104Root MSE 1.524 1.269 1.270Individual FE Yes Yes Yes Robust standard errors in parentheses ∗ p < . ∗∗ p < . ∗∗∗ p < . Our results have two main implications for the Italian vaccine roll-out strategy. Whenmoving to vaccinate the health and active share of the population, essential workers andworkers not eligible for remote working should be prioritized, since they increase mobilitythus inducing higher excess mortality. Second, based on the results of the econometric anal-ysis we propose to prioritize those areas in which the effect of mobility on Wage GuaranteeFund is stronger. Areas are identified according to the estimates of the individual-specificfixed effects as from the full-sample IV model (specification [A](4), table 5), consideringthe Wage Guarantee Fund as the dependent variable. Fixed effects coefficients are graphi-cally represented in figure 2, where panel ’b’ shows which NUTS 3 regions explain a higher18igure 2: Province-specific fixed effects estimates (a) Excess Deaths (b) Wage Guarantee FundNotes: the map shows a graphical representation of the province-specific fixed effects estimated throughequation ln (WGF FTE) it = β + β Mob.Range i ( t − + δLockdown t + (cid:80) Nj =2 pv j d j,it + ε it , when instrumentingMobility Range by both provice centrality and the share of essential residents (IV model, specification (4),section [A], tables 3 and 5). Color intensity for each province i is proportional to coefficient pv j . increase in the Wage Guarantee Fund (FTE units). Fixed effects estimates account fortime-invariant provinces’ effects like demographic and socio-economic characteristics (whichreasonably remain stable in the period we consider).The provinces that have been most in need for wages supplementation schemes havebeen identified according to fixed effects estimates performed on the entire period, spanningfrom March to August 2020, and on the lockdown time only (March to May). We assumethat stricter restrictions are likely to be enforced in the months when the last steps ofthe campaign are about to start. We could instead refer to the estimates obtained whenfocusing the post-lockdown period (June to August) if we expect milder (or almost absent)restrictions to be enforced.We compare our allocation criterion with a benchmark based on Working population ,i.e. the number of people employed per administrative region.Each province i is ranked according to the two criteria explained above, and we indicate The number of people employed in province i has been obtained by multiplying the 2019 value of theemployment rate of people aged 20 to 64 as from ISTAT (Labour Force Survey), by the number of residentsaged between 20 and 64. W GFi with R W Pi and R W GFi the position of the province i in the Working Population, and WageGuarantee Fund criteria. To highlight possible inequalities we compare the criteria bysubtracting WP to WGF rankings∆ W GFi = R W GFi − R W Pi (4)where ∆ i W GF is the difference in the ranking positions between { W GF, EM } and theworking force ranking. The distribution of ∆ is reported in figure 3. The intensity of thecolor is proportional to ∆ i W GF . Areas in light colors between blue and red tones arethose with a similar position in both criteria (∆ ∼ ± i W GF <
0, then they would be disadvantaged in case of WP criterion. On the otherhand, provinces in red and dark red shades (∆ > Final discussion
In this paper we analyse the impact of human mobility on excess mortality and the useof furlough schemes in Italy. We assume that, safe return-to-work will be possible forvaccinated workers, reactivating mobility and restoring full production capacity. This isbecause the negative health consequences of human mobility will be neutralized. Therefore,we propose a vaccine prioritization policy of the health and active share of the populationin two stages. First, access to vaccination should be guaranteed to essential workers andthe ones not eligible for remote working. Then, return-to-work should be facilitated forthe beneficiaries of wage guarantee schemes. This will be beneficial both in terms of areallocation and more efficient use of public funds and in terms of reduction of potentialjob losses . It is important to highlight that our recommendations refer to the last phaseof the vaccination campaign, when vulnerable categories according to the national strategicplan have already been vaccinated and immunized against the virus (della Sanit`a, 2020).The proposed strategy puts in advantage those workers employed in the administrativeareas in which wage integration measures have been used more, allowing them to come backsooner to a safe workplace , triggering a gradual economic recovery. The expected benefitof this policy can be interpreted mostly in terms of a gradual resumption of most economicactivities and in terms of potential alternative allocations of public funds. We recall that,according to the European Commission (Commission, 2021), the Italian government hascommitted around 19 billion euros to cover wage supplementation schemes , accountingfor around 70% of the total amount committed to employment support measures. With theapproval of the 2021 Italian Budget Law , the use of wage guarantee schemes against theCOVID-19 crisis has been extended until the end of March 2021 and until the end of June2021, in the latter case with regard only to the Derogatory Wage Guarantee Fund andthe Wage Integration Fund . Further measures on employment protection are currentlyunder discussion.To support our proposal we explored the link between the drop in mobility and theamount of the Wage Guarantee Fund expressed in full time equivalent units, also providingevidence on the association between changes in mobility and measures related to the spread A similar argument could be extended even to those workers who lost their job because of the pandemiccrisis: immunity against COVID-19 could facilitate applying for and starting a new job in a safer way. The prospected scenario does not take into account potential market labour flows (especially firings)which could occur when public policies issued to increase employment protection, among which firing freeze,are lifted. The 2021 Italian Budget Law (Law n. 178/2020) has extended the firing freeze until the endof March 2021 together with the extension of wage supplementation schemes against the COVID-19 crisis.Further measures are under discussion. As from the same document, since March 2020 the Italian government has committed around 100.3billion euros in accordance with three fiscal packages as from Law Decree no. 18 from 17 March, Law Decreeno. 34 from 19 May, Law Decree no. 104 from 14 August, including, among the others, measures to supportfirms and employment. Law n. 178/2020. Cassa integrazione guadagni in deroga. Fondo di Integrazione Salariale, FIS.
21f COVID-19 infections, proxied by the number of excess deaths.Results highlight a negative and significant relationship between mobility changes andthe amount of the Wage Guarantee Fund (in full time equivalent units) over the periodMarch-August 2020. Moreover, we find that a 1% contraction in mobility (w.r.t. thebaseline) explains a 5% growth in the amount of the Wage Guarantee Fund (FTE units)allowed in the following month. Looking at the interpretation of a full time equivalentunit, a drop in human mobility explains an increase in the number of full-time workingemployees enrolled in wage guarantee schemes in the following month. The associationbecomes stronger if restricting the analysis when the first national lockdown was in force(March to May 2020), then gets milder and less significant after mobility restrictions areloosened (June to August 2021).Under the public health point of view, results show the existence of a positive andsignificant association between ’one month-lagged’ mobility changes and the excess deathsrecorded: a one percent drop in mobility (w.r.t. the baseline) explains a 0.5 percent dropin the number of excess deaths in the following month.Our finding are in agreement with the literature, as a positive association betweenmobility changes and deaths has already been observed by Glaeser et al. (2020), amongothers. In addition, Borri et al. (2020) highlighted a significant reduction in excess deaths(especially with respect to older people) in those municipalities experiencing more restrictinglockdown measures, then, the authors put in evidence how municipalities with a higher dropin the share of active people following business shutdowns are those showing a strongercontraction in mobility. Similarly, we notice that lower shares of essential working residentsin a administrative region (province) are associated to a higher mobility contraction andthat provinces with a higher centrality are those experiencing a higher drop in mobilityflows. In line with this evidence, Krenz and Strulik (2020) detected a higher decline ofmobility flows in areas which are less remote (lower travel time to the next urban center).The empirical evidence points out an association between the share of people employedin essential industries and excess deaths going through human mobility flows. These re-sults provides support for the inclusion of workers in essential sectors among the prioritycategories.Concerning the last stage of the vaccines delivery plan, we propose a prioritizationcriterion addressing the beneficiaries of furlough schemes and we test it against a benchmarkbased on the resident working force. We notice that while in some cases the two criteriaare substantially equivalent, in other cases the choice of criterion is detrimental and leadsto the significant disadvantages. From the economic point of view, we suggest that acriterion based on public expenditure review yields a co-benefit due to the substitutionpotential of the funds saved by the WGF reduction, that could be effectively invested bothin strengthening the sanitary system and in supporting the national economy.Although we analyze the Italian case, our results are relevant for an international audi-ence as well, since similar employment protection measures have been issued by Europeangovernments as a response to the pandemic. Short-time furlough schemes meant to supportthe firms affected by the crisis have been introduced or extended in Europe (Commission,22021). European Union member states are allowed to ask for European funds in order tocover such employment protection measures: financial support in the form of loans grantedon favourable terms is provided under the SURE instrument (temporary Support to mitigateUnemployment Risks in an Emergency) .As further data covering the period of the second epidemic wave and the effects ofthe COVID-19 crisis are expected to be released, future work will be devoted to a bettercharacterisation of the models developed and to a further refinement of the prioritisationindex. Acknowledgements
AF is supported by SoBigData++ (EC grant n. 871042). AF VP and MR thank FrancescaSantucci for help in comparing Facebook with Census data. AF VP and MR are gratefulto Francesco Serti for valuable comments and suggestions.
A Facebook and census data
In this section we test how Facebook data represent a reliable approximation of the popula-tion commuting for work and study, i.e. excluding those that do not habitually move withinthe provinces and between the provinces. Our idea is to test this reliability by looking atthe data provided by the Italian statistical office in 2011, i.e the last census data collectionin Italy.Both data are presented in the form of a origin destination matrix, where the linksexpress the number of commuters who travel between and within provinces. In the case ofISTAT data, people move for study or work, in all time slots, and data are averaged overa period of one year. On the other hand, Facebook data account for the people movingdaily, sampled at 8 hours intervals. In order to compare data, we averaged over the wholeavailable period the Facebook data obtaining an averaged origin destination matrix. Ourhypothesis is that the temporal averaging leads to a origin destination matrix accountingfor the habitual commuting of the study and workforce. Italy is among those member states which will benefit the most from the allocation of the resourcesprovided under the SURE: around 27.4 billions out of the total amount of 90.3 billion euros approvedby the European Council are gradually provided to Italy. Other member states which have been allowedto receive a financial support under the SURE are Belgium (7.8 billions), Spain (21.3 billions), Poland(11.2 billions), Portugal (5.9 billions), Greece (2.7 billions), Romania (4.1 billions). For further informationsee https://ec.europa.eu/info/business-economy-euro/economic-and-fiscal-policy-coordination/financial-assistance-eu/funding-mechanisms-and-facilities/sure_en R = 0 . Maps and plots
Figure 6: Province-specific fixed effects estimates: lockdown and post-lockdown periods (a) Excess Deaths: lockdown (b) Wage Guarantee Fund: lockdown(c) Excess Deaths: post-lockdown (d) Wage Guarantee Fund: post-lockdownNotes: the map shows a graphical representation of the individual fixed effects estimated through equation ln ( y ) it = β + β Mob.Range i ( t − + (cid:80) Nj =2 pv j d j,it + ε it . Estimates from the lockdown and post-lockdownperiods are represented. (a) January (b) February(c) March (d) April e) May (f) June(g) July (h) August eferences D. Acemoglu, V. Chernozhukov, I. Werning, and M. D. Whinston. A multi-risk sir modelwith optimally targeted lockdown. Technical report, National Bureau of Economic Re-search, 2020. URL .G. Bonaccorsi, F. Pierri, M. Cinelli, A. Flori, A. Galeazzi, F. Porcelli, A. L. Schmidt, C. M.Valensise, A. Scala, W. Quattrociocchi, and F. Pammolli. Economic and social conse-quences of human mobility restrictions under COVID-19.
Proceedings of the NationalAcademy of Sciences of the United States of America , 117(27):15530–15535, 2020. ISSN10916490. doi: 10.1073/pnas.2007658117.N. Borri, F. Drago, C. Santantonio, and F. Sobbrio. The ’Great Lockdown’: Inactive Work-ers and Mortality by COVID-19. 2020. URL https://cepr.org/active/publications/discussion_papers/dp.php?dpno=15317 .M. Borsati, S. Nocera, and M. Percoco. Questioning the spatial association between thespread of Covid-19 and transit usage in Italy. 2020.P. Buonanno, S. Galletta, and M. Puca. Estimating the severity of COVID-19: Evidencefrom the Italian epicenter.
PLoS ONE , 15(10 October):1–13, 2020. doi: 10.1371/journal.pone.0239569. URL http://dx.doi.org/10.1371/journal.pone.0239569 .A. Casarico and S. Lattanzio. The heterogeneous effects of COVID-19 on labor marketflows: Evidence from administrative data. 2020.R. Chetty, J.N. Friedman, N. Hendren, and M. Stepner. How did COVID-19 and stabiliza-tion policies affect spending and employment? A new real-time economic tracker basedon private sector data. 2020. URL .European Commission. Policy measures taken against the spread and impact of the coro-navirus - January 2021. Technical report, 2021. URL https://ec.europa.eu/info/sites/info/files/coronovirus_policy_measures_14_january_2021.pdf .Ministero della Sanit`a. Piano strategico: Elementi di preparazione e di implementazionedella strategia vaccinale. Technical report, 2020. URL .M. Farboodi, G. Jarosch, and R. J. Shimer. Internal and External Effects of Social Dis-tancing in a Pandemic. 2020. URL .C. A. Favero, A. Ichino, and A. Rustichini. Restarting the Economy While Saving LivesUnder COVID-19. 2020. URL http://dx.doi.org/10.2139/ssrn.3580626 .E. L. Glaeser, C. Gorback, and S. J. Redding. JUE insight: How much does COVID-19increase with mobility? Evidence from New York and four other U.S. cities.
Journal ofUrban Economics , (July):103292, 2020. ISSN 00941190. doi: 10.1016/j.jue.2020.103292.29. Goolsbee and C. Syverson. Fear, lockdown, and diversion: Comparing drivers of pan-demic economic decline 2020.
Journal of Public Economics , 193:104311, 2021. ISSN00472727. doi: 10.1016/j.jpubeco.2020.104311.INPS. Opendata portal, 2020. URL .INPS and Banca d’Italia. Le imprese e i lavoratori in cassa integrazione COVID neimesi di marzo e aprile, 2020. URL .ISTAT. Decessi e cause di morte, 2020. URL .A. Krenz and H. Strulik. The Benefits of Remoteness - Digital Mobility Data, RegionalRoad Infrastructure, and COVID-19 Infection. 2020. URL .P. Maas, A. Gros, L. McGorman, P. Alex Dow, S. Iyer, W. Park, and C. Nayak. Facebookdisaster maps: Aggregate insights for crisis response & recovery. In
Proceedings of theInternational ISCRAM Conference , volume 2019-May, pages 836–847, 2019.J. Medlock and A. P. Galvani. Optimizing influenza vaccine distribution.
Science , 325:1705,2009.M.E.J. Newman.
Networks: An introduction . Oxford University Press, 2010. ISBN 978-0199206650.P. Sah, J. Medlock, M. C. Fitzpatrick, B. H. Singer, and A.P. Galvani. Optimizing theimpact of low-efficacy influenza vaccines.
Proc. Natl. Acad. Sci. , 115:5151, 2018.E. Viviano. Alcune stime preliminari degli effetti delle misure di sostegno sul mercato dellavoro, 2020. URL .M. S. Warren and S. W. Skillman. Mobility Changes in Response to COVID-19. 2020. URL https://arxiv.org/abs/2003.14228 .WHO. Guidelines on the use of vaccines and antivirals during influenza pandemics.Technical report, 2004. URL .World Bank. Global Economic Prospects, January 2021, 2021. URL .S. Yoo and S. Managi. Global mortality benefits of covid-19 action.