The effects of the Covid-related lockdowns are massive (see McKibbin and Fernando 2020). In Italy, we estimate a 44% reduction in the value of potential output produced, peaking at 69% for construction and real estate and 63% for mechanics. Activities were closed following an administrative decision of the government, differently from other countries like Germany or France, where the lockdown was in most cases the indirect effect of social confinement (Barrot et al. 2020, Fadinger and Schymik 2020). As a result, GDP in Italy is expected to drop by around 10% in 2020, according to most forecasts.
In Italy, most activities were reopened in May, although with social distancing and health safety guidelines. In this column, we argue that a targeted exit from the lockdown could have been implemented instead, giving priority to those activities with the greatest impact on the national economy. This targeted strategy, combined with an assessment of the inherent health risks of each activity (as done by the government), would have reduced the risks of a second wave of contagion for those returning to work and for the country at large, while achieving a reactivation of gross output and jobs similar to the general reopening actually implemented.
In a recent paper (Barba Navaretti et al. 2020), we propose an approach to identify production activities for which total or partial closures (and consequently reopening) would have the greatest impact on the country's GDP, output, and employment, using input-output tables and network centrality measures in production chains. Our approach provides an operative tool for clearly identifying the economic effects of lockdowns at a granular level. We show, for example, that the activation of just 20 core micro-sectors would raise the capacity of the Italian economy, as measured by the value of gross output, from 56% (under the administrative lockdown as of 4 May) to 76%, with respect to the pre-Covid-19 level. The impact would be especially sizeable in core production chains. For instance, the value of gross output of mechanics would increase from 37% to 84% and of constructions from 31% to 77%. We estimate that the administrative lockdown, as of 4 May, if kept for an entire year, would have implied a drop in GDP of 52%. The reopening of the sectors identified in our analysis would limit this fall to 16% (assuming the other activities remain closed for a whole year).
Understanding how to minimise the impact of the lockdown on the economy is of course important to design the re-opening strategy, but it is even more crucial in the event of having to face a possible resurgence of the Covid-19 pandemic or a new pandemic in the future. We all wish these events would never happen, but the main lesson from these tragic months is that we can no longer be unprepared in confronting them.
Identifying core sectors of the economy is not easy, because size is not the only issue. Given the tangled nature of value chains, there are activities that may seem insignificant from a quantitative point of view, but are fundamental links in several production chains, and therefore have a significant indirect impact on the production capacity of the country. Input-output (IO) tables are the starting point to unfold the interconnections among different sectors of activity. A parallel strand of analysis, closer to the business and management literature, has developed from the concept of production or supply chains, building on the seminal contribution of Porter (1985). We merge these two strands of the literature, integrating information from the IO matrices of the Italian economy (produced by Istat), with those from the structure of the value chains of the Italian economy, built by Prometeia. In our analysis, we exploit two sets of analytical tools that have been developed in the recent years: the methodology to extract some sectors of economic activities from IO matrices proposed by Dietzenbacher and Lahr (2013), and the techniques used by social network analysis to identify key players within a system (Jackson 2010).1
The tangled web of the economy
To disentangle the intricacies of the Italian economy we move in three steps.
First, among the 63 sectors of the national IO tables2, we identify those in which closure causes the largest drop of GDP, thus simulating the effect of their total or partial lockdown as implemented by the government decrees of 22 March and 10 April (Dietzenbacher and Lahr 2013).
Table 1 below lists the sectors for which a yearly closure (as of Italy’s Prime Minister Decrees of 22 March and 10 April 2020) would have an impact on GDP greater than 3%.
Table 1 Impact of one year of administrative lockdown on GDP by industry, as of IO tables
Notes: The table reports the estimates of the drop in GDP caused by the lockdown of a sector’s economic activities as of Italy’s Prime Minister Decrees of 22 March and 10 April 2020, if kept for an entire year. Only sectors with an estimated drop larger than 3% of GDP are listed.
The size of the decline in GDP can be decomposed into three factors: (i) the size of the sector, (ii) the degree of interconnection between the sector and other upstream and downstream sectors, and (iii) the degree of closure of the sector imposed by the Ministerial Decrees (i.e. the larger the extent of the lockdown, the larger the economic impact).
The sectors identified at NACE 2 digits are large and made of heterogeneous activities. At the same time, activities which are relatively small and have small weights in input-output tables may provide crucial inputs or crucial outlets to more than one production chain. Hence their closure may nevertheless endanger a large share of national output. This effect would not be detected by input-output table.
For this reason, as our second step, we revert to finer industry statistics, and also to the analysis of specific production chains.3
Our units of analysis are the 192 micro-sectors included in production chains. We focus on those included in the core NACE 2 digits sectors identified above in Table 1. We use social network analysis to study the links among each micro-sector and identify the most central ones within and between production chains.4
The literature on social networks has proposed several measures to characterise the relevance of each node in addition to the number of connections. We choose eigenvector centrality, which estimates the relative relevance of a node within a network.5 Micro-sectors with a higher eigenvector centrality, therefore, are more relevant within each production chain and across a larger number of production chains.
Figure 1 provides a representation of our network.6 Each dot represents a micro-sector, with its size proportional to its eigenvector centrality. The ‘tails’ in the picture represent production chains and their dots correspond to micro-sectors which only belong to a single production chain, therefore having lower centrality values. Dots in the centre represent micro-sectors with high centrality values because they interact with many micro-sectors across different production chains (e.g. wholesalers of intermediate industrial goods).
Figure 1 Micro-sector’s network
The micro-sectors are then ranked according to their eigenvector centrality. We have identified three groups of micro-sectors. A first group includes the 20 most central micro-sectors, with a total value of production when fully open of over €820 billion (23% of the total production of the Italian economy). As of 4 May, these micro-sectors were operating at a mere 13% of their potential output. Of the approximately four million people employed before the crisis, just over half a million were at work on 4 May. If these 20 micro-sectors were allowed to operate at full capacity, the total value of output of the Italian economy would rise from 56% of its potential to as much 76%.
Nonetheless, while inspecting production chains, one immediately realises that in some of them (e.g. fashion) many core activities, such as clothing and footwear, would still remain closed, making the opening up of the identified micro-sectors ineffective for these production chains. We therefore consider a second set of the next 20 micro-sectors in terms of centrality, which are less central than the initial 20, but still large and with a crucial role in core production chains.
Finally, by combining information on the centrality of the network with qualitative assessments regarding the articulation of the individual production chains, one can also identify ten additional micro-sectors. Since they are relatively small, accounting for only 3.9% of total production, they cannot be identified by our procedure. Yet they are crucial enablers of entire production chains (e.g. packaging paper for the food industry or textile finishing in fashion).
Table 2 reports output capacity for each production chain (measured as gross value of output) compared to pre Covid levels, under four scenarios: the administrative lockdown as of 4 May (only essential activities are allowed - column 1) and following the opening up of the core 20 (column 2), 40 (column 3), and 50 (column 4) micro-sectors.
With just 50 micro-sectors operating at full capacity in addition to those deemed essential, out of a total number of 192, production chains such as Agrifood, Media and TLC, Transport and logistics, Energy and utilities, Health and Mechanics would be almost entirely active (reaching a capacity between 93% and 100%). All other industries would operate above 80% of capacity.
Table 2 Central micro-sectors and total production
Notes: The table reports output capacity (gross value of output) compared to pre-Covid levels under administrative lockdown and the complete opening of 20, 40, and 50 core micro-sectors
Back to IO and impact on GDP
The final step of our methodology is to estimate the impact on GDP of the core micro-sectors identified above. We therefore revert to the IO tables. As a benchmark, we report in Table 3 column 1 the impact of the administrative lockdown (also reported in Table 1)compared to pre-Covid levels when everything was open. We than report in columns 2 and 3 what would be the decline in GDP, if closures were to exclude the core micro-sectors. All closures are assumed to last for one year.
Allowing production just in the first 20 micro-sectors identified above would reduce the negative impact on GDP of a lockdown of the construction industry by more than 10%. Equally sizeable would be the impact on wholesale trade, excluding that of cars and motorcycles (from 9.2% to 1.7%) and accommodation and restaurant services (from 8.7% to 1.4%). Opening the second group of 20 micro-sectors would in turn have a sizeable impact for the manufacture of vehicles, trailers and semi-trailers, narrowing the reduction in GDP from 8.3% to 1.4% and also for the manufacturing of metal products and textiles.
Table 3 Core micro-sectors and GDP
Notes: The table reports the estimates of the drop in GDP caused by the lockdown as of 4 May (column 1) and if the lockdown excluded the first 20 (column 2) and 40 (column 3) core sectors.
All economies are exiting the lockdown imposed by the Covid-19 pandemic. Governments are planning the re-opening of most activities, according to both health and economic criteria. In May, the Italian government lifted the administrative lockdown on most activities. Of course, measures of social distancing and health safety guidelines will still hold. Yet neither the lockdown, nor its lifting, was based on a careful analysis of the economic impact of specific activities. Targeted action, opening or closing just the core sectors of economic activity, identified by the intricacies of production chains, could have an almost equivalent economic impact as a general re-opening, yet implicitly limiting health risks. We propose an approach to identify these core activities.
Barba Navaretti, B, G Calzolari, A Dossena, A Lanza and A F Pozzolo (2020), “In and out lockdowns: Identifying the centrality of economic activities”, Covid Economics: Vetted and Real-Time Papers 17.
Barrot, J N, B Grassi and J Sauvagnat (2020), “Sectoral Effects of Social Distancing”, Covid Economics: Vetted and Real-Time Papers 3.
Dietzenbacher, E and M L Lahr (2013), ”Expanding extractions”, Economic Systems Research 25(3): 341-360.
Fadinger, H and J Schimik (2020), “The Costs and Benefits of Home Office during the Covid-19 Pandemic - Evidence from Infections and an Input-Output Model for Germany”, Covid Economics: Vetted and Real-Time Papers 9.
Jackson, M O (2010), Social and economic networks, Princeton University Press.
McKibbin, W and R Fernando (2020) “The economic impact of COVID-19”, In R Baldwin and B Weder di Mauro (eds.), Economics in the Time of COVID-19, a VoxEU eBook, CEPR Press.
Miller, R E and P D Blair (2009), Input-output analysis: foundations and extensions, Cambridge University Press.
Newman, M (2018), Networks, Oxford University Press.
Porter, M (1985), Competitive Advantage: Creating and Sustaining Superior Performance, The Free Press.
Voth, J (2020), “Trade and Travel in the Time of Epidemics”, in R Baldwin and B Weder di Mauro (eds.), Economics in the Time of COVID-19, a VoxEU eBook, CEPR Press.
1 In this work we do not consider the health risks involved in each economic activity, but we strictly focus on their economic impact. We also do not take into account international connections (see Voth (2020) for an analysis of the trade effects of Covid-19) or local effects on regions and provinces.
2 The National Institute of Statistics, Istat, provide IO tables that report the value of intermediate flows of goods among the 63 sectors of the Italian economy, according to the classification of NACE revision 2.
3 Prometeia has classified the entire Italian economy into 12 production chains (“filiere produttive”): Agrifood, Automotive, Home: furniture and design, Shipbuilding and aerospace, Construction and real estate, Energy and utility, Mechanics and Engineering, Fashion & beauty, Health, Media and TLC, Land transport and logistic, Tourism and travel.
4 Using the terminology of social network analysis, each production chain naturally maps into a weighted and possibly directional graph, in which the node is the micro-sector, the link corresponds to a business relationship, the orientation corresponds to the supplier-customer direction, and the weight can be measured by the relevance of the links, such as the value or the number of upstream and downstream connections. See Jackson (2010) and Newman (2018) for a thorough introduction to social network analysis and the methodologies used in this paper.
5 The measure of eigenvector centrality increases with the number of connections (production chain links) and with the centrality of the nodes with which each node is connected. This measure refers to the unrestricted network. Alternative approaches can be considered, see our working paper for a discussion.
6 The graph is obtained with the Python library igraph using the Kamada-Kawai display algorithm. To allow a neater presentation, the graph is first reduced through a maximum spanning tree algorithm (which retains only the strongest links between micro-sectors) and then plotted according to a force directed layout (to spread the sectors dependently on their proximity).