A regional input-output model of the COVID-19 crisis in Italy: decomposing demand and supply factors

Abstract

We propose an empirically estimated inter-regional input-output model of the Italian economy designed for COVID-19 impact assessment, intended as a tool for public authorities facing comparable adverse events and requiring timely estimates of sectoral and regional economic impacts. We evaluate the contributions of demand- and supply-side factors to output losses in Italy during the pandemic, providing insights on the suitability of demand- and supply-side policies. Supply-side shocks, as a consequence of mandated closures, are the primary driver of output losses only during the nationwide lockdown of spring 2020. During the following stages, changes in final demand due to income losses and changes in mobility play a pivotal role at the aggregate, regional, and sectoral levels. While this result supports demand-side policies, the efficacy of such policies may be hampered when final consumption demand is low chiefly due to reduced mobility rather than income losses.

Keywords:

• Input-output
• COVID-19
• lockdown
• Italy
• demand and supply shocks

JEL Codes:

• C63
• C67
• D57
• E17
• I18
• R15

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data Availability Statement

The open-source files necessary to reproduce the simulations shown in the paper can be accessed at https://github.com/SReissl/CovidIO2. The link also contains instructions on how to request access to the proprietary IO tables used.

Notes

1 For example, a shift in the spending of Italian families towards Food, Beverages and Tobacco likely explains that industrial production in this sector was left almost unaffected by the COVID-19 crisis, despite the negative macroeconomic outlook and the lower demand from downstream sectors affected by closures. Conversely, accommodation services such as hotels were never mandated to close. Nonetheless, they experienced a dramatic decline in activity levels since consumers' mobility plummeted, initially due to mandated mobility restrictions and subsequently autonomous precautionary responses (cf. Alexander & Karger, 2021).

2 The latter represented a non-negligible part of the fiscal packages implemented by governments in the early phases of the pandemic. For instance, the ‘Relaunch Decree’ approved by the Italian government in May 2020 included several tax breaks and monetary transfers such as the ‘Mobility bonus’, providing discounts and tax breaks up to €750 on the purchase of goods and services connected to green mobility, the €500 ‘Holiday Bonus’ to stimulate the domestic tourism sector, and the ‘Bonus 110%’, intended to relaunch the construction sector by entitling home-owners to a tax credit of up to 110% on the cost of renovations aimed at upgrading the energy-efficiency of dwellings.

3 Pichler et al. (2020) employ a modified Leontief production function featuring a distinction between critical and non-critical inputs.

4 In the equation shown, we purposely abstract from the presence of imported inputs (which do exist in the model) for simplicity. As is often done in the literature applying IO models to disaster analysis (see, for example, Hallegatte, 2008), we assume that all imported inputs demanded are always delivered, such that no supply constraints can arise along this dimension.

5 The procedure for the derivation of the labor shocks is described in Appendix B, with additional details given in Section 1 of the online supplement to this article. It combines a systematic tracking of government lockdown decrees with regional data on sectoral employment at the 5-digit ATECO level to map the sectors affected by decrees into the 32 sectors of our model. Survey data on occupations and tasks are employed to assess the share of teleworkable occupations within each sector, following Cetrulo et al. (2020). The idea of depicting lockdown shocks as changes in labor availability is common in the literature using IO-frameworks to analyze the COVID-19 crisis, though the precise approaches to specifying or estimating these shocks differ substantially (e.g. Baqaee & Farhi, 2021b; Barrot et al., 2021; del Rio-Chanona et al., 2020; Guan et al., 2020; Haddad et al., 2021b; Pichler & Farmer, 2021; Pichler et al., 2021).

6 https://data.humdata.org/dataset/movement-range-maps

7 The index, therefore, takes a positive value if mobility on a given day exceeds the baseline and a negative value in the opposite case.

8 Mobility indexes were chosen over epidemiological indicators such as new infections, new fatalities, and the Rt index – with all of which we also experimented – for several reasons. The Rt index is a leading indicator of the future evolution of contagion rather than an indicator of the current (perceived) severity of the epidemic. During the late stages of the spring 2020 nationwide lockdown in Italy, for example, a low Rt coexisted with dramatically high numbers of hospitalizations and deaths for an extended period. In addition, Rt, at times, behaves erratically, particularly when infection numbers are low. The use of infection data implies other problems, as recorded new infections also depend on the amount of tests carried out and the efficacy of the tracing system. The peak of new infections during the first wave of the epidemic in Italy was far lower than during the second wave. The peaks of new fatalities instead suggest that the two waves were quite similar in terms of severity. Finally, new fatalities are limited in their ability to provide a picture of the current severity of the epidemic in that they substantially lag other indicators such as new infections or the mobility index.

9 Using a predefined weighting matrix is a common practice employed in both GMM and MSM. As argued by Cochrane (2005, p.199), the use of such a predefined matrix allows ‘to emphasize economically interesting results’ and obtain estimates that ‘may give up something in asymptotic efficiency, but […] are still consistent, and […] can be more robust to statistical and economic problems.’

10 Declines of this or even greater magnitude were frequently observed across all regions during the pandemic.

11 All remaining parameters are set to their estimated values provided in Table 1.

12 Recall that the mobility- and output-related effects are not additive, as the level of consumption demand in each region is determined by the stronger of the two (see Section 2.4).

13 Under the zone-based regional system, different extents of restrictions were automatically enforced at the regional level depending on a predefined set of epidemiological indicators.

Additional information

Funding

F. Lamperti acknowledges support from the H2020 project Growth Welfare Innovation Productivity (GROWINPRO), grant agreement No. 822781.