What effects did the principles of supply and demand have on business and industries during this time period?

Oxford Review of Economic Policy. 2020 Aug 29 : graa033.

Abstract

We provide quantitative predictions of first-order supply and demand shocks for the US economy associated with the COVID-19 pandemic at the level of individual occupations and industries. To analyse the supply shock, we classify industries as essential or non-essential and construct a Remote Labour Index, which measures the ability of different occupations to work from home. Demand shocks are based on a study of the likely effect of a severe influenza epidemic developed by the US Congressional Budget Office. Compared to the pre-COVID period, these shocks would threaten around 20 per cent of the US economy’s GDP, jeopardize 23 per cent of jobs, and reduce total wage income by 16 per cent. At the industry level, sectors such as transport are likely to be output-constrained by demand shocks, while sectors relating to manufacturing, mining, and services are more likely to be constrained by supply shocks. Entertainment, restaurants, and tourism face large supply and demand shocks. At the occupation level, we show that high-wage occupations are relatively immune from adverse supply- and demand-side shocks, while low-wage occupations are much more vulnerable. We should emphasize that our results are only first-order shocks—we expect them to be substantially amplified by feedback effects in the production network.

Keywords: COVID-19, shocks, economic growth, unemployment

I. Introduction

The COVID-19 pandemic is having an unprecedented impact on societies around the world.1 As governments mandate social distancing practices and instruct non-essential businesses to close to slow the spread of the outbreak, there is significant uncertainty about the effect such measures will have on lives and livelihoods. While demand for specific sectors such as grocery stores increased in the early weeks of the pandemic, other sectors such as air transportation and tourism have seen demand for their services evaporate. At the same time, many sectors are experiencing issues on the supply side, as governments curtail the activities of non-essential industries and workers are confined to their homes.

In this paper, we aim to provide analytical clarity about the supply and demand shocks caused by public health measures and changes in preferences caused by avoidance of infection. We estimate (i) supply-side reductions due to the closure of non-essential industries and workers not being able to perform their activities at home, and (ii) demand-side changes due to peoples’ immediate response to the pandemic, such as reduced demand for goods or services that are likely to place people at risk of infection (e.g. tourism).

Several researchers have already provided estimates of the supply shock from labour supply (Dingel and Neiman, 2020; Hicks et al., 2020; Koren and Petö, 2020). Here we improve on these efforts in three ways: (i) we propose a methodology for estimating how much work can be done from home, based on work activities, (ii) we identify industries for which working from home is irrelevant because the industries are considered essential, and (iii) we compare our estimated supply shocks to estimates of the demand shock, which in many industries is the more relevant constraint on output.

A number of papers have also emphasized the importance of demand-side factors in the pandemic. For example, an early paper by Guerrieri et al. (2020) showed that, in a two-sector new Keynesian model with low substitutability in consumption, asymmetric labour supply shocks can lead to reductions in demand that are higher than the initial shock. In their framework, sectoral heterogeneity is necessary for supply shocks to lead to a larger demand impact. The scenario of a drop in demand larger than the drop in supply is also plausible if long-term labour supply constraints lead to a collapse of investment. This discussion underscores that supply and demand interact, and supply shocks can lead to decreases in demand. But as we argue here, the COVID pandemic also created exogenous and instantaneous changes to consumer demand, both in magnitude and in composition as consumers’ preferences change in response to factors such as infection risk, lower positive externalities in social consumption, explicit guidelines from the government, etc.

To see why it is important to compare supply and demand shocks, consider the following thought experiment. Following social-distancing measures, suppose industry i is capable of producing only 70 per cent of its pre-crisis output, e.g. because workers can produce only 70 per cent of the output while working from home. If consumers reduce their demand by 90 per cent, the industry will produce only what will be bought, that is, 10 per cent. If instead consumers reduce their demand by 20 per cent, the industry will not be able to satisfy demand but will produce everything it can, that is, 70 per cent. In other words, the experienced first-order reduction in output from the immediate shock will be the greater of the supply and the demand shock.

It is important to stress that the shocks that we predict here should not be interpreted as the overall impact of the COVID-19 pandemic on the economy. Again, we expect that as wages from work drop, there will be potentially larger second-order negative impacts on demand, and the potential for a self-reinforcing downward spiral in output, employment, income, and demand. Deriving overall impact estimates involves modelling second-order effects, such as the additional reductions in demand as workers who are stood down or laid off experience a reduction in income, and additional reductions in supply as potential shortages propagate through supply chains. Further effects, such as cascading firm defaults, which can trigger bank failures and systemic risk in the financial system, could also arise. Understanding these impacts requires a model of the macro-economy and financial sector. In a companion paper (Pichler et al., 2020), we present results from such an economic model, but we make our estimates of first-order impacts available separately here, for other researchers or governments to build upon or use in their own models.

Overall, we find that the supply and demand shocks considered in this paper represent a reduction of around one-fifth of the US economy’s value added, one-quarter of current employment, and about 16 per cent of the US total wage income.2 Supply shocks account for the majority of this reduction. These effects vary substantially across different industries. While we find no negative effects on value added for industries like Legal Services, Power Generation and Distribution, or Scientific Research, the expected loss of value added reaches up to 80 per cent for Accommodation, Food Services, and Independent Artists.

We find that sectors such as Transport are likely to experience immediate demand-side reductions that are larger than their corresponding supply-side shocks. Other industries such as Manufacturing, Mining, and certain service sectors are likely to experience larger immediate supply-side shocks relative to demand-side shocks. Entertainment, restaurants, and hotels experience very large supply and demand shocks, with the demand shock dominating. These results are important because supply and demand shocks might have different degrees of persistence, and industries will react differently to policies depending on the constraints that they face. Overall, however, we find that aggregate effects are dominated by supply shocks, with a large part of manufacturing and services being classified as non-essential while its labour force is unable to work from home.

We also break down our results by occupation and show that there is a strong negative relationship between the overall immediate shock experienced by an occupation and its wage. Relative to the pre-COVID period, 41 per cent of the jobs for workers in the bottom quartile of the wage distribution are predicted to be vulnerable. (And bear in mind that this is only a first-order shock—second-order shocks may significantly increase this.) In contrast, most high-wage occupations are relatively immune from adverse shocks, with only 6 per cent of the jobs at risk for the 25 per cent of workers working in the highest pay occupations. Absent strong support from governments, most of the economic burden of the pandemic will fall on lower wage workers.

We neglect several effects that, while important, are small compared to those we consider here. First, we have not sought to quantify the reduction in labour supply due to workers contracting COVID-19. A rough estimate suggests that this effect is relatively minor in comparison to the shocks associated with social-distancing measures that are being taken in most developed countries.3 We have also not explicitly included the effect of school closures. However, in Appendix D2. we argue that this is not the largest effect and is already partially included in our estimates through indirect channels.

A more serious problem is caused by the need to assume that within a given occupation, being unable to perform some work activities does not harm the performance of other work activities. Within an industry, we also assume that if workers in a given occupation cannot work, they do not produce output, but this does not prevent other workers in different occupations from producing. In both cases we assume that the effects of labour on production are linear, i.e. that production is proportional to the fraction of workers who can work. In reality, however, it is clear that there are important complementarities leading to nonlinear effects. There are many situations where production requires a combination of different occupations, such that if workers in key occupations cannot work at home, production is not possible. For example, while the accountants in a steel plant might be able to work from home, if the steelworkers needed to run the plant cannot come to work, no steel is made. We cannot avoid making linear assumptions because, as far as we know, there is no detailed understanding of the labour production function and these interdependencies at an industry level. By neglecting nonlinear effects, our work here should consequently be regarded as an approximate lower bound on the size of the first-order shocks.

This paper focuses on the United States. We have chosen it as our initial test case because input–output tables are more disaggregated than those of most other countries, and because the Occupational Information Network (O*NET) database, which we rely on for information about occupations, was developed based on US data. With some additional assumptions it is possible to apply the analysis we perform here to other developed countries.

This paper is structured as follows. In section II we review the most relevant literature on the economic impact of the pandemic and the associated supply and demand shocks. In section III we describe our methodology for estimating supply shocks, which involves developing a new Remote Labour Index (RLI) for occupations and combining it with a list of essential industries. Section IV discusses likely demand shocks based on estimates developed by the US Congressional Budget Office (2006) to predict the potential economic effects of an influenza pandemic. In section V, we show a comparison of the supply and demand shocks across different industries and occupations and identify the extent to which different activities are likely to be constrained by supply or demand. In this section, we also explore which occupations are more exposed to infection and make comparisons to wage and occupation-specific shocks. Finally, in section VI we discuss our findings in light of existing research and outline avenues for future work. We also make all of our data available in a continuously updated online repository (https://zenodo.org/record/3744959).

II. Literature

Many economists and commentators believe that the economic impact could be dramatic (Baldwin and Weder di Mauro, 2020). To give an example based on survey data in an economy under lockdown, the French statistical office estimated on 26 March 2020 that the economy is currently at around 65 per cent of its normal level.4Bullard (2020) provides an undocumented estimate that around a half of the US economy would be considered either essential, or able to operate without creating risks of diffusing the virus. Inoue and Todo (2020) modelled how shutting down firms in Tokyo would cause a loss of output in other parts of the economy through supply chain linkages, and estimate that after a month, daily output would be 86 per cent lower than pre-shock (i.e. the economy would be operating at only 14 per cent of its capacity!). Using a calibrated extended consumption function, and assuming a labour income shock of 16 per cent and various consumption shocks by expenditure categories, Muellbauer (2020) estimates a fall of quarterly consumption of 20 per cent. Roughly speaking, most of these estimates, like ours, are estimates of instantaneous declines, and would translate to losses of annual GDP if the lockdown lasted for a year.

Based on aggregating industry-level shocks, the OECD (2020) estimates a drop in immediate GDP of around 25 per cent. Another study by Barrot et al. (2020) estimates industry-level shocks by considering the list of essential industries, the closure of schools, and an estimate of the ability to work from home (based on ICT use surveys). Using these shocks in a multisector input–output model, they find that 6 weeks of social distancing would bring annual GDP down by 5.6 per cent.

Our study predicts supply and demand shocks at a disaggregated level, and proposes a simple method to calculate aggregate shocks from these. We take a short-term approach, and assume that the immediate drop in output is driven by the most binding constraint—the worse of the supply and demand shock, essentially assuming that prices do not adjust and markets do not clear. An alternative, standard in empirical macroeconomics, is to observe aggregate changes in prices and quantities to infer the relative size of the supply and demand shocks. For instance, Brinca et al. (2020) use data on wage and hours worked; Balleer et al. (2020) use data on planned price changes in German firms; and Bekaert et al. (2020) use surveys of inflation forecasts. While these studies do not agree on the relative importance of supply and demand shocks, an emerging consensus is that both supply and demand shocks co-exist and are vastly different across sectors, and over time.

III. Supply shocks

Supply shocks from pandemics are mostly thought of as labour supply shocks. Several pre-COVID-19 studies focused on the direct loss of labour from death and sickness (e.g. McKibbin and Sidorenko (2006), Santos et al. (2013)), although some have also noted the potentially large impact of school closure (Keogh-Brown et al., 2010). McKibbin and Fernando (2020) consider (among other shocks) reduced labour supply due to mortality, morbidity due to infection, and morbidity due to the need to care for affected family members. In countries where social distancing measures are in place, such measures will have a much larger economic effect than the direct effects from mortality and morbidity. This is in part because if social distancing measures work, only a small share of population will be infected and die eventually. Appendix D1. provides more quantitative estimates of the direct mortality and morbidity effects and argues that they are likely to be at least an order of magnitude smaller than those due to social-distancing measures, especially if the pandemic is contained.

For convenience we neglect mortality and morbidity and assume that the supply shocks are determined only by the amount of labour that is withdrawn due to social distancing. We consider two key factors: (i) the extent to which workers in given occupations can perform their requisite activities at home, and (ii) the extent to which workers are likely to be unable to come to work due to being in non-essential industries. We quantify these effects on both industries and occupations. Figure 1 gives a schematic overview of how we predict industry and occupation specific supply shocks. We explain this in qualitative terms in the next few pages; for a formal mathematical description see Appendix A.1.

A schematic network representation of supply-side shocksNotes: The nodes to the left represent the list of essential industries at the NAICS 6-digit level. A green node indicates essential, a red node non-essential. The orange nodes (centre-left) are more aggregate industry categories (e.g. 4-dig. NAICS or the BLS industry categories) for which further economic data are available. These two sets of nodes are connected through industry concordance tables. The blue nodes (centre-right) are different occupations. A weighted link connecting an industry category with an occupation represents the number of people of a given occupation employed in each industry. Nodes on the very right are O*NET work activities. Green work activities mean that they can be performed from home, while red means that they cannot. O*NET provides a mapping of work activities to occupations.

(i) How much work can be performed from home?

One way to assess the degree to which workers are able to work from home during the COVID-19 pandemic is by direct survey. For example, Zhang et al. (2020) conducted a survey of Chinese citizens in late February (1 month into the coronavirus-induced lockdown in China) and found that 27 per cent of the labour force continued working at the office, 38 per cent worked from home, and 25 per cent stopped working. Adams-Prassl et al. (2020) surveyed US and UK citizens in late March, and reported that the share of tasks that can be performed from home varies widely between occupations (from around 20 to 70 per cent), and that higher wage occupations tend to be more able to work from home.

Other recent work has instead drawn on occupation-level data from O*NET to determine labour shocks due to the COVID-19 pandemic. For example, Hicks et al. (2020) drew on O*NET’s occupational Work Context Questionnaire and considered the degree to which an occupation is required to ‘work with others’ or involves ‘physical proximity to others’ in order to assess which occupations are likely to be most impacted by social distancing. Dingel and Neiman (2020) aimed to quantify the number of jobs that could be performed at home by analysing responses on O*NET’s Work Context Questionnaire (such as whether the average respondent for an occupation spends the majority of time walking or running or uses email less than once per month) as well as responses on O*NET’s Generalized Work Activities Questionnaire (such as whether performing general physical activities or handling and moving objects is very important for a given occupation).

In this study, we go to a more granular level than both the Work Context Questionnaire and Generalized Work Activities Questionnaire, and instead draw on O*NET’s ‘intermediate work activity’ data, which provide a list of the activities performed by each occupation based on a list of 332 possible work activities. For example, a nurse undertakes activities such as ‘maintain health or medical records’, ‘develop patient or client care or treatment plans’, and ‘operate medical equipment’, while a computer programmer performs activities such as ‘resolve computer programs’, ‘program computer systems or production equipment’, and ‘document technical designs, producers or activities’.5 In Figure 1 these work activities are illustrated by the rightmost set of nodes.

Which work activities can be performed from home?

Four of us independently assigned a subjective binary rating to each work activity as to whether it could successfully be performed at home. The individual results were in broad agreement. Based on the responses, we assigned an overall consensus rating to each work activity.6 Ratings for each work activity are available in an online data repository.7 While O*NET maps each intermediate work activity to 6-digit O*NET occupation codes, employment information from the US Bureau of Labor Statistics (BLS) is available for the 4-digit 2010 Standard Occupation Scheme (SOC) codes, so we mapped O*NET and SOC codes using a crosswalk available from O*NET.8 Our final sample contains 740 occupations.

From work activities to occupations.

We then created a Remote Labour Index (RLI) for each occupation by calculating the proportion of an occupation’s work activities that can be performed at home. An RLI of 1 would indicate that all of the activities associated with an occupation could be undertaken at home, while an RLI of 0 would indicate that none of the occupation’s activities could be performed at home.9 The resulting ranking of each of the 740 occupations can be found in the online repository (see footnote 6). In contrast to previous work that has tended to arrive at binary assessments of whether an occupation can be performed at home, our approach has the advantage of providing a unique indication of the amount of work performed by a given occupation that can be done remotely. While the results are not perfect,10 most of the rankings make sense. For example, in Table 1, we show the top 20 occupations having the highest RLI ranking. Some occupations such as credit analysts, tax preparers, and mathematical technician occupations are estimated to be able to perform 100 per cent of their work activities from home. Table 1 also shows a sample of the 43 occupations with an RLI ranking of zero, i.e. those for which there are no activities that can be performed at home.

Table 1:

Top and bottom 20 occupations ranked by Remote Labour Index (RLI), based on proportion of work activities that can be performed at home

To provide a broader perspective of how the RLI differs across occupation categories, Figure 2 shows a series of box-plots indicating the distribution of RLI for each 4-digit occupation in each 2-digit SOC occupation category. We have ordered 2-digit SOC occupations in accordance with their median values. Occupations with the highest RLI relate to Education, Training and Library, Computer and Mathematical, and Business and Financial roles, while occupations relating to Production, Farming, Fishing and Forestry, and Construction and Extraction tend to have lower RLI.

Distribution of RLI across occupationsNote: We provide boxplots showing distribution of RLI for each 4-digit occupation in each 2-digit SOC occupation category.

From occupations to industries.

We next map the RLI to industry categories to quantify industry-specific supply shocks from social distancing measures. We obtain occupational compositions per industry from the BLS, which allows us to match 740 occupations to 277 industries.11

In Figure 3, we show the RLI distribution for each 4-digit occupation category falling within each broad 2-digit NAICS category. Similar to Figure 2, we have ordered the 2-digit NAICS industry categories in accordance with the median values of each underpinning distribution. As there is a greater variety of different types of occupations within these broader industry categories, distributions tend to be much wider. Industries with the highest median RLI values relate to Information, Finance and Insurance, and Professional, Science and Technical Services, while industries with the lowest median RLI relate to Agriculture, Forestry, Fishing and Hunting and Accommodation and Food Services.

Distribution of RLI across industriesNote: We provide boxplots showing distribution of RLI for each 4-digit occupation in each 2-digit NAICS Industry category.

In Appendix B, we show industry-specific RLI values for the more detailed 4-digit NAICS industries. To arrive at a single number for each 4-digit industry, we compute the employment-weighted average of occupation-specific RLIs. The resulting industry-specific RLI can be interpreted as a rough estimate of the fraction of jobs which can be performed from home for each industry.

(ii) Which industries are ‘essential’?

Across the world, many governments have mandated that certain industries deemed ‘essential’ should remain open over the COVID-19 crisis duration. What constitutes an ‘essential’ industry has been the subject of significant debate, and it is likely that the endorsed set of essential industries will vary across countries. As the US government has not produced a definitive list, here we draw on the list of essential industries developed by Italy and assume it can be applied, at least as an approximation, to other countries, such as the US, as well. This list has two key advantages. First, as Italy was one of the countries affected earliest and most severely, it was one of the first countries to invest significant effort considering which industries should be deemed essential. Second, Italy’s list of essential industries includes NACE industrial classification codes, which can be mapped to the NAICS industry classification we use to classify industrial employment in this paper.12

Table 2 shows the total numbers of NAICS essential industries at the 6-digit and 4-digit level. More than 50 per cent of 6-digit NAICS industries are considered essential. At the 6-digit level the industries are either classified as essential, and assigned essential score 1, or non-essential and assigned essential score 0. Unfortunately, it is not possible to translate this directly into a labour force proportion as BLS employment data at detailed occupation and industry levels are only available at the NAICS 4-digit level. To derive an estimate at the 4-digit level, we assume that labour in a NAICS 4-digit code is uniformly distributed over its associated 6-digit codes. We then assign an essential ‘share’ to each 4-digit NAICS industry based on the proportion of its 6-digit NAICS industries that are considered essential. (The distribution of the essential share over 4-digit NAICS industries is shown in Appendix B.) Based on this analysis, we estimate that about 92m (or 67 per cent) of US workers are currently employed in essential industries.

Table 2:

Essential industries

(iii) Supply shock: non-essential industries unable to work from home

Having analysed both the extent to which jobs in each industry are essential and the likelihood that workers in a given occupation can perform their requisite activities at home, we now combine these to consider the overall first-order effect on labour supply in the US. In Figure 4, we plot the RLI of each occupation against the fraction of that occupation employed in an essential industry. Each circle in the scatter plot represents an occupation; the circles are sized proportional to current employment and colour coded according to the median wage in each occupation.

Fraction employed in an essential industry vs Remote Labour Index for each occupationNotes: Omitting the effect of demand reduction, the occupations in the lower left corner, with a small proportion of workers in essential industries and a low Remote Labour Index, are the most vulnerable to loss of employment due to social distancing.

Figure 4 indicates the vulnerability of occupations due to supply-side shocks. Occupations in the lower left-hand side of the plot (such as Dishwashers, Rock Splitters, and Logging Equipment Operators) have lower RLI scores (indicating they are less able to work from home) and are less likely to be employed in an essential industry. If we consider only the immediate supply-side effects of social distancing, workers in these occupations are more likely to face reduced work hours or be at risk of losing their jobs altogether. In contrast, occupations on the upper right-hand side of the plot (such as Credit Analysis, Political Scientists, and Operations Research Analysts) have higher RLI scores and are more likely to be employed in an essential industry. These occupations are less economically vulnerable to the supply-side shocks (though, as we discuss in the next section, they could still face employment risks due to first-order demand-side effects). Occupations in the upper-left hand side of the plot (such as Farmworkers, Healthcare Support Workers, and Respiratory Therapists) are less likely to be able to perform their job at home, but since they are more likely to be employed in an essential industry their economic vulnerability from supply-side shocks is lower. Interestingly, there are relatively few occupations on the lower-right hand side of the plot. This indicates that occupations that are predominantly employed in non-essential industries tend to be less able to perform their activities at home.

To help visualize the aggregate numbers we provide a summary in the form of a Venn diagram in Figure 5. Before the pandemic, 33 per cent of workers were employed in non-essential jobs. 56 per cent of workers are estimated to be unable to perform their job remotely. 19 per cent of workers are in the intersection corresponding to non-essential jobs that cannot be performed remotely. In addition, there are 30 per cent of workers in essential industries that can also work from home.13

Workers that cannot workNotes: On the left is the percentage of workers in a non-essential job (33 per cent in total). On the right is the percentage of workers that cannot work remotely (56 per cent in total). The intersection is the set of workers that cannot work, which is 19 per cent of all workers. A remaining 30 per cent of workers are in essential jobs where they can work remotely.

IV. Demand shock

The pre-COVID-19 literature on epidemics and the discussions of the current crisis make it clear that epidemics strongly influence patterns of consumer spending. Consumers are likely to seek to reduce their risk of exposure to the virus and decrease demand for products and services that involve close contact with others. In the early days of the outbreak, stockpiling behaviour also drives a direct demand increase in the retail sector (Baker et al., 2020).

Estimates from the CBO

Our estimates of the demand shock are based on expert estimates developed by the US Congressional Budget Office (2006) that attempted to predict the potential impact of an influenza pandemic. Similar to the current COVID-19 pandemic, this analysis assumes that demand is reduced due to the desire to avoid infection. While the analysis is highly relevant to the present COVID-19 situation, it is important to note that the estimates are ‘extremely rough’ and ‘based loosely on Hong Kong’s experience with SARS’. The CBO provides estimates for two scenarios (mild and severe). We draw on the severe scenario, which

describes a pandemic that is similar to the 1918–1919 Spanish flu outbreak. It incorporates the assumption that a particularly virulent strain of influenza infects roughly 90 million people in the United States and kills more than 2 million of them.

In this paper, we simply take the CBO estimates as immediate (first-order) demand-side shocks. The CBO lists demand-side estimates for broad industry categories, which we mapped to the 2-digit NAICS codes by hand. Table 3 shows the CBO’s estimates of the percent decrease in demand by industry, and Table 8 in Appendix E shows the full mapping to 2-digit NAICS.

Table 3:

These estimates, of course, are far from perfect. They are based on expert estimates made more than 10 years ago for a hypothetical pandemic scenario. It is not entirely clear if they are for gross output or for final (consumer) demand. However, in Appendix E, we describe three other sources of consumption shocks (Keogh-Brown et al., 2010; Muellbauer, 2020; OECD, 2020) that provide broadly similar estimates by industry or spending category. We also review papers that have appeared more recently and contained estimates of consumption changes based on transaction data. Taken together, these papers suggest that the shocks from the CBO were qualitatively accurate: very large declines in the hospitality, entertainment, and transport industries, milder declines in manufacturing, and a more resistant business services sector. The main features that have been missed are the increase in demand, at least early on, in some specific retail categories (groceries), and the decline in health consumption, in sharp contrast with the CBO prediction of a 15 per cent increase.

Aggregate consumption vs composition of the shocks

The shocks from the CBO include two separate effects: a shift of preferences, where the utility of healthcare relative to restaurants, say, increases; and an aggregate consumption effect. Here, we do not go further in distinguishing these effects, although this becomes necessary in a more fully-fledged model (Pichler et al., 2020). Yet, it remains instructive to note that, in Muellbauer’s (2020) consumption function estimates, the decline in aggregate consumption is not only due to direct changes in consumption in specific sectors, but also to lower income, rising income insecurity (due to unemployment in particular), and wealth effects (due in particular to falling asset prices).

Transitory and permanent shocks

An important question is whether demand reductions are just postponed expenses, and if they are permanent (Keogh-Brown et al., 2010; Mann, 2020). Baldwin and Weder di Mauro (2020) also distinguish between ‘practical’ (the impossibility to shop) and ‘psychological’ (the wait-and-see attitude adopted by consumers facing strong uncertainty demand shocks). We see three possibilities: (i) expenses in a specific good or service are just delayed but will take place later, for instance if households do not go to the restaurant this quarter, but go twice as often as they would normally during the next quarter; (ii) expenses are not incurred this quarter, but will come back to their normal level after the crisis, meaning that restaurants will have a one-quarter loss of sales; and (iii) expenses decrease to a permanently lower level, as household change their preferences in view of the ‘new normal’. Appendix E reproduces the scenario adopted by Keogh-Brown et al. (2010), which distinguishes between delay and permanently lost expenses.

Other components of aggregate demand

We do not include direct shocks to investment, net exports, and net inventories. Investment is typically very pro-cyclical and is likely to be strongly affected, with direct factors including cash-flow reductions and high uncertainty (Boone, 2020). The impact on trade is likely to be strong and possibly permanent (Baldwin and Weder di Mauro, 2020), but would affect exports and imports in a relatively similar way, so the overall effect on net exports is unclear. Finally, it is likely that due to the disruption of supply chains, inventories will be run down so the change in inventories will be negative (Boone, 2020).

V. Combining supply and demand shocks

Having described both supply- and demand-side shocks, we now compare the two at the industry and occupation level.

(i) Industry-level supply and demand shocks

Figure 6 plots the demand shock against the supply shock for each industry. The radius of the circles is proportional to the gross output of the industry.14 Essential industries have no supply shock and so lie on the horizontal ‘0’ line. Of these industries, sectors such as Utilities and Government experience no demand shock either, since immediate demand for their output is assumed to remain the same. Following the CBO predictions, Health experiences an increase in demand and consequently lies below the identity line. Transport, on the other hand, experiences a reduction in demand and lies well above the identity line. This reflects the current situation, where trains and buses are running because they are deemed essential, but they are mostly empty. Non-essential industries such as Entertainment, Restaurants, and Hotels, experience both a demand reduction (due to consumers seeking to avoid infection) and a supply reduction (as many workers are unable to perform their activities at home). Since the demand shock is bigger than the supply shock, they lie above the identity line. Other non-essential industries, such as Manufacturing, Mining, and Retail, have supply shocks that are larger than their demand shocks and consequently lie below the identity line.

Supply and demand shocks for industriesNotes: Each circle is an industry, with radius proportional to gross output. Many industries experience exactly the same shock, hence the superposition of some of the circles. Labels correspond to broad classifications of industries.

(ii) Occupation-level supply and demand shocks

In Figure 7 we show the supply and demand shocks for occupations rather than industries. For each occupation this comparison indicates whether it faces a risk of unemployment due the lack of demand or a lack of supply in its industry.

Supply and demand shocks for occupationsNotes: Each circle is an occupation with radius proportional to employment. Circles are colour coded by the log median wage of the occupation. The correlation between median wages and demand shocks is 0.26 (p-value = 2.8 × 10–13) and between median wages and supply shocks is 0.41 (p-value = 1.5 × 10–30).

Several health-related occupations, such as Nurses, Medical Equipment Preparers, and Healthcare Social Workers, are employed in industries experiencing increased demand. Occupations such as Airline Pilots, Lodging Managers, and Hotel Desk Clerks face relatively mild supply shocks and strong demand shocks (as consumers reduce their demand for travel and hotel accommodation) and consequently lie above the identity line. Other occupations such as Stonemasons, Rock Splitters, Roofers, and Floor Layers face a much stronger supply shock as it is very difficult for these workers to perform their job at home. Finally, occupations such as Cooks, Dishwashers, and Waiters suffer both adverse demand shocks (since demand for restaurants is reduced) and supply shocks (since they cannot work from home and tend not to work in essential industries).

For the majority of occupations, the supply shock is larger than the demand shock. This is not surprising given that we only consider immediate shocks and no feedback-loops in the economy. We expect that once second-order effects are considered the demand shocks are likely to be much larger.

(iii) Aggregate shocks

We now aggregate shocks to obtain estimates for the whole economy. We assume that, in a given industry, the total shock will be the largest of the supply or demand shocks. For example, if an industry faces a 30 per cent demand shock and 50 per cent supply shock because 50 per cent of the industry’s workforce cannot work, the industry is assumed to experience an overall 50 per cent shock to output. For simplicity, we assume a linear relationship between output and labour: i.e. when industries are supply constrained, output is reduced by the same fraction as the reduction in labour supply. This assumption also implies that the demand shock that workers of an industry experience equals the industry’s output demand shock in percentage terms. For example, if transport faces a 67 per cent demand shock and no supply shock, bus drivers working in this industry will experience an overall 67 per cent employment shock. The shock on occupations depends on the prevalence of each occupation in each industry (see Appendix A for details). We then aggregate shocks in three different ways.

First, we estimate the decline in employment by weighting occupation-level shocks by the number of workers in each occupation. Second, we estimate the decline in total wages paid by weighting occupation-level shocks by the share of occupations in the total wage bill. Finally, we estimate the decline in GDP by weighting industry-level shocks by the share of industries in GDP.15

Table 4 shows the results. In all cases, by definition, the total shock is larger than both the supply and demand shock, but smaller than the sum. Overall, the supply shock appears to contribute more to the total shock than does the demand shock.

Table 4:

Aggregate shocks to employment, wages, and value added

The wage shock is around 16 per cent and is lower than the employment shock (23 per cent). This makes sense, and reflects a fact already well acknowledged in the literature (Office for National Statistics, 2020; Adams-Prassl et al., 2020) that occupations that are most affected tend to have lower wages. We discuss this more below.

For industries and occupations in the health sector, which experience an increase in demand in our predictions, there is no corresponding increase in supply. Table 6 in Appendix A.7 provides the same estimates as Table 4, but now assuming that the increased demand for health will be matched by increased supply. This corresponds to a scenario where the healthcare sector would be immediately able to hire as many workers as necessary and pay them at the normal rate. This assumption does not, however, make a significant difference to the aggregate total shock. In other words, the increase in activity in the health sector is unlikely to be large enough to compensate significantly for the losses from other sectors.

(iv) Shocks by wage level

To understand how the pandemic has affected workers of different income levels differently, we present results for each wage quartile. The results are in Table 5, columns q1... q4,16 where we show employment shocks by wage quartile. This table shows that workers whose wages are in the lowest quartile (lowest 25 per cent) will bear much higher relative losses than workers whose wages are in the highest quartile. Our results confirm the survey evidence reported by the Office for National Statistics (2020) and Adams-Prassl et al. (2020), showing that low-wage workers are more strongly affected by the COVID crisis in terms of lost employment and lost income. Furthermore, Table 5 shows how the total loss of wages in the economy is split amongst the different quartiles. Even though those in the lowest quartile have lower salary, the shock is so high that they bear the highest share of the total loss.

Table 5:

Total wages or employment shocks by wage quartile

Next we estimate labour shocks at the occupation level. We define the labour shocks as the declines in employment due to the total shocks in the industries associated with each occupation. We use Eq. (14) (Appendix A.7) to compute the labour shocks, which allows for positive shocks in healthcare workers, to suggest an interpretation in terms of a change in labour demand. Figure 8 plots the relationship between labour shocks and median wage. A strong positive correlation (Pearson ρ = 0.40, p-value = 3.5 × 10–30) is clearly evident, with almost no high-wage occupations facing a serious shock.

Labour shock vs median wage for different occupationsNotes: We colour occupations by their exposure to disease and infection. There is a 0.40 correlation between wages and the labour shock (p-value = 3.5 × 10–30). Note the striking lack of high-wage occupations with large labour demand shocks.

We have also coloured occupations by their exposure to disease and infection using an index developed by O*NET17 (for brevity we refer to this index as ‘exposure to infection’). As most occupations facing a positive labour shock relate to healthcare,18 it is not surprising to see that they have a much higher risk of being exposed to disease and infection. However, other occupations such as janitors, cleaners, maids, and childcare workers also face higher risk of infection. Appendix C explores the relationship between exposure to infection and wage in more detail.

VI. Conclusion

This paper has sought to provide quantitative predictions for the US economy of the supply and demand shocks associated with the COVID-19 pandemic. To characterize supply shocks, we developed a Remote Labour Index (RLI) to estimate the extent to which workers can perform activities associated with their occupation at home and identified which industries are classified as essential vs non-essential. We also reported plausible estimates of the demand shocks, in an attempt to acknowledge that some industries will have an immediate reduction in output due to a shortfall in demand, rather than due to an impossibility to work. We would like to emphasize that these are predictions, not measurements. The estimates of the demand shocks were made in 2006, and the RLI and the list of non-essential industries contain no pandemic-specific information, and could have been made at any time. Putting these predictions together, we estimate that the first-order aggregate shock to the economy represents a reduction of roughly a fifth of the economy.

This is the first study seeking to compare supply-side shocks with corresponding demand-side shocks at the occupation and industry level. At the time of writing (mid-April), the most relevant demand-side estimates available are admittedly highly ‘rough’ and only available for very aggregate (2-digit) industries. Yet, this suggests that sectors such as transport are more likely to have output constrained by demand-side shocks, while sectors relating to manufacturing, mining, and services are more likely to be constrained by supply-side shocks. Entertainment, restaurants, and tourism face both very large supply and demand constraints, with demand shocks dominating in our estimates. By quantifying supply and demand shocks by industry, our paper speaks to the debate on the possibility of inflation after the crisis. Goodhart and Pradhan (2020) argue that the lockdown causes a massive supply shock that will lead to inflation when demand comes back after the crisis. But as Miles and Scott (2020) note, in many sectors it is not obvious that demand will come back immediately after the crisis, and if a gradual reopening of the economy takes place, it may be that supply and demand rise slowly together. However, our paper is the first to raise the fact that because supply and demand shocks are so different by sectors, even a gradual reopening may leave important supply–demand imbalances within industries. Such mismatches could consequently lead to an unusual level of heterogeneity in the inflation for different goods.

When considering total shocks at the occupation level, we find that high-wage occupations are relatively immune from both supply- and demand-side shocks, while many low-wage occupations are much more economically vulnerable to both. Interestingly, low-wage occupations that are not vulnerable to supply- and/or demand-side shocks are nonetheless at higher risk of being exposed to coronavirus (see colour code in Figure 8). Such findings suggest that the COVID-19 pandemic is likely to exacerbate income inequality in what is already a highly unequal society.

For policy-makers there are three key implications from this study. First, the magnitude of the shocks being experienced by the US economy is very large, with around a fifth of the economy not functioning. As Table 4 shows, even including positive shocks, our estimates of the potential impacts are a drop in employment of 23 per cent, a decline in wages of 16 per cent, and loss in value added of 20 per cent. Bearing in mind the caveats about shocks vs total impacts, the potential impacts are a multiple of what was experienced during the Global Financial Crisis (e.g. where employment dropped 3.28 percentage points)19 and comparable only to the Great Depression (e.g. where employment dropped 21.7 per cent 1929–32 (Wallis, 1989, Table 2)). Second, as the largest shocks are from the supply side, strategies for returning people to work as quickly as possible without endangering public health must be a priority. Virus mitigation and containment are clearly essential first steps, but strategies such as widespread antibody testing to identify people who are safe to return to work, and rapid testing, tracing, and isolation to minimize future lock-downs, will also be vital until, if and when, a vaccine is available. Furthermore, aggressive fiscal and monetary policies to minimize first-order shocks cascading into second-order shocks are essential, in particular policies to keep workers in employment and maintain incomes (e.g. the ‘paycheck protection’ schemes announced by several countries), as well as policies to preserve business and financial solvency. Third, and finally, the inequalities highlighted by this study will also require policy responses. Again, higher-income knowledge and service workers will likely see relatively little impact, while lower-income workers will bear the brunt of the employment, income, and health impacts. In order to ensure that burdens from the crisis are shared as fairly as possible, assistance should be targeted at those most affected, while taxes to support such programmes should be drawn primarily from those least affected.

To reiterate an important point, our predictions of the shocks are not estimates of the overall impact of the COVID-19 on the economy, but are rather estimates of the first-order shocks. Overall impacts can be very different from first-order shocks for several reasons. First, shocks to a particular sector propagate and may be amplified as each industry faces a shock and reduces its demand for intermediate goods from other industries (Pichler et al., 2020). Second, industries with decreased output will stop paying wages of furloughed workers, thereby reducing income and demand; importantly, reduction in supply in capacity-constrained sectors lead to decreases in expenditures in other sectors, and the details of these imbalances will determine the overall impact (Guerrieri et al., 2020). Third, the few industries facing higher demand will increase supply, if they can overcome labour mobility frictions (del Rio-Chanona et al., 2019). Fourth, the final outcomes will very much depend on the policy response, and in particular the ability of government to maintain (consumption and investment) demand and limit the collapse of the labour market, in a context where the shocks are extremely heterogenous across industries, occupations, and income levels. We make our predictions of the shocks available here so that other researchers can improve upon them and use them in their own models.20 We intend to update and use these shocks ourselves in our models in the near future.

We have made a number of strong assumptions and used data from different sources. To recapitulate, we assume that the production function for an industry is linear and that it does not depend on the composition of occupations who are still able to work; we neglect absenteeism due to mortality and morbidity, as well as loss of productivity due to school closures (though we have argued these effects are small—see Appendix D.1). We have constructed our Remote Labour Index based on a subjective rating of work activities and we assumed that all work activities are equally important and they are additive. We have also applied a rating of essential industries for Italy to the US. Nonetheless, we believe that the analysis here provides a useful starting point for macroeconomic models attempting to measure the impact of the COVID-19 pandemic on the economy.

As new data become available we will be able to test whether our predictions are correct and improve our shock estimate across industries and occupations. Several countries have already started to release survey data. New measurements about the ability to perform work remotely in different occupations are also becoming available. New York and Pennsylvania have released a list of industries that are considered essential21 (though this is not currently associated with any industrial classification such as NACE or NAICS). Germany and Spain have also released a list of essential industries (Fana et al., 2020). As new data become available for the mitigation measures different states and countries are taking, we can also refine our analysis to account for different government actions. Thus we hope that the usefulness of methodology we have presented here goes beyond the immediate application, and will provide a useful framework for predicting economic shocks as the pandemic develops.

Appendix

A. Derivation of total shocks

A.1 Derivation of supply shocks

As discussed in the main text, we estimate the supply shock by computing an estimate of the share of work that will not be performed, which we compute by estimating the share of work that is not in an essential industry and that cannot be performed from home. We had to use several concordance tables, and make a number of assumptions, which we describe in detail here.

Figure 9 illustrates our method. There are four sets of nodes which are connected by three bipartite networks. The first set of nodes are the 6-digit NAICS industries which are classified to be essential or non-essential. This information is encoded in the K-dimensional column vector u which element uk = 1 if NAICS 6-digit industry k is essential and 0 otherwise. Second, there are N different industry categories on which our economic analysis is based. The 6-digit NAICS codes are connected to these industries by the incidence matrix (concordance table) S. The third set of nodes are the J occupations obtained from the BLS and O*NET data. The weighted incidence matrix M couples industries with occupations where the element Mnj denotes the number of people in occupation j being employed in industry n. Fourth, we also have a list of I work activities. Each activity was rated whether it can be performed from home. If activity i can be done from home, the ith element of the vector v is equal to 1, and otherwise it is equal to 0. The incidence matrix T denotes whether an occupation is associated with any given work activity, i.e. Tji = 1 if activity i is relevant for occupation j.

Figure 9:

The same schematic network representation of supply-side shocks as in the main text, but now also including mathematical notationNotes: The K-dimensional vector u below the NAICS 6-dig. (left nodes) encodes essential industries with binary elements. This set of nodes is connected to relevant industry categories by concordance tables (incidence matrix S). Matrix M connects the N industry categories with J occupations where an element represents the corresponding employment number. The ability to perform work activities (right nodes) from home is represented in vector v, also by binary elements. We use occupation-activity mappings provided from O*NET, represented as incidence matrix T. The grey arrows show the direction of shocks to industries and employment. The shock originating from the list of essential industries is mapped directly on to the broader industry categories, before it can be computed for occupations. Conversely, the Remote Labour Index is first mapped on to occupations and then projected on to industries.

The analysis presented here is based on I = 332 unique work activities, J = 740 occupations, and K = 1,057 6-digit NAICS industries. When relating to industry-specific results we use the BLS industry categories of the input–output accounts. We are able to derive supply shocks for N = 170 industries (out of 182 industries in the BLS data) for which we also have reliable data on value added, total output, and other key statistics. Employment, occupation, and wage statistics are available on a more fine-grained 4-digit NAICS level. We therefore use these N = 277 industries for deriving labour-specific results.

A.2 Industry-specific shocks

We can use this simple framework for deriving the supply shocks to industries.

(Non-)essential industries

To estimate the extent to which an industry category is affected by a shutdown of non-essential economic activities, we measure the fraction of its 6-digit NAICS sub-industries which are classified as non-essential. In mathematical terms, the essential-score for every industry is therefore a weighted sum which can be written compactly in matrix notation as

where S’ is the row-normalized version of matrix S with elements S’nk = Snk/ ∑ hSnh.

Note that this assumes that the fine-grained NAICS codes contribute uniformly to the more aggregate industry categories. Although this assumption might be violated in several cases, in absence of further information, we use this assumption throughout the text. Finally, we revised the essential score e of all industries. With the help of two colleagues with knowledge of the current situation in Italy we reclassified a small subset of industries with implausible essential scores (see Appendix B for details).

Industry Remote Labour Index

We can similarly estimate the extent to which the production of occupations or industries can take place by working from home. Since work activities are linked to occupations, but not directly to industries, we need to take two weighted averages to obtain the industry-specific RLI.

For each occupation we first measure the fraction of work activities that can be done from home. We interpret this as the share of work of an occupation that can be performed from home, or ‘occupation-level RLI’. This interpretation makes two assumptions: (i) that every work activity contributes equally to an occupation, which is our best guess since we do not have better data, and (ii) that if z per cent of activities cannot be done from home, the other 1 – z per cent of activities can still be carried out and are as productive as before.

For each industry i we then take a weighted average of the occupation-level RLIs, where the weights are the shares of workers employed in each occupation and in industry i. Let T’ denote the row-normalized version of matrix T, i.e. T’ji = Tji/ ∑ hTjh and similarly let the element of matrix M’ be M’nj = Mnj/ ∑ hMnh. Then the industry-specific RLI is given by the vector

We interpret the RLI for an industry, rn, as the fraction of work in an industry n that can be performed from home. As for assumption (ii) above for the occupation-level RLI, this assumes that if z per cent of the work of occupations cannot be done, the other 1 – z per cent of work can still be carried out.

Immediate industry supply shock

To derive industry supply shocks from the scores above, we need to take into account that industries might be exposed to both effects at the same time, but with different magnitudes. For example, consider the illustrative case of Chemical Manufacturing in Figure 9. Half of the industry is non-essential (red node ‘325130’) and could therefore be directly affected by an economic shutdown. But different occupations can be found in this industry that are affected heterogeneously. In this simple example, Chemical Manufacturing draws heavily on Boilermakers who have only work activities that cannot be done from home. On the other hand, this industry also has a tiny share of accountants and a larger share of Chemical Engineers who are able to do half of their work activities from home.

As stated above, the essential score en and the RLI rn can be interpreted as shares of industry-specific work which can be performed, either thanks to being essential or thanks to being adequately done from home. To compute the share of industry-specific work that can be performed due to either effect, we interpret shares as probabilities and assume independence,

where ISS stands for ‘industry supply shock’. We have multiplied the probability by minus one to obtain negative shocks. Although independence is a strong assumption, we have no reason to believe that the work that can be done from home is more or less likely to be judged essential. The empirical correlation coefficient of e and r is 0.04 and is far from being significant (p-value of 0.5), indicating that the independence assumption should have only minor effects on our results.

When applying these industry supply shocks to value added, we make the implicit assumption that a z per cent decrease in labour will cause a z per cent decrease in value added.

A.3 Occupation-specific shocks

We now describe how we compute shocks for specific occupations, rather than specific industries.

Occupations in (non-)essential industries

Occupations are mapped to industries through the weighted incidence matrix M, where an element denotes the number of jobs per occupation and industry.

The column-normalized matrix M∗with elements Mnj∗= M nj/∑hMhj denotes the share of an occupation carried out in a particular industry.22 The essential-score for occupations is taken as weighted average of the essential score for industries (computed in Eq. 1),

Occupation Remote Labour Index

As already indicated in the derivation of the industry-specific RLI, r, in Eq. (2), the occupation-specific RLI, y, is a weighted average of all the corresponding work activities that can be done from home. Formally, the occupation-based RLI is given by

Total supply-driven occupation shock

Following the same procedure as in Eq. (3), we can get the total immediate shock on occupations from the economy’s supply side.23 The combined immediate shock to occupations is then given as

OSSj=− (1− xj)  (1− yj).

(6)

Here, the correlation between RLI and the essential-score is larger, ρ(x, y) = 0.32 (p-value =2.8 × 10–19), and significant, which can also be seen from Figure 4. It should therefore be noted that the labour-specific results are expected to be more sensitive with respect to the independence assumption, as is the case for industry-related results.

A.4 Derivation of demand shocks

Since we have demand shocks only on the 2-digit NAICS level, disaggregating them into the more fine-grained relevant industry categories is straightforward when assuming that the demand shock holds equally for all sub-industries. We let the industry demand shock in percentages for industry n be –IDSn.

To map the demand shocks on to occupations, we can invoke the same matrix algebra as above. The occupation-specific shock originating from the economy’s demand side is then given by the projection

A.5 Total immediate (first-order) shocks

We now combine supply- and demand-driven shocks to total immediate shocks for occupations and industries.

Let us turn to industries first. As discussed in more depth in the main text, the shock experienced in the very short term is likely to be the worse of the two (supply and demand) shocks. Since we have expressed shocks as negative if they lead to decrease in output, in more mathematical terms, the industry total shock then is

and the occupation total shock is

OTSj = min(OSSj,ODS j).

(9)

Under these assumptions, the health sector will not experience a positive shock. We provide an alternative treatment in Appendix A.7.

A.6 Aggregate total shocks

To provide an economy-wide estimate of the shocks, we aggregate industry- or occupation-level shocks. We do this using different sets of weights.

First of all, consider the interpretation that our shocks at the occupation level represent the share of work that will not be performed. If we assume that if z per cent of the work cannot be done, z per cent of the workers will become unemployed, we can weigh the occupation shocks by the share of employment in each occupation. Using the vector L to denote the share of employed workers that are employed by occupation j, we have

Employment total shock  = OTSTL.

(10)

The employment supply (demand) shock is computed similarly but using OSS (ODS) instead of OTS.

Instead of computing how many workers may lose their jobs, we can compute by how much paid wages will decrease. For each occupation, we compute the total wage bill by multiplying the number of workers by the average wage. We then create a vector w where wj is the share of occupation j in the total wage bill. Then,

Wage total shock = OTSTw,

(11)

and similarly for the OSS and ODS. Note that we omit three occupations for which we do not have wages (but had employment).

Finally, to get an estimate of the loss of GDP, we can aggregate shocks by industry, weighting by the share of an industry in GDP. Denoting by Y the vector where Yn is the VA of industry n divided by GDP,24

Value added total shock = IT STY,

(12)

and similarly for the industry supply and demand shocks (ISS and IDS). Note that we could have used shares of gross output and compute a shock to gross output rather than to GDP.

A.7 Aggregate total shocks with growth of the health sector

Here we make a different assumption about how to construct the total shock for occupations and industries. For industries, we assume that if they experience a positive demand shock, the industries are able to increase their supply to meet the new demand. Instead of Eq. (8) we use

ITSnh={ITSn, if IDS n≤0IDSn,if IDSn>0.

(13)

Since occupations are employed by different industries, the total shock to an occupation can be influenced by positive demand shocks from the healthcare sector and negative demand shocks from non-essential industries. In Eq. (9) we consider that occupations only experience the negative shocks. An alternative is to consider both the negative shock caused by non-essential industries and the positive shock caused by the health industries. This gives

OTSjh={O TSj, if ODSj≤0ODSj+OSSj,if ODSj>0.

(14)

In section IV, specifically Figure 8, we use this convention for the y-axis, the Labour Shock. Using Eq. (14) allows us to observe how health-related occupations experience a positive shock.

In Table 6 we show the aggregate total shocks when using Eqs. (13) and (14). There is very little difference with the results in the main text. The health sector and its increase in demand are not large enough to make a big difference to aggregate results.

Table 6:

Main results allowing for growth in the health sector

B. Data

In this section we give more details about how we constructed all our variables. We stress that our goal was to produce useful results quickly and transparently, and make them available so that anyone can update and use them. We intend to improve these estimates ourselves in the future, as more information becomes available on the ability to work from home, which industries are essential, and how consumers react to the crisis by shifting their spending patterns.

(i) Italian list of essential industries

The Italian list of essential industries25 is based on the Statistical Classification of Economic Activities in the European Community, commonly referred to as NACE. Essential industries are listed with NACE 2-digit, 4-digit, and 6-digit codes. We automatically map industries listed at the 2- or 4-digit NACE level to NAICS 6-digit industries using the crosswalk made available by the European Commission.26 The 6-digit NACE level classification is country-specific and thus there is no official crosswalk to NAICS codes. We map the 6-digit industries by hand. In some cases, a 6-digit industry NAICS code maps into more than one NACE industry code. When this happens, we consider the NAICS industry to be essential if it maps into at least one essential NACE industry code. We then build the essential score for industries at the NAICS 4-digit level; the essential score of a 4-digit NAICS industry is the fraction of NAICS 6-digit subcategories that are essential.

In a second step, we looked at the resulting list of 4-digit NAICS industries and their essential score and discovered a few implausible cases, resulting from the complex mapping between the various classification systems at different levels. For instance, because Transport is essential, ‘Scenic and sightseeing transportation, other’ was considered essential. In contrast, ‘Death care services’ was classified as non-essential. Three of us, as well as two independent colleagues with knowledge of the current situation in Italy, evaluated the list and we proceeded to editing the 4-digit NAICS essential scores as follows. From non-essential to essential: grocery stores; health and personal care stores; gasoline stations; death care services. From essential (sometimes only partly) to non-essential: scenic and sightseeing transportation; independent artists, writers, and performers; software publishers; motion picture and video industries; sound recording industries; and other amusement and recreation industries. Finally, owner-occupied dwellings and federal, state, and local government were not classified, and we classified them as essential.

(ii) Data for occupations

O*NET has work activities data for 775 occupations, out of which 765 occupations have more than five work activities. We compute the Remote Labour Index for the 765 occupations with more than five work activities. From the May 2018 Occupational Employment Statistics (OES) estimates on the level of 4-digit NAICS (North American Industry Classification System), file nat4d_M2018_dl, which is available at https://www.bls.gov/oes/tables.htm under All Data, we find data for the number of employed workers of 807 occupations in 277 industries. These data cover 144m workers.27 From the sample of 765 occupations with RLI, and from the sample of 807 occupations with employment data from the BLS, we are able to match 740 occupations, which cover 136.8m workers. Therefore, our final sample has 740 occupations and 136.8m workers.28

With the occupation-industry employment data and the essential score of each industry, we estimate the share of essential jobs within each occupation. Additionally, we have wage information for most occupations (i.e. we have median and mean wage data for 732 and 737 occupations). We computed all correlations for median wage considering all occupations we had median wage data for. For the three occupations for which median wage data were missing, the colour coding of occupations in Figures 4, ​7, and the x-axis in Figure 8 corresponds to the average (across all occupations) of the median wage. We used the mean wages and the employment of occupations to define the wage quartiles of our sample. We excluded the three occupations for which we did not have mean wage data from these calculations.

Figure 10:

Left: Relationship between Remote Labour Index and median wage. The Pearson correlation is 0.46 (p-value = 5.6 × 10–39). Right: Relationship between fraction of workers in essential industries and wage. The Pearson correlation is 0.36 (p-value = 1.5 × 10–24).

Figure 11:

Left: Distribution of the RLI for the 740 occupations. Right: Distribution of the share of essential jobs within each of the 740 occupations

Figure 12:

Left: Supply shock distribution across occupations. Right: Demand shock distribution across occupations.

Figure 13:

Left: Shock distribution for occupations. Right: Distribution of exposure to disease

Finally, we use the O*NET data on exposure to disease and infection of occupations for the colour coding in Figure 8. We explain these data further in Appendix C. In the following charts we show the distribution of the RLI, exposure to disease and infection, supply, demand, and overall shocks across occupations.

(iii) Data for industries

Matching all data to BLS I-O industries

A key motivation of this paper is to provide relevant economic data which can be used by other researchers and policy-makers to model the economic impact of the COVID-19 pandemic. We therefore bring the supply and demand shock data into a format that matches directly to US input–output data.

We use the BLS 2018 input–output account, which allows us to discern 179 private sectors. There are the additional industries Private Households, NAICS 814, and Postal Service, NAICS 491. The data also contain 19 different industries relating to governmental activities. Since these industries are not classified with NAICS codes, we aggregate all governmental industries into a single node Government, which can be interpreted as the NAICS 2-digit industry 92. This leaves us with 182 industry categories which are a mixture of 2- to 6-digit NAICS industries. Moreover, the data contains one special industry ‘Owner-occupied dwellings’ which is not classified by NAICS codes yet relevant for GDP accounting.

We are able to match occupational data to 170 out of the 182 industry categories, accounting for 97 per cent of total value added (excluding Owner-occupied dwellings). For this subset we compute industry-specific RLIs, essential scores, and supply shocks as spelled out in Appendix A.1, as well as employment-weighted infection exposures.

Since we have demand shocks only at the 2-digit NAICS level, disaggregating them into the more fine-grained BLS input–output data is straightforward when assuming that the demand shocks hold equally for all sub-industries.

In the online data repository, we also report total wages and total employment per industry. We use the same OES estimates as for the occupational data, but match every industry category according to the corresponding NAICS 2- to 6-digit digit levels.

Figures 14 to 16 show distributions of supply and demand shock-related variables on the industry level. Table 7 summarizes a few key statistics for these industries, when further aggregated to 72 industry categories.

Table 7:

Key statistics for different 2- and 3-digit NAICS industries

Figure 14:

Left: Distribution of the Remote Labour Index, aggregated to 170 industries. Right: Fractions of essential sub-industries per industry category

Figure 15:

Left: Supply shock distribution across industries. Right: Demand shock distribution across industries

Figure 16:

Shock distribution across industries

C. Occupations most at risk of contracting SARS-CoV-2

O*NET makes available online work context data for occupations that describe the physical and social factors that influence the nature of work. The ‘Exposed to disease and infection’ work context,29 which we refer to as ‘exposure to infection’ for short, describes the frequency with which a worker in a given occupation is exposed to disease or infection. It ranges from 0 to 100, where 0 means ‘never’ and 100 ‘every day’; an exposed to infection rating of 50 means an exposure of ‘once a month or more but not every week’ and 75 means ‘Once a week or more but not every day’. We have exposure to infection data for 737 of the 740 occupations in our sample. For those occupations for which we did not have the exposure to infection, we coloured them as if they had zero exposure to infection.

As we see in Figure 17, there is a U-shaped relationship between wages and exposure to infection. There is a correlation of 0.08 (p-value = 0.02) between wages and exposure to infection, but this is misleading.30 Though many high-wage occupations are highly exposed to infection (highly paid doctors), there are also many low-wage occupations with high probability of infection.

Figure 17:

Relationship between wage and probability of infectionNotes: The Pearson correlation is 0.08 (p-value = 0.2). However, we consider that this correlation is mostly driven by high salaries in the health sector, but there are many low-wage occupations with a significant exposure to infection.

D. Discussion of labour supply shocks which we do not include

(i) Labour supply shocks from mortality and morbidity

Typical estimates

McKibbin and Fernando (2020) consider attack rates (share of population who become sick) in the range 1–30 per cent and case-fatality rates (share of those infected who die) in the range 2–3 per cent. From attack rates and case fatality rates, they compute mortality rates. They also assume that sick people stay out of work for 14 days. A third effect they assume is that workers would be care-givers to family members.

For their severe scenario of an influenza pandemic, Congressional Budget Office (2006) assumed that 30 per cent of the workers in each sector (except for Farms, which is 10 per cent) would become ill and would lose 3 weeks of work, at best, or die (2.5 per cent case fatality rate).

Best guess for current effect of COVID-19

In the case of COVID-19, estimating a labour supply shock is made difficult by several uncertainties. First of all, at the time of writing there are very large uncertainties on the ascertainment rate (the share of infected people who are registered as confirmed cases), making it difficult to know the actual death rate.

We report the result from a recent and careful study by Verity et al. (2020), who estimated an infection fatality ratio of 0.145 per cent (0.08–0.32) for people younger than 60, and 3.28 per cent (1.62–6.18) for people aged 60 or more. The age bracket 60–69, which in many countries will still be part of the labour force, was reported as 1.93 per cent (1.11–3.89).

Taking the infection fatality ratio for granted, the next question is the attack rate. In Verity et al. (2020), the infection fatality ratios are roughly one-fourth of the case fatality ratios, suggesting that three-quarters of the cases are undetected. For the sake of the argument, consider Italy, a country that has been strongly affected and appears to have reached a peak (at least of a first wave). There are at the time of writing 132,547 cases in Italy.31 In 2018 the population of Italy32 was 60,431,283. If we assume that Italy is at the peak today and the curve is symmetric, the total number of cases will be double the current number, that is 265,094, which is 0.44 per cent of the population. If we assume that the true number of cases is four times higher, the attack rate is, roughly speaking, 1.76 per cent. These numbers are more than an order of magnitude smaller than the number who cannot work due to social distancing.

Thus, while it is clear that the virus is causing deep pain and suffering throughout Italy, the actual decrease in labour supply, which is massive, is unlikely to be mostly caused by people being sick, and is much more a result of social distancing measures.

Uncontrolled epidemic

Now, it may be informative to consider the case of an uncontrolled epidemic. If we assume that the uncontrolled epidemic has an attack rate of 80 per cent (a number quoted in Verity et al. (2020)), an infection fatality ratio for people in the labour force of 1 per cent (an arbitrary number between 0.145 per cent for people younger than 60, and 1.93 per cent for the 60–69 age bracket) implies an 0.8 per cent permanent decrease of the labour force. If we assume that those who do not die are out of work for 3 weeks, on an annual basis of 48 worked weeks, we have (3/48)*(0.80–0.01)=4.94 per cent decrease of the labour supply.

Overall, this exercise suggests that left uncontrolled, the epidemic can have a serious effect on labour supply. However, in the current context, the effect on the economy is vastly more a result of social distancing than direct sickness and death.

(ii) Labour supply shocks from school closure

School closures are a major disruption to the functioning of the economy as parents can no longer count on the school system to care for their children during the day.

Chen et al. (2011) surveyed households following a school closure in Taiwan during the H1N1 outbreak, and found that 27 per cent reported workplace absenteeism. Lempel et al. (2009) attempted to estimate the cost of school closure in the US in case of an influenza pandemic. They note that 23 per cent of all civilian workers live in households with a child under 16 and no stay-at-home adults. Their baseline scenario assumes that around half of these workers will miss some work leading to a loss of 10 per cent of all labour hours in the civilian US economy, for as long as the school closure lasts.

Some of these effects would already be accounted for in our shocks. For instance, some workers are made redundant because of a supply or demand shock, so while they have to stay at home to care for their children, this is as much a result of labour and supply shocks as a result of school closure. For those working from home, we might expect a decline in productivity. Finally, for those in essential industries, it is likely that schools are not closed. For instance, in the UK, schools are opened for children of essential workers. Our list of essential industries from Italy includes Education.

Overall, school closures indeed have large effects, but in the current context these may already be accounted for by supply and demand shocks, or non-existent because schools are not fully closed. The loss of productivity from parents working from home remains an open question.

E. Additional estimates of demand or consumption shocks

In this appendix we provide additional data on the demand shock. Table 8 shows our crosswalk between the industry classification of the Congressional Budget Office (2006) and NAICS 2-digit industry codes, and, in addition to the ‘severe’ shocks used here, shows the CBO’s ‘mild’ shocks. We have created this concordance table ourselves, by reading the titles of the categories and making a judgement. Whenever NAICS was more detailed, we reported the CBO’s numbers in each more fine-grained NAICS.

Table 8:

Mapping of CBO shocks to NAICS 2-digits

We also provide three sources of consumption shocks (in principle, these estimates are meant to reflect actual decreases in consumption rather than shifts of the demand curve). Table 9 shows the consumption shocks used by Keogh-Brown et al. (2010) to model the impact of potential severe influenza outbreak in the UK. Table 10 shows the consumption shocks used by Muellbauer (2020) to model the impact of the COVID-19 on quarterly US consumption. OECD (2020) provided two other sources, both reported in Table 10. The first one is based on assumptions of shocks at the industry level, while the other shows assumptions of shocks by expenditure categories (COICOP: Classification of individual consumption by purpose).

Table 9:

Table 10:

Estimates of consumption shocks from various sources

The aim of this paper was a timely prediction of first-order shocks before relevant data became available. While realized consumption is in principle different from demand shocks, it is instructive to look at the various studies of sectoral consumption that have appeared since our first paper.

Baker et al. (2020) use transaction-level data from a non-profit Fintech company to measure changes in consumption behaviour in the US. They find an increase in consumer spending at the early stage of the pandemic due to stockpiling, and sharp declines in most consumption categories in the subsequent weeks with public transportation, air travel, and restaurants experiencing the largest impacts.

Based on a survey of roughly 14,000 respondents, Coibion et al. (2020) analyse spending in several consumption categories as well as plans to buy durable goods. They find negative changes in all categories with the largest drops in travel, clothing, debt payments, and housing, and a decline in total spending by around 30 log percentage points.

Using daily data on bank card transactions of the second largest Spanish bank, Carvalho et al. (2020) study changes in consumption behaviours for 2.2m merchants between 1 January 2019 to 30 March 2020. Their results indicate that total consumption was rising before the enactment of the nationwide lockdown, and drastically falling thereafter (almost by 50 per cent compared to previous year levels). They also note substitution effects from offline to online payments. In line with our analysis they find substantial adjustments in the market shares of different expenditure categories. While categories like food shops and supermarkets, tobacconists, and pharmacies have experienced the largest increase in the consumption basket, restaurants, night clubs, furniture stores, and clothes shops experienced the largest decline in relative importance.

Analysing transactions of one million credit card users in Japan, Watanabe (2020) shows that aggregate consumption declined by 14 per cent. Travel spending experienced the largest decrease of 57 per cent. Declines in spending on goods tend to be smaller than declines in the consumption of services, with supermarkets and e-commerce experiencing even positive consumption impacts.

Using Chinese offline transaction data, Chen et al. (2020) find a substantial decrease in both goods and services consumption of around one-third. In line with other studies, the largest drops are in dining and entertainment as well as in travel, falling by 64 per cent and 59 per cent, respectively. Consumption response varies in magnitude for different Chinese cities, depending on how strongly they have been affected by the pandemic.

Andersen et al. (2020) analyse transaction-level customer data from the largest bank in Denmark. They estimate consumption levels to be 27 per cent below counterfactual levels without the pandemic. For retail, restaurants, and travel they find consumption drops of 24.7, 64, and 84.5 per cent, respectively, and report an increase of 9 per cent in grocery consumption.

Finally, according to the Opportunity Insights Economic recovery tracker (Chetty et al., 2020)33, compared to January 2020, US consumption fell very sharply up to 33 per cent before starting a slow recovery. Out of the six categories tracked, Groceries is the only sector exhibiting an increased consumption, while Health care, in sharp contrast to our predicted demand shocks, shows an impressive fall, by up to 58 per cent.

Notes

We would like to thank Eric Beinhocker, Stefania Innocenti, John Muellbauer, Marco Pangallo, and David Vines for many comments and discussions. We are also grateful to Andrea Bacilieri and Luca Mungo for their help with the list of essential industries. We thank Baillie Gifford, IARPA, the Mexican Energy Ministry (SENER), the Mexican Science and Technology Research Council (CONACyT Mexico), and the Oxford Martin School for the funding that made this possible. This research is based upon work supported in part by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via IARPA-BAA-17-01, Contract No. 2019-19020100003. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of ODNI, IARPA, or the US government. The US government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright annotation therein.

Footnotes

1This paper was prepared in March and early April 2020 and released on 16 April. This version contains only minor changes rather than comprehensive updates.

2The results reported here are a slightly updated version of our work released on 16 April; we use a revised and updated list of essential industries, and we no longer exclude owner-occupied dwellings (imputed rents) from GDP (we assume no shocks to this sector). Our overall results do not change substantially.

3See Appendix D1. for rough quantitative estimates in support of this argument.

4 https://www.insee.fr/en/statistiques/4473305?sommaire=4473307

5In the future we intend to redo this using O*NET’s ‘detailed’ work activity data, which involve over 2,000 individual activities associated with different occupations. We believe this would somewhat improve our analysis, but think that the intermediate activity list provides a good approximation. All updates will be made available in the online data repository (see footnote 6).

6An activity was considered to be able to be performed at home if three or more respondents rated this as true. We also undertook a robustness analysis where an activity was considered to be able to be performed at home based on two or more true ratings. Results remained fairly similar. In post-survey discussion, we agreed that the most contentious point is that some work activities might be done from home or not, depending on the industry in which they are performed.

7 https://zenodo.org/record/3744959

8Available at https://www.onetcenter.org/crosswalks.html

9We omitted ten occupations that had fewer than five work activities associated with them. These occupations include Insurance Appraisers Auto Damage; Animal Scientists; Court Reporters; Title Examiners, Abstractors, and Searchers; Athletes and Sports Competitors; Shampooers; Models; Fabric Menders, Except Garment; Slaughterers and Meat Packers; and Dredge Operators.

10There are a few cases that we believe are misclassified. For example, two occupations with a high RLI that we think cannot be performed remotely are real estate agents (RLI = 0.7) and retail salespersons (RLI = 0.63). However, these are exceptions—in most cases the rankings make sense.

11We use the May 2018 Occupational Employment Statistics (OES) estimates on the level of 4-digit NAICS (North American Industry Classification System), file nat4d_M2018_dl, which is available at https://www.bls.gov/oes/tables.htm under All Data. Our merged dataset covers 136.8 out of 144 million employed people (95 per cent) initially reported in the OES.

12Mapping NACE industries to NAICS industries is not straightforward. NACE industry codes at the 4-digit level are internationally defined. However, 6-digit level NACE codes are country specific. Moreover, the list of essential industries developed by Italy involves industries defined by varying levels of aggregation. Most essential industries are defined at the NACE 2-digit and 4-digit level, with a few 6-digit categories thrown in for good measure. As such, much of our industrial mapping methodology involved mapping from one classification to the other by hand. We provide a detailed description of this process in Appendix B.1.

13In fact we allow for a continuum between the ability to work from home, and an industry can be partially essential.

14Since relevant economic variables such as total output per industry are not extensively available on the NAICS 4-digit level, we need to further aggregate the data. We derive industry-specific total output and value added for the year 2018 from the BLS input–output accounts, allowing us to distinguish 170 industries for which we can also match the relevant occupation data. The data can be downloaded from https://www.bls.gov/emp/data/input-output-matrix.htm.

15Since rents account for an important part of GDP, we make an additional robustness check by considering the Real Estate sector essential. In this scenario the supply and total shocks drop by 3 percentage points.

16As before, Table 6 in Appendix A.7 gives the results assuming positive total shocks for the health sector, but shows that it makes very little difference.

17 https://www.onetonline.org/find/descriptor/result/4.C.2.c.1.b

18Our demand shocks do not have an increase in retail but, in the UK, supermarkets have been trying to hire several tens of thousands of workers (Source: BBC, 21 March, https://www.bbc.co.uk/news/business-51976075). Baker et al. (2020) document stock-piling behaviour in the US.

19Employment Rate, aged 15–64, all persons for the US (FRED LREM64TTUSM156N) fell from 71.51 in December 2007 to 68.23 in June 2009, the employment peak to trough during the dates of recession as defined by the NBER.

20Our data repository is at https://zenodo.org/record/3744959, where we will post any update.

21 https://esd.ny.gov/guidance-executive-order-2026

22Note that we column-normalize M to map from industries to occupations and row-normalize when mapping from occupations to industries.

23To be clear, this is a product market supply-side shock, but this translates into a reduction in labour demand in each occupation.

24Our estimate of GDP is the sum of VA of industries in our sample.

25Available at http://www.governo.it/sites/new.governo.it/files/dpcm_20200322.pdf, 22 March.

26 https://ec.europa.eu/eurostat/ramon/relations/index.cfm?TargetUrl=LST_REL&StrLanguageCode=EN&IntCurrentPage=11

27The US economy had 156m workers mid-2018, see https://fred.stlouisfed.org/series/CE16OV

28Note that the BLS employment data we use here do not include self-employed workers (which currently accounts for about 16m people).

29 https://www.onetonline.org/find/descriptor/result/4.C.2.c.1.b?s=2

30For example, Adams-Prassl et al. (2020), using survey evidence for ~4,000 US individuals, found that workers without paid sick leave are more likely to go to work in close proximity to others, which may have suggested a negative correlation between wages and exposures. Note, however, that our correlation is based on occupations, not individuals, and that wages are not necessarily an excellent predictor of having paid sick leave or not.

31 https://coronavirus.jhu.edu/map.html

32 https://data.worldbank.org/indicator/SP.POP.TOTL?locations=IT

33 https://tracktherecovery.org, accessed 27 June 2020

References

• Adams-Prassl A., Boneva T., Golin M., and Rauh C (2020), ‘The Large and Unequal Impact of Covid-19 on Workers’, VoxEU.org, 8 April, https://voxeu.org/article/large-and-unequal-impact-covid-19-workers [Google Scholar]
• Andersen A. L., Hansen E. T., Johannesen N., and Sheridan A (2020), ‘Consumer Responses to the COVID-19 Crisis: Evidence from Bank Account Transaction Data’, available at SSRN 3609814.
• Baker S. R., Farrokhnia R., Meyer S., Pagel M., and Yannelis C (2020), ‘How Does Household Spending Respond to an Epidemic? Consumption during the Covid-19 Pandemic’, NBER WP 26949. [Google Scholar]
• Baldwin R., and Weder di Mauro B (2020), ‘Introduction’, ch. 1 in Baldwin R. and Weder di Mauro B. (eds), Economics in the Time of COVID-19, London, CEPR, 1–30. [Google Scholar]
• Balleer A., Zorn P., Link S., and Menkhoff M (2020), ‘Demand or Supply? Price Adjustment during the Covid-19 Pandemic’, CEPR DP14907. [Google Scholar]
• Barrot J.-N., Grassi B., and Sauvagnat J (2020), ‘Sectoral Effects of Social Distancing’, Covid Economics 3, 10 April, 85–102. [Google Scholar]
• Bekaert G., Engstrom E., and Ermolov A (2020), ‘Aggregate Demand and Aggregate Supply Effects of COVID-19: A Real-time Analysis’, available at SSRN 3611399. [Google Scholar]
• Boone L. (2020), ‘Tackling the Fallout from Covid-19’, ch. 2 in Baldwin R. and Weder di Mauro B. (eds), Economics in the Time of COVID-19, London, CEPR, 37–44. [Google Scholar]
• Brinca P., Duarte J. B., and Faria-e-Castro M (2020), ‘Measuring Sectoral Supply and Demand Shocks during COVID-19’, FRB St Louis Working Paper, 2020-011. [Google Scholar]
• Bullard J. (2020), ‘Expected US Macroeconomic Performance during the Pandemic Adjustment Period’, VoxEU.org, 4 April, available at https://voxeu.org/article/expected-us-macroeconomic-performance-during-pandemic-adjustment-period [Google Scholar]
• Carvalho V. M., Hansen S., Ortiz Á., Garcia J. R., Rodrigo T., Rodriguez Mora S., and Ruiz de Aguirre P (2020), ‘Tracking the Covid-19 Crisis with High-resolution Transaction Data’, CEPR DP14642. [Google Scholar]
• Chen H., Qian W., and Wen Q (2020), ‘The Impact of the COVID-19 Pandemic on Consumption: Learning from High Frequency Transaction Data’, available at SSRN 3568574. [Google Scholar]
• Chen W.-C., Huang A. S., Chuang J.-H., Chiu C.-C., and Kuo H.-S (2011), ‘Social and Economic Impact of School Closure Resulting from Pandemic Influenza a/h2n1’, Journal of Infection, 62(3), 200–3. [PubMed] [Google Scholar]
• Chetty R., Friedman J. N., Hendren N., and Stepner M (2020), ‘How Did COVID-19 and Stabilization Policies Affect Spending and Employment? A New Real-Time Economic Tracker Based on Private Sector Data’, Opportunity Insights Working Paper. [Google Scholar]
• Coibion O., Gorodnichenko Y., and Weber M (2020), ‘Labor Markets during the Covid-19 Crisis: A Preliminary View’, NBER WP 27017. [Google Scholar]
• Congressional Budget Office (2006), ‘Potential Influenza Pandemic: Possible Macroeconomic Effects and Policy Issues’, https://www.cbo.gov/sites/default/files/109th-congress-2005–2006/reports/12-08-birdflu.pdf
• del Rio-Chanona R. M., Mealy P., Beguerisse-Díaz M., Lafond F., and Farmer J. D (2019), ‘Automation and Occupational Mobility: A Data-driven Network Model’, arXiv:1906.04086. [Google Scholar]
• Dingel J.. and Neiman B (2020), ‘How Many Jobs can be Done at Home?’, Becker Friedman Institute; White Paper, https://bfi.uchicago.edu/wp-content/uploads/BFI_White-Paper_Dingel_Neiman_3.2020.pdf [PMC free article] [PubMed] [Google Scholar]
• Fana M., et al. (2020), ‘The COVID Confinement Measures and EU Labour Markets’, No. JRC120578, Joint Research Centre (Seville site). [Google Scholar]
• Goodhart C., and Pradhan M (2020), ‘Future Imperfect After Coronavirus’, VoxEU.org, 27 March. https://voxeu.org/article/future-imperfect-after-coronavirus [Google Scholar]
• Guerrieri V., Lorenzoni G., Straub L., and Werning I (2020), ‘Macroeconomic Implications of COVID-19: Can Negative Supply Shocks Cause Demand Shortages?’, No. w26918, National Bureau of Economic Research. [Google Scholar]
• Hicks Michael J., Faulk D., and Devaraj S (2020), ‘Occupational Exposure to Social Distancing: A Preliminary Analysis Using O*NET data’, Becker Friedman Institute; White Paper. https://projects.cberdata.org/170/social-distance. [Google Scholar]
• Inoue H., and Todo Y (2020), ‘The Propagation of the Economic Impact Through Supply Chains: The Case of a Mega-city Lockdown against the Spread of covid-19’, SSRN 3564898. [PMC free article] [PubMed] [Google Scholar]
• Keogh-Brown M. R., Wren-Lewis S., Edmunds W. J., Beutels P., and Smith R. D (2010), ‘The Possible Macroeconomic Impact on the UK of an Influenza Pandemic’, Health Economics, 19(11), 1345–60. [PubMed] [Google Scholar]
• Koren M., and Petö R (2020), ‘Business Disruptions from Social Distancing’, arXiv:2003.13983. [Google Scholar]
• Lempel H., Epstein J. M., and Hammond R. A (2009), ‘Economic Cost and Health Care Workforce Effects of School Closures in the US’, PLoS currents, 1. [PMC free article] [PubMed] [Google Scholar]
• McKibbin W. J., and Fernando R (2020), ‘The Global Macroeconomic Impacts of Covid-19: Seven Scenarios’, CAMA Working Paper No. 19/2020. [Google Scholar]
• — Sidorenko A. A (2006), ‘Global Macroeconomic Consequences of Pandemic Influenza’, Lowy Institute for International Policy, Sydney. [Google Scholar]
• Mann C. (2020), ‘Real and financial lenses to assess the economic consequences of covid-19’, ch. 8 in Baldwin R. and Weder di Mauro B. (eds), Economics in the Time of COVID-19, London, CEPR, 81–6. [Google Scholar]
• Miles D., and Scott A (2020), ‘Will Inflation Make a Comeback after the Crisis Ends?’, VoxEU.org, 4 April, https://voxeu.org/article/will-inflation-make-comeback-after-crisis-ends [Google Scholar]
• Muellbauer J. (2020), ‘The Coronavirus Pandemic and US Consumption’, VoxEU.org, 11 April, https://voxeu.org/article/coronavirus-pandemic-and-us-consumption [Google Scholar]
• OECD (2020), ‘Evaluating the Initial Impact of Covid-19 Containment Measures on Economic Activity’, OECD Technical Report, March 2020, https://read.oecd-ilibrary.org/view/?ref=126_126496-Evgsi2gmqj&title=Evaluating_the_initial_impact_of_COVID-19_containment_measures_on_economic_activity. [Google Scholar]
• Office for National Statistics (2020), ‘Coronavirus and Homeworking in the UK Labour Market: 2019’, ONS Technical Report, https://www.ons.gov.uk/employmentandlabourmarket/peopleinwork/employmentandemployeetypes/articles/coronavirusandhomeworkingintheuklabourmarket/2019. [Google Scholar]
• Pichler A., Pangallo M., del Rio-Chanona R. M., Lafond F., and Farmer J. D (2020), ‘Production Networks and Epidemic Spreading: How to Restart the UK Economy?’, arXiv preprint arXiv:2005.10585. [Google Scholar]
• Santos J. R., May L., and Haimar A. E (2013), ‘Risk-based Input-Output Analysis of Influenza Epidemic Consequences on Interdependent Workforce Sectors’, Risk Analysis, 33(9), 1620–35. [PMC free article] [PubMed] [Google Scholar]
• Verity R., Okell L. C., Dorigatti I., Winskill P., Whittaker C., Imai N., Cuomo-Dannenburg G., Thompson H., Walker P. G., Fu H., et al. (2020), ‘Estimates of the Severity of Coronavirus Disease 2019: A Model-based Analysis’, The Lancet Infectious Diseases, 20(6), 669–77. [PMC free article] [PubMed] [Google Scholar]
• Wallis J. J. (1989), ‘Employment in the Great Depression: New Data and Hypotheses’, Explorations in Economic History, 26, 45–72. [Google Scholar]
• Watanabe T. (2020), ‘The Responses of Consumption and Prices in Japan to the COVID-19 Crisis and the Tohoku Earthquake’, Working Papers on Central Bank Communication 020, University of Tokyo, Graduate School of Economics. [Google Scholar]
• Zhang S. X., Wang Y., Rauch A., and Wei F (2020), ‘Health, Distress and Life Satisfaction of People in China One Month into the Covid-19 Outbreak’, 10.1101/2020.03.13.20034496. [PMC free article] [PubMed] [CrossRef] [Google Scholar]