Literature Review

The literature on the economic impact of the COVID-19 pandemic is in a rapid state of expansion. Our study connects with a number of important strands. First, a number of papers such as Dingel and Neiman (2020); Mongey, Pilossoph and Weinberg (2020); Coibion, Gorod-nichenko and Weber (2020) explore the impact of COVID-19 on labor markets. Like Dingel and Neiman (2020), we use data from the Occupational Information Network (O*NET) to inform the model about sectoral supply and demand shocks. Second, some papers such as Goolsbee and Syverson (2020); Chetty, Friedman, Hendren, Stepner and Team (2020); Cavallo (2020); Cox, Ganong, Noel, Vavra, Wong, Farrell and Greig (2020) use real-time data to understand the impact of and recovery from COVID-19. Third, some papers such as Baqaee and Farhi (2020a,b); Barrot et al. (2020); Woodford (2020); Gottlieb, Grobovsek, Poschke and Saltiel (2020), explore the importance of networks and linkages for sectoral shocks and their aggregate consequences. Fourth, papers such as Barrero et al. (2020); Guerrieri, Lorenzoni, Straub and Werning (2020); Krueger, Uhlig and Xie (2020), explore the distinction between the demand and supply component of the COVID-19 shock, and the sectoral reallocation it induces. More closely related with our exercise, a number of papers also explore the consequences of COVID-19 for business failures (Demmou, Franco, Sara and Dlugosch, 2020; Carletti et al., 2020; Acharya and Steffen, 2020; Schivardi and Romano, 2020; Guerini, Nesta, Ragot and Schiavo, 2020). Unlike our approach, these papers do not rely on a structural model of the firm and often consider a simple empirical rule to project cash-flow under COVID-19.⁹ Finally, some papers such as Granja, Makridis, Yannelis and Zwick (2020); Elenev, Landvoigt and Van Nieuwerburgh (2020); Core and De Marco (2020) evaluate the targeting and effectiveness of small business support programs like the PPP in the U.S.

2.1 Supply

The economy consists of S sectors. In each sector s ∈ S there is a mass N_s of firms, indexed by i. Throughout the analysis, we consider the mass of firms in each sector as given. We assume that each firm i in sector s produces according to the following sector-specific production function:

In Eq. (1), y_is denotes gross output, k_is represents any fixed factor, including capital, entrepreneurial talent etc.., n_is is the labor input, while m_is denotes other variable inputs such as materials or intermediate inputs, including output produced by other firms in the same or other sectors. A_s is a sector-specific labor-augmenting productivity so that A_sn_is is the effective labor supply in firm i, while z_is is a firm-specific productivity. Because our analysis is essentially static, we ignore time subscripts. We assume that, regardless of fixed factors, firms need both labor and intermediate goods to produce, so that f_s(., 0, .)= f_s(., ., 0)= 0.

We define the corresponding prices: p_is is the price of output of firm i in sector s, w_s denotes the wage rate per effective unit of labor, r_s is the user cost for fixed factors and p_ms is the price of other variable inputs. Factor prices only vary at the sector level. Prices, both for factors and output are assumed constant in the short run, perhaps because of nominal rigidities.¹⁰

2.2 Demand

Each firm within a sector sells a differentiated variety. We assume a nested CES demand structure, for both final and intermediate uses, of the form:

In Eq. (2), D denotes aggregate demand, D_s is sectoral demand, ζ_s is a sectoral demand shifter, and η is the elasticity of substitution between sectors. For simplicity, we assume that sectors are symmetric before the COVID-19 shock, and set N_sζ_s = 1,∀s. We denote ${\xi }_s^{\text{'}}$ the sectoral demand shifter during COVID-19. For many sectors, we expect ${\xi }_s^{\text{'}}<1$, i.e. sectoral demand falls. This can happen because final demand declines. For instance, the demand for restaurants declines as people are concerned about enclosed spaces. This can happen also because downstream industries are negatively affected and their demand for intermediate inputs declines. For instance, the shutdown of restaurants and open air markets may reduce the demand for fresh produce from local growers.¹¹ For some sectors, demand during the COVID-19 shock may increase, i.e. ${\xi }_s^{\text{'}}>1$. For instance, the demand for some online services or home delivery may increase during confinement.

In turn, sectoral demands D_s satisfy:

where ρ_s is the sector-specific elasticity of substitution between varieties.

From Eqs. (2) and (3), the demand for variety i in sector s is given by Eq. (4):

where P_s denotes the average sectoral price index per unit of expenditure, and P the overall price level. They satisfy:¹²

Because we assume that the price of individual varieties p_is and the mass of firms N_s are constant, sectoral price indices P_s given in Eq. (5) are also constant. The aggregate price index P, however, can change because of the demand shifters ζ_s. We denote with a prime the value of variables during COVID-19 and with a ‘hat’ the ratio of variables between normal and COVID-19 times, e.g. ${\hat{\xi }}_s\equiv {{\xi }^{\prime }}_s/{\xi }_s$. From Eq. (4), we can use exact-hat algebra to express the relative change in demand from normal to COVID-19 times as:

Under the assumption that the pre-COVID-19 equilibrium is symmetric, P_sN_s = PS^1/(η-1), ∀s, and we can write:¹³

Putting the two previous equations together, we obtain a very simple expression for the change in demand under COVID-19:

Eq. (7) indicates that the total change in sectoral demand is a function of two drivers: a relative and an aggregate one. First, sector-specific demand shocks ${\hat{\xi }}_s$ reallocate a given aggregate expenditure across sectors. Importantly, it is the relative pattern of sector-specific demand shocks that matters, not their absolute level. For instance, suppose there is no change in aggregate demand so $\widehat{PD}=1$ and the economy consists of two sectors with ${\hat{\xi }}_s<{\hat{\xi }}_{s^{\prime }}$, then ${\hat{d}}_s<1<{\hat{d}}_{s^{\prime }}:$ one sector is in recession, and the other must be in a boom. The elasticity of substitution across sectors η mediates the sectoral demand shocks £: when goods are very substitutable (high η), small sectoral demand shocks lead to large demand responses. Conversely when demand is very inelastic (low η) demand responds is more similar across sectors (in the limit of η = 0, we obtain $\hat{d}=\widehat{PD}$). Second, for a given pattern of sector-specific demand shocks, all sectors respond proportionately to changes in aggregate demand. For instance, if all sectors are affected uniformly so that ${\hat{\xi }}_s=\hat{\xi },\forall s$, then Eq. (7) indicates that total demand in all sectors is affected uniformly with ${\hat{d}}_{is}=\widehat{PD}$. For future reference, we define $${\tilde{\xi }}_s^{\eta }={\hat{\xi }}_s^{\eta }/({\Sigma }_{\sigma }{\hat{\xi }}_{\sigma }^{\eta }/S).{\tilde{\xi }}_s^{\eta }$$ succinctly summarizes the impact of sector-specific demand shocks on total demand and satisfies ${\Sigma }_s{\tilde{\xi }}_s^{\eta }/S=1$. With this notation, we can rewrite total demand as:

2.3 Modeling the COVID-19 Shock

We model the COVID-19 shock as a flexible combination of supply, productivity, and demand shocks at the sectoral and aggregate level.

First, on the supply side, we assume that fixed factors are immobile. Moreover, only a fraction of workers are allowed to work in each sector. This approach follows Mongey et al. (2020) and Dingel and Neiman (2020). Specifically, consider firm i with pre-COVID employment level n_is. We assume that this firm can only employ up to ${x^{\prime }}_s{n}_{is}$ workers during the COVID-19 shock. Of course, the firm may decide to employ even fewer workers – for instance if demand for its goods declines significantly. Thus, COVID-19 introduces the following labor supply constraint in the firm cost-minimization problem:

where ${n^{\prime }}_{is}$ is the level of employment chosen by the firm during COVID-19.

It is natural to consider that ${x^{\prime }}_s$ varies by sector. For instance, for some essential sectors we may have ${x^{\prime }}_s=\infty$, implying that Eq. (9) never binds. This captures the intuition that workers in these sectors are not sent home.¹⁴ For non-essential sectors, we expect ${x^{\prime }}_s\le 1$. This captures the idea that a firm may retain workers who can work from home as well as a fraction of the current workers who cannot work from home, but may have to lay-off temporarily or permanently any remaining workers. For instance, a university may be able to shift all its courses online, so that ${x^{\prime }}_s\approx 1$. By contrast, a construction company may be able to shift only part of its workers online (project managers, accountants, payroll, HR...) and lay-off temporarily the construction workers. In that case ${x^{\prime }}_s<1$.

In addition, we allow sectoral productivity A_s to change. This reflects the fact that, in the short term, workers who work remotely may perform their duties less efficiently. Some workers may face adjustment costs, others may face additional constraints such as taking care of dependents such as young children. Productivity may decrease even for on-site workers, since COVID-19 introduces additional constraints in the spatial organization of production. We denote ${A^{\prime }}_s$ the labor productivity in sector s during COVID and expect ${A^{\prime }}_s/A_s\le 1$.

On the demand side, we allow for both sectoral and aggregate demand shocks. Relative sector-specific demand shocks are represented by changes in the sectoral demand shifter from ${\xi }_{s}to{{\xi }^{\prime }}_s$. These relative sector-specific demand shocks may represent adjustments in final or intermediate demand patterns as discussed previously. They may also reflect changes in behavior or in policy. For instance, most households may choose to stay away from restaurants during COVID-19, regardless of official instructions (see e.g. Goolsbee and Syverson (2020)). By contrast, some sectors may be shut down because of official policy. For instance, the government may implement a shelter-in-place policy which would imply ${{\xi }^{\prime }}_s=0$ for restaurants, beauty salons and gyms. Our approach encompasses equally well changes in demand that arise from either source. As discussed above, the set of sector-specific demand shifters $\left\{{{\xi }^{\prime }}_s\right\}$ simply redistributes total demand for a given level of aggregate expenditures.

Our approach also allows us to model aggregate demand shocks, measured as the shift in aggregate nominal gross expenditures $\widehat{PD}$. In this paper, we take these changes in aggregate expenditures as exogenous and focus on the implications for business failures. A more ambitious agenda – left for future work – would loop back and derive the change in aggregate demand from more primitive economic forces, taking into account the impact of business failures. We simply note that aggregate demand could change through a variety of channels. First, increased precautionary savings by households and firms may delay spending on consumption or investment. Second, as explored by Guerrieri et al. (2020), the supply shock itself could generate an even larger decline in aggregate demand, when markets are incomplete. Third, as studied by Woodford (2020) in an economy with a ‘circular flow of payments,’ the decline of production (and income) in some sectors of the economy has the potential to dramatically reduce aggregate demand.

We assume that the COVID-19 shock is temporary and maintain the assumption that prices of goods and factors are sticky at that horizon. We also assume that labor cannot reallocate across sectors in the short run, so workers who cannot work for their original place of employment are laid off, either temporarily or permanently. In some countries, like the U.S., these workers may have access to unemployment insurance. In others, such as Germany, the U.K. or France, the government may cover part of the wage bill, allowing the workers to be on a temporary layoff. Either way, we assume that the workers who are not actively employed by the firm are not on the firm’s payroll and generate no drain on its cashflow.¹⁵

Because prices are sticky, firms produce the level of output that is demanded. In Section 6, we consider an extension where firms can optimally ‘mothball,’ i.e. to temporarily shut down, if cash flows are lower under production.

2.4 The Firm’s Cost Minimization Problem

Consider the cost minimization problem of a single firm. For simplicity we omit the firm and sector indices i, s in what follows. We specialize the problem by assuming that the production function f_s(.) is Cobb Douglas:

with the (sector-specific) exponents α, β and γ summing to one.¹⁶

The cost-minimization problem of the firm can be written as:

where d’ is the level of demand faced by the firm, obtained from Eq. (8). The second line indicates that, if the firm produces, it must meet the demand. The third line is the labor supply constraint. We have two cases to consider: when the labor supply constraint doesn’t bind, and when it does.

2.5 Business Failures

To evaluate business failure, we assume that firms follow a simple ‘static’ decision rule – they remain in business as long as their cash balances and their operating cashflow are sufficient to cover their financial expenses. Otherwise, we assume that they are forced to close. Operating cash flow of the firm is defined as:

where the first term represents sales, the other two terms the wage and intermediate input bills, F_is represents any costs associated with fixed factors (rent, utilities, management compensation etc.), including capital costs, r_sk_i,s, and T_is denotes business taxes. The last expression writes operating cash flow in terms of the variable profits, minus payments to fixed factors and taxes. As long as fixed costs and taxes are not affected by COVID-19, we can difference them out by considering the change in cash-flows from CF to CF’, i.e. from the observed to the predicted cash flows.¹⁸ Since the predicted cashflows of the firms depend on whether it is labor constrained, there are two cases to consider.

• Case 1: When the labor supply does not bind $$(\hat{x}>{\hat{x}}^c)$$, the change in cashflow during COVID-19 (compared to the non-COVID scenario) can be expressed using Eq. (13) as:

  $$\begin{array}{cc}C{F}_{is}^{\text{'}}-C{F}_{is}=p_{is}{d}_{is}[{\tilde{\xi }}_s^{\eta }\widehat{PD}-1+(s_{n,is}+s_{m,is}\left)\right(1-\hat{x}\left)\right].& \left(17\right)\end{array}$$

• Case 2: When the labor supply binds $$(\hat{x}>{\hat{x}}^c)$$, the change in cashflow under COVID-19 (compared to the non-COVID scenario) can be expressed using Eq. (15) as:

We assume that the decision to close the firm occurs if there isn’t enough cash, Z_is, to cover operating cash flow minus financial expenses, ιL_is, i.e. if

where the firm’s financial expenses, ιL_is, are defined as interest and principal repayments. Subtracting CF_is from both sides, we obtain:

The term on the right hand side of Eq. (20) can be observed in our firm-level data. The term on the left hand side can be constructed using Eqs. (17) and (18).

The business failure condition Eq. (20) calls for a number of observations. First, while this rule has the advantage of simplicity it assumes that firms with a temporary cashflow shortfall cannot access credit markets and borrow against future profits. To the extent that future profits are sufficiently large, it would be optimal to do so to keep the business afloat. In other words, we are looking at situations where illiquidity turns into insolvency. In our baseline, we estimate bankruptcy rates weekly. This effectively imposes a very tight borrowing constraint: a firm that fails in week t is unable to borrow from cash flows at any later date t'>t regardless of its long term viability. In our view, the focus on liquidity shortfalls is appropriate for SMEs since the immediate danger for small businesses is that they will be forced to shut down in the short run. Our estimates directly get at this issue. In Section 6 we will instead evaluate the business failure condition at the end of the calendar year, even if the length of the COVID-19 related lockdown is much shorter (8 weeks in our baseline). This will allow the firm to smooth cash flow shortages over the calendar year.

A second caveat is that we ignore the role of bankruptcy courts. In theory as long as a business remains viable, the failure to repay creditors in the short run does not mean that it ceases to operate. Instead, business liabilities should optimally be restructured under bankruptcy proceedings. In practice, however, there is substantial variation in bankruptcy regimes across countries. The bankruptcy code may also work more efficiently for large corporations – for instance, over the years, many airlines in the U.S. have continued operating despite undergoing Chapter 11 restructuring – but it is less well suited for SMEs. Moreover, bankruptcy courts in many countries may not be able to efficiently preserve viable businesses in the middle of a pandemic if a wave of small business failures congests the courts. Our estimates should thus be interpreted as the predicted business failures in a scenario where no fresh capital is available and liquidation is the only possible outcome.

There are additional practical reasons why we focus on a liquidity criterion. First, we cannot hope to construct estimates of future revenues and costs at the firm level, which would be important for a solvency criterion. In addition, it is difficult to estimate accurately the initial equity position of SMEs since most are unlisted. In practice, this means that evaluating the equity shortfall that occurs because of COVID-19 is a difficult exercise. Second, we do not have direct information on the firms continued access to credit. There is mounting evidence that many firms responded to the very early phase of the COVID-19 crisis by borrowing (Acharya and Steffen, 2020). However, our understanding is that this was primarily relevant for large firms that increased their cash holdings by drawing upon pre-existing credit lines. Small and medium enterprises typically have more limited access to credit, and at a higher cost (Almeida and Campello, 2013; Almeida and Ippollito, 2014; Gopinath, Kalemli-Özcan, Karabarbounis and Villegas-Sanchez, 2017).

3.1 Firm-Level Data

We use a large firm level data set, Orbis, from BvD-Moody’s. The advantage of Orbis is that BvD collects data from various sources, in particular, national business registries, and harmonizes the data into an internationally comparable format. The Orbis database covers more than 200 countries and over 200 million firms (private and publicly listed), with the longitudinal dimension and representativeness of firms varying from country to country depending on which firms are required to file information with business registries. We report the results for seventeen countries and use 2017 as our base year, as 2018 data is not yet available. The countries included are Belgium, Czech Republic, Finland, France, Germany, Greece, Hungary, Italy, Japan, Korea, Poland, Portugal, Romania, Slovak Republic, Slovenia, Spain, and the United Kingdom.

As depicted in Table 1, we have good coverage of aggregate revenue for the countries in our sample, both for all firms and SMEs specifically. Our coverage ranges from 27 percent in Germany to nearly 66 percent in Finland.¹⁹ Even after imposing additional data requirements for analysis, such as availability of intermediate costs, our data cover more than half of the aggregate revenue of SMEs for most countries – the exceptions are highlighted in grey, where we cover under 40% of aggregate revenue.²⁰

^{Table 1:}

Orbis Coverage (2017)

Notes: OECD revenue (for all firms and SMEs) in 2017 is obtained from the Structural Business Statistics Database. The SBSD provides data for a subset of sectors – for most countries the covered NACE 1-digit sectors are B, C, D, E, F, G, H, I, J, L, M, and N. For Japan, data on revenue are obtained from the Economic Census in 2015 (Orbis data are also for 2015). Only sectors covered in both the OECD (or Census) and Orbis data are used in calculating coverage statistics. To calculate coverage, Orbis revenue (for all firms and SMEs) is summed and divided by the total revenue (for all firms and SMEs) reported by OECD. The coverage rates are computed using cleaned Orbis data. Additional cleaning is done to generate the analysis data, including conditioning on variables needed to compute the bankruptcy condition. Highlighted in grey are countries where the coverage rate falls below 40% in the analysis data. Japan is also highlighted in grey because revenue data on SMEs is not available to evaluate coverage.

^{Table 1:}

Orbis Coverage (2017)

	% of OECD Revenue
	(1) All	(2) SMEs
Belgium	60.4	52.1
Czech Republic	63.4	62.8
Finland	66.0	68.3
France	46.3	46.3
Germany	27.2	17.7
Greece	48.0	48.1
Hungary	63.9	48.7
Italy	63.5	75.8
Japan	42.5	.
Korea	61.9	34.0
Poland	47.5	44.5
Portugal	63.2	72.9
Romania	60.6	40.0
Slovak Republic	52.0	73.2
Slovenia	49.3	61.0
Spain	58.4	69.9
United Kingdom	49.2	41.4

Notes: OECD revenue (for all firms and SMEs) in 2017 is obtained from the Structural Business Statistics Database. The SBSD provides data for a subset of sectors – for most countries the covered NACE 1-digit sectors are B, C, D, E, F, G, H, I, J, L, M, and N. For Japan, data on revenue are obtained from the Economic Census in 2015 (Orbis data are also for 2015). Only sectors covered in both the OECD (or Census) and Orbis data are used in calculating coverage statistics. To calculate coverage, Orbis revenue (for all firms and SMEs) is summed and divided by the total revenue (for all firms and SMEs) reported by OECD. The coverage rates are computed using cleaned Orbis data. Additional cleaning is done to generate the analysis data, including conditioning on variables needed to compute the bankruptcy condition. Highlighted in grey are countries where the coverage rate falls below 40% in the analysis data. Japan is also highlighted in grey because revenue data on SMEs is not available to evaluate coverage.

View Table

^{Table 1:}

Orbis Coverage (2017)

	% of OECD Revenue
	(1) All	(2) SMEs
Belgium	60.4	52.1
Czech Republic	63.4	62.8
Finland	66.0	68.3
France	46.3	46.3
Germany	27.2	17.7
Greece	48.0	48.1
Hungary	63.9	48.7
Italy	63.5	75.8
Japan	42.5	.
Korea	61.9	34.0
Poland	47.5	44.5
Portugal	63.2	72.9
Romania	60.6	40.0
Slovak Republic	52.0	73.2
Slovenia	49.3	61.0
Spain	58.4	69.9
United Kingdom	49.2	41.4

Notes: OECD revenue (for all firms and SMEs) in 2017 is obtained from the Structural Business Statistics Database. The SBSD provides data for a subset of sectors – for most countries the covered NACE 1-digit sectors are B, C, D, E, F, G, H, I, J, L, M, and N. For Japan, data on revenue are obtained from the Economic Census in 2015 (Orbis data are also for 2015). Only sectors covered in both the OECD (or Census) and Orbis data are used in calculating coverage statistics. To calculate coverage, Orbis revenue (for all firms and SMEs) is summed and divided by the total revenue (for all firms and SMEs) reported by OECD. The coverage rates are computed using cleaned Orbis data. Additional cleaning is done to generate the analysis data, including conditioning on variables needed to compute the bankruptcy condition. Highlighted in grey are countries where the coverage rate falls below 40% in the analysis data. Japan is also highlighted in grey because revenue data on SMEs is not available to evaluate coverage.

The evaluate bankruptcy rates, our analysis requires data on firm revenue, wage bill, material cost, number of employees, net income, depreciation, cash stock and financial expenses.²¹ Cash flow is calculated as the sum of net income and depreciation, less financial profits. Sectoral labor and material cost shares are constructed as the revenue weighted average of the firm-level wage bill over revenue and material costs over revenue at the 2-digit NACE level. The analysis focuses on private, non-financial firms.²²

In addition to capturing a high share of aggregate revenue, Table 2 shows that firm failure rates reported by the OECD in normal times are broadly in line with those obtained using Orbis data for many of the high coverage countries.²³ For most countries, business failure rates are higher than those reported by OECD. We suspect the reason for this is the fact that our failure rate calculation assume that all illiquid firms fail, not taking into account access to credit or possible debt restructuring that might allow a firm to continue operating when faced with a liquidity shortfall. Because of these differences in failure rates given by OECD and our estimation, we emphasize changes in the business failure rates before and after COVID-19 instead of levels.

^{Table 2:}

Pre-COVID Business Failure Rates

Notes: Data on firm failure rates are obtained from the OECD’s SDBS Business Demography Indicators. Failure rates are available for a subset of sectors – NACE 1-digit sectors B, C, D, E, F, G, H, I, J, L, M, N, P, Q, R, S. The coverage of sectors varies across countries. Only sectors available in both the OECD and Orbis data are used in calculating failure rates. Sector-level gross value added (GVA) shares in 2017 (OECD) are used for aggregation of both Orbis and OECD data to the country level (the exception is the United Kingdom where 2016 GVA is used because this is the last year in which sector level data are made available for the country). The failure rate comparison is only done for the subset of countries covered in the OECD data.

^{Table 2:}

Pre-COVID Business Failure Rates

	OECD	Orbis
Belgium	3.0	8.3
Czech Republic	7.9	8.3
Finland	5.4	9.3
France	4.7	8.1
Germany	6.7	11.2
Greece	4.0	10.3
Hungary	8.7	8.9
Italy	6.7	9.7
Portugal	11.5	12.5
Romania	8.6	15.5
Slovak Republic	10.0	11.1
Slovenia	3.9	6.5
Spain	7.4	9.2
United Kingdom	13.9	11.4

Notes: Data on firm failure rates are obtained from the OECD’s SDBS Business Demography Indicators. Failure rates are available for a subset of sectors – NACE 1-digit sectors B, C, D, E, F, G, H, I, J, L, M, N, P, Q, R, S. The coverage of sectors varies across countries. Only sectors available in both the OECD and Orbis data are used in calculating failure rates. Sector-level gross value added (GVA) shares in 2017 (OECD) are used for aggregation of both Orbis and OECD data to the country level (the exception is the United Kingdom where 2016 GVA is used because this is the last year in which sector level data are made available for the country). The failure rate comparison is only done for the subset of countries covered in the OECD data.

View Table

^{Table 2:}

Pre-COVID Business Failure Rates

	OECD	Orbis
Belgium	3.0	8.3
Czech Republic	7.9	8.3
Finland	5.4	9.3
France	4.7	8.1
Germany	6.7	11.2
Greece	4.0	10.3
Hungary	8.7	8.9
Italy	6.7	9.7
Portugal	11.5	12.5
Romania	8.6	15.5
Slovak Republic	10.0	11.1
Slovenia	3.9	6.5
Spain	7.4	9.2
United Kingdom	13.9	11.4

Notes: Data on firm failure rates are obtained from the OECD’s SDBS Business Demography Indicators. Failure rates are available for a subset of sectors – NACE 1-digit sectors B, C, D, E, F, G, H, I, J, L, M, N, P, Q, R, S. The coverage of sectors varies across countries. Only sectors available in both the OECD and Orbis data are used in calculating failure rates. Sector-level gross value added (GVA) shares in 2017 (OECD) are used for aggregation of both Orbis and OECD data to the country level (the exception is the United Kingdom where 2016 GVA is used because this is the last year in which sector level data are made available for the country). The failure rate comparison is only done for the subset of countries covered in the OECD data.

An important feature of most economies is the over-sized contribution of SMEs to economic activity, defined as firms with less than 250 employees, play in the economy. As Fig. 1 shows, SMEs account for 53.4% of employment and over 48.9% of payroll, 50.1% of revenue, and 45.6% of total assets across our “good coverage” European countries (Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain).²⁴ It is precisely these SMEs that are most vulnerable to the COVID-19 shock because they tend to be bank-dependent and have limited ability to draw on credit lines. Both of these features make them vulnerable to solvency problems that can follow the liquidity shortage.

Share of SMEs in Employment, Payroll, Revenue, and Total Assets

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Figures depict the SME share of employment (top-left), payroll (top-right), revenue (bottom-left), and total assets (bottom-right). The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain. The SME shares are first calculated at the country level and aggregated across countries using country GVA for weighting.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Share of SMEs in Employment, Payroll, Revenue, and Total Assets

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Figures depict the SME share of employment (top-left), payroll (top-right), revenue (bottom-left), and total assets (bottom-right). The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain. The SME shares are first calculated at the country level and aggregated across countries using country GVA for weighting.

• Download Figure
• Download figure as PowerPoint slide

Share of SMEs in Employment, Payroll, Revenue, and Total Assets

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Figures depict the SME share of employment (top-left), payroll (top-right), revenue (bottom-left), and total assets (bottom-right). The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain. The SME shares are first calculated at the country level and aggregated across countries using country GVA for weighting.

• Download Figure
• Download figure as PowerPoint slide

3.2 Demand and Supply Shocks

In addition to firm-level data, we require information on demand, supply, and productivity shocks. As a first step, we separate sectors, at the 4-digit NACE level, into essential and non-essential sectors based on the U.S. Department of Homeland Security Guidance on the Essential Critical Infrastructure Workforce.²⁵ While the DHS does not provide a list of industry codes that are considered essential, we classify sectors based on the information provided regarding the types of workers and activities considered as part of essential critical infrastructure. Among those workers considered essential are those working in public health, public safety, food supply chain, energy infrastructure, transportation and logistics, critical manufacturing, hygiene products and services, among others.

To measure the sectoral supply shock, $${\hat{x}}_s$$, we follow Dingel and Neiman (2020) and measure the feasibility of remote work by industry. To construct the measure, we start with the “work context” and “generalized work activities” surveys conducted by the Occupational Information Network (O*NET). Following Dingel and Neiman (2020), we classify occupations into those that can be performed remotely versus those that cannot based on characteristics such as reliance on being outdoors, interacting with patients or prisoners, repairing and inspecting structures and equipment, controlling machines, handling and moving objects, among others. We then use information from the U.S. Bureau of Labor Statistics (BLS) on the prevalence of each occupation by NAICS code, along with a cross-walk between NAICS and NACE industry codes to arrive at the fraction of employees that can and cannot perform their work remotely by 4-digit NACE code. To construct the COVID-19 sectoral supply shock $$\left({\hat{x}}_s\right)$$, we assume that no workers in essential sectors lose their jobs, and that in non-essential sectors all workers that cannot work remotely lose their jobs temporarily. Fig. 2 illustrates the severity of the supply shock at the 1-digit NACE level by first averaging 4-digit NACE shocks to the 1-digit level in each country and then using the gross value added sector share of each country to aggregate 1-digit sector shocks across countries. As we would expect, the Accommodation and Food Service and Arts, Entertainment and Recreation sectors are among the most affected, while essential infrastructure sectors including Electricity and Water and Waste remain unaffected.

Supply Shock by Sector

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Depicts the COVID-19 demand shock by 1-digit NACE sector, as the percent change relative to the non-COVID scenario. Supply shocks are first aggregate from the 4-digit NACE to 1-digit NACE by taking a simple average across 4-digit sectors within each country. The gross value added sector share of each country is used to aggregate 1-digit sector shocks across countries. The countries used in aggregation are the the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Korea, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Supply Shock by Sector

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Depicts the COVID-19 demand shock by 1-digit NACE sector, as the percent change relative to the non-COVID scenario. Supply shocks are first aggregate from the 4-digit NACE to 1-digit NACE by taking a simple average across 4-digit sectors within each country. The gross value added sector share of each country is used to aggregate 1-digit sector shocks across countries. The countries used in aggregation are the the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Korea, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain.

• Download Figure
• Download figure as PowerPoint slide

Supply Shock by Sector

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Depicts the COVID-19 demand shock by 1-digit NACE sector, as the percent change relative to the non-COVID scenario. Supply shocks are first aggregate from the 4-digit NACE to 1-digit NACE by taking a simple average across 4-digit sectors within each country. The gross value added sector share of each country is used to aggregate 1-digit sector shocks across countries. The countries used in aggregation are the the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Korea, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain.

• Download Figure
• Download figure as PowerPoint slide

To calculate sector-specific demand, we follow a similar approach. Specifically, we first use the same O*NET surveys to classify occupations based on reliance on face-to-face interactions. We consider occupations as highly reliant on face-to-face interactions when working with external customers or in physical proximity assisting and caring for others, working with the public, and selling to others are deemed important. Using the BLS data and NAICS-NACE crosswalks, we aggregate these occupation-level data to arrive at an estimate of the fraction of employees reliant on face-to-face interactions at the 4-digit NACE level. We assume that demand is unaffected in essential sectors and is one minus the “interaction share” for non-essential industries. We interpret the resulting estimate as a measure of $${\tilde{\xi }}_s^{\eta }$$.²⁶ The last step is to normalize our sectoral demand shocks to be consistent with aggregate demand Eq. (7). This is done by constructing $${\tilde{\xi }}_s^{\eta }={\tilde{\xi }}_s^{\eta }/({\mathrm{\Sigma }}_{\sigma }{\tilde{\xi }}_{\sigma }^{\eta }/S)$$.

Fig. 3 illustrates the size of the sector-specific demand shock $$\left({\tilde{\xi }}_s^{\eta }\right)$$ at the 1-digit NACE level. The figure illustrates that COVID-19 reallocates aggregate expenditure from highly affected non-essential sectors such as Arts, Entertainment, and Recreation to non-affected essential sectors including Water & Waste and Electricity²⁷

Demand Shock by Sector

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Depicts the COVID-19 demand shock by 1-digit NACE sector, as the percent change relative to the non-COVID scenario. Demand shocks are first aggregated from the 4-digit NACE to 1-digit NACE by taking a simple average across 4-digit sectors within each country. The gross value added sector share of each country is used to aggregate 1-digit sector shocks across countries. The countries used in aggregation are the the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Korea, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Demand Shock by Sector

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Depicts the COVID-19 demand shock by 1-digit NACE sector, as the percent change relative to the non-COVID scenario. Demand shocks are first aggregated from the 4-digit NACE to 1-digit NACE by taking a simple average across 4-digit sectors within each country. The gross value added sector share of each country is used to aggregate 1-digit sector shocks across countries. The countries used in aggregation are the the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Korea, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain.

• Download Figure
• Download figure as PowerPoint slide

Demand Shock by Sector

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Depicts the COVID-19 demand shock by 1-digit NACE sector, as the percent change relative to the non-COVID scenario. Demand shocks are first aggregated from the 4-digit NACE to 1-digit NACE by taking a simple average across 4-digit sectors within each country. The gross value added sector share of each country is used to aggregate 1-digit sector shocks across countries. The countries used in aggregation are the the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Korea, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain.

• Download Figure
• Download figure as PowerPoint slide

In addition to the sector-specific demand shock, we also measure changes in aggregate demand $$\left(\widehat{PD}\right)$$ using projections of quarterly and annualized changes in GDP from the International Monetary Fund (IMF).²⁸ What matters for the estimation of bankruptcy rates is the combination of sector-specific demand and aggregate demand shocks, $${\hat{d}}_s={\tilde{\xi }}_s^{\eta }\widehat{PD}$$, which we refer to simply as the total demand shock.

3.3 Productivity shock

To measure the sectoral productivity shock (Â), we start by assuming that sectoral productivity is a combination of the productivity of on-site and remote work, according to:

where all variables vary at the sector level, ω_s is the fraction of on-site workers in total employment, A^work is productivity of workers at work and A^home is productivity of telecommuting workers.

If we assume that A^work and A^home are the same before and during COVID-19 then we can write the ratio Â_s as:

Under the assumption that non-essential industries don’t have onsite workers during the lockdown period, $${\omega }_s^{\prime }=0$$ and this expression collapses to:

We use data from the 2018 American Community Survey (ACS) on the share of remote workers by industry to measure ω_s. Absent any good data on this quantity, we opt to calibrate $$\frac{A^{remote}}{A^{work}}=0.8$$. This implies that  = 0.8 is the maximum decline in productivity possible for a sector where none of the workers worked remotely before COVID-19 to 100% during COVID-19.

3.4 Production Function Parameters

Labor and materials elasticities (β_s and γ_S) are estimated at the 2-digit NACE level for each country.²⁹ Taking into account our modeling assumption that labor and intermediate inputs are variable inputs, and recent critiques of the key identifying assumptions of popular production function estimation techniques, we estimate elasticities as the weighted average of the firm revenue share of input expenditures (e.g., labor cost share of revenue and material cost share of revenue), where the weights are given by firm revenue.³⁰ Due to the lack of price data, the elasticities we estimate are revenue, rather than output, elasticities.

4.1 Cross-Sector Heterogeneity

In our model, the rise in bankruptcy rates under COVID-19 is driven by the deterioration in firms’ cashflow. This deterioration, in turn, is largely driven by the total demand, supply, and productivity shocks under COVID-19. Therefore, to understand why aggregate baseline bankruptcy rates increases by 9 percentage points under COVID-19, it is important to first understand heterogeneity in sectoral vulnerability to the crisis.

As for the previous table, Table 4 reports the non-COVID (col. 1) and COVID-19 (col. 2) bankruptcy rates and the difference between the two (∆, col. 3). Given both the customer orientation and limited scope of remote work, some service sectors (Accommodation & Food Service, Arts, Entertainment & Recreation) experience an increase in bankruptcy rates under COVID-19 exceeding 20 percentage points. In fact, among the top five most affected sectors, the majority belong to the service sector.³⁴ In stark contrast, sectors with a high fraction of essential sub-sectors, with no sectoral supply shocks and higher sector-specific demand, including Agriculture, Health, and Water & Waste, are the least impacted by COVID-19, with bankruptcy rates rising less than 5 percentage point. Sectors with fewer essential workers, but relatively low total demand shocks (Manufacturing) and high scope for remote work (Information & Communications, Professional, Scientific & Technical Activities) are moderately effected, experiencing a rise in bankruptcy rates under 10 percentage points.

^{Table 4:}

Sector SME Bankruptcy Rates

Notes: Sector bankruptcy rates are first calculated at the 1-digit NACE level for each country, and then aggregated across countries using (country x sector) gross value added from the OECD as weights. The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Sloveian, and Spain.

^{Table 4:}

Sector SME Bankruptcy Rates

	(1) Non-COVID	(2) COVID-19	(3) Δ
Agriculture	9.44	13.52	4.08
Mining	12.50	36.03	23.54
Manufacturing	8.48	16.73	8.25
Electric, Gas & Air Con	9.35	11.31	1.96
Water & Waste	6.72	9.65	2.93
Construction	7.97	10.19	2.21
Wholesale & Retail	9.12	18.21	9.10
Transport & Storage	7.64	13.28	5.63
Accom. & Food Service	13.15	38.59	25.44
Info. & Comms	10.00	15.92	5.92
Real Estate	11.61	17.38	5.76
Prof., Sci., & Technical	10.24	18.85	8.60
Administration	8.32	19.39	11.06
Education	10.86	30.04	19.18
Health & Social Work	7.74	11.22	3.48
Arts, Ent., & Recreation	12.95	36.55	23.60
Other Services	12.80	31.42	18.62

Notes: Sector bankruptcy rates are first calculated at the 1-digit NACE level for each country, and then aggregated across countries using (country x sector) gross value added from the OECD as weights. The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Sloveian, and Spain.

View Table

^{Table 4:}

Sector SME Bankruptcy Rates

	(1) Non-COVID	(2) COVID-19	(3) Δ
Agriculture	9.44	13.52	4.08
Mining	12.50	36.03	23.54
Manufacturing	8.48	16.73	8.25
Electric, Gas & Air Con	9.35	11.31	1.96
Water & Waste	6.72	9.65	2.93
Construction	7.97	10.19	2.21
Wholesale & Retail	9.12	18.21	9.10
Transport & Storage	7.64	13.28	5.63
Accom. & Food Service	13.15	38.59	25.44
Info. & Comms	10.00	15.92	5.92
Real Estate	11.61	17.38	5.76
Prof., Sci., & Technical	10.24	18.85	8.60
Administration	8.32	19.39	11.06
Education	10.86	30.04	19.18
Health & Social Work	7.74	11.22	3.48
Arts, Ent., & Recreation	12.95	36.55	23.60
Other Services	12.80	31.42	18.62

Notes: Sector bankruptcy rates are first calculated at the 1-digit NACE level for each country, and then aggregated across countries using (country x sector) gross value added from the OECD as weights. The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Sloveian, and Spain.

Because bankruptcy rates are calculated at a weekly frequency, we are able to evaluate the evolution of key variables to better understand the sources of cross-sector heterogeneity in bankruptcy rates under COVID-19. Fig. 4 depicts the evolution of six variables – the increase in bankruptcy rates due to COVID-19 (top left), average firm cash balance (top right), sector-specific demand shocks (middle left), total demand shocks (middle right), sectoral supply shocks (bottom left), and fraction of firms constrained (bottom right).^35,36

Weekly Evolution (Sector)

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Figures show the weekly evolution of six key variables – the increase in bankruptcy rates under COVID-19 (top left), average firm cash balance (top right), sector-specific demand shock (middle left), total demand shock (interaction between sector-specific demand and aggregate demand shock, middle right), sectoral supply shock (bottom let), and fraction of firms labor supply constrained (bottom right). In each week, sector variables are first calculated at the 1-digit NACE level for each country, and then aggregated across countries using (country x sector) gross value added from the OECD as weights. The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Weekly Evolution (Sector)

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Figures show the weekly evolution of six key variables – the increase in bankruptcy rates under COVID-19 (top left), average firm cash balance (top right), sector-specific demand shock (middle left), total demand shock (interaction between sector-specific demand and aggregate demand shock, middle right), sectoral supply shock (bottom let), and fraction of firms labor supply constrained (bottom right). In each week, sector variables are first calculated at the 1-digit NACE level for each country, and then aggregated across countries using (country x sector) gross value added from the OECD as weights. The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain.

• Download Figure
• Download figure as PowerPoint slide

Weekly Evolution (Sector)

Citation: IMF Working Papers 2020, 207; 10.5089/9781513557748.001.A001

Notes: Figures show the weekly evolution of six key variables – the increase in bankruptcy rates under COVID-19 (top left), average firm cash balance (top right), sector-specific demand shock (middle left), total demand shock (interaction between sector-specific demand and aggregate demand shock, middle right), sectoral supply shock (bottom let), and fraction of firms labor supply constrained (bottom right). In each week, sector variables are first calculated at the 1-digit NACE level for each country, and then aggregated across countries using (country x sector) gross value added from the OECD as weights. The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain.

• Download Figure
• Download figure as PowerPoint slide

Water & Waste (sector E) is only mildly affected by the crisis, experiencing a small uptick in bankruptcy rates during the lockdown and quickly leveling off at just under 10% through the rest of the year (top left). This minimal impact is due to the fact that most sub-sectors of sector E are classified as essential. Consequently, during the lockdown, the sector does not experience a labor supply shock (bottom left), and benefits from a reallocation of demand (middle right). During the post-lockdown period, the demand for sector E is redistributed back toward other sectors in the economy, causing total demand for this sector to fall.

More interesting is the comparison between a highly affected sector (Arts, Entertainment & Recreation, sector R) and a moderately affected one (Wholesale & Retail, sector G). The average cash balance (top right) of each sector prior to the onset of the COVID-19 crisis reflects firms’ initial liquidity position. The fact that the average cash balance in sector G exceeds that of sector R indicates that sector R enters the crisis in a more precarious liquidity position than sector G, and helps explain the higher non-COVID bankruptcy rates in the former sector. As the crisis proceeds, and more firms go bankrupt in sector R, its average cash balance deteriorates relatively more than it does in sector G.

Although non-COVID bankruptcy rates (top left) in sector R exceed those of sector G, the gap between the two expands substantially, especially during the lockdown period. The core reason for this divergence is the severity of the sector-specific demand shock (middle left) in sector R, relative to sector G. Notice that the total demand shocks (middle right) are more severe than the sector-specific ones because all countries face negative aggregate demand shocks during the COVID-19 crisis. Importantly, at first glance, it appears that sector R faces a substantially more severe sectoral supply shock (bottom left) than sector G. However, the degree to which this shock impacts the sector is determined by the fraction of firms that become labor supply constrained (bottom right). Under 10% of firms in sector R are constrained by the near 80% decline in labor supply while over 20% of firms in sector G are constrained by the 30% decline in labor supply. The reason why the labor constraint binds for a smaller fraction of firms in sector R is because the sector faces a severe total demand shock that lowers the optimal labor demand of most firms below the labor supply constraint (i.e., the labor constraint does not bind). In contrast, firms in sector G face a more severe sectoral labor supply shock than demand shock, which results in optimal labor demand exceeding the labor constraint for over 20% of firms. Ultimately, the crisis is felt more severely in Arts, Entertainment & Recreation, which results in bankruptcy rates reaching 36.5% by the end of the year, compared to 18.2% in Wholesale & Retail.

Fig. 4 highlights the importance of sectoral shocks and their interactions in driving cross-sector heterogeneity in COVID-19 bankruptcy rates. To further decompose the effects of different shocks, Table 5 evaluates changes in bankruptcy rates (COVID minus non-COVID) under five alternative scenarios that differ in the composition of shocks. The first column only includes the aggregate demand shock $$\left(\widehat{PD}\right)$$. The second column includes both aggregate demand and sectoral supply shocks $$(\widehat{PD},{\hat{x}}_s)$$. The third includes total demand shocks $$(\widehat{PD},{\tilde{\xi }}_s^{\eta })$$, which incorporate both aggregate demand and sector-specific demand shocks. The fourth includes total demand and sectoral supply shocks $$(\widehat{PD}{\tilde{\xi }}_s^{\eta },{\hat{x}}_s)$$. The last is our baseline, which adds sectoral productivity shocks to column (4).

^{Table 5:}

∆ Bankruptcy Rate Comparison (Alternative Shock Combinations)

Notes: The table reports the change in bankruptcy rates (COVID-19 – non-COVID) under 5 alternative scenarios – aggregate demand shock only $$\widehat{PD}$$ both aggregate demand and sectoral supply shocks $$(\widehat{PD},{\hat{x}}_s)$$ both aggregate demand and sector-specific demand shocks $$\left(\widehat{PD}{\tilde{\xi }}_s^{\eta }\right)$$; total demand and supply shocks $$(\widehat{PD}{\tilde{\xi }}_s^{\eta },{\hat{x}}_s)$$; and the baseline $$(\widehat{PD}{\tilde{\xi }}_s^{\eta },{\hat{x}}_s,{\hat{A}}_s)$$. Sector changes in bankruptcy rates are first calculated at the 1-digit NACE level for each country and then aggregated across countries using (country x sector) gross value added from the OECD as weights. The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain. The last row is the sector GVA weighted average.

^{Table 5:}

∆ Bankruptcy Rate Comparison (Alternative Shock Combinations)

	(1) $$\widehat{PD}$$	(2) $$\widehat{PD},{\hat{x}}_s$$	(3) $$\widehat{PD}{\tilde{\xi }}_s^{\eta }$$	(4) $$(\widehat{PD}{\tilde{\xi }}_s^{\eta },{\hat{x}}_s)$$	(5) Baseline
Agriculture	0.82	2.89	1.01	3.82	4.08
Mining	0.61	19.05	1.17	19.73	23.54
Manufacturing	1.00	6.11	0.95	6.56	8.25
Electric, Gas & Air Con	1.98	1.98	1.94	1.94	1.96
Water & Waste	3.33	3.33	2.73	2.73	2.93
Construction	1.48	1.77	2.00	2.00	2.21
Wholesale & Retail	1.65	4.83	8.84	8.73	9.10
Transport & Storage	6.83	7.00	5.13	5.16	5.63
Accom. & Food Service	0.07	75.04	9.20	20.04	25.44
Info. & Comms	2.12	3.56	4.99	4.99	5.92
Real Estate	1.60	2.18	5.76	5.71	5.76
Prof., Sci., & Technical	3.25	3.88	7.35	7.48	8.60
Administration	3.67	5.87	10.61	10.72	11.06
Education	2.01	49.45	18.60	18.60	19.18
Health & Social Work	1.85	12.16	3.14	3.14	3.48
Arts, Ent., & Recreation	1.92	51.04	18.92	21.27	23.60
Other Services	0.13	47.27	16.90	17.62	18.62
Average	2.02	11.78	6.19	7.75	8.63

Notes: The table reports the change in bankruptcy rates (COVID-19 – non-COVID) under 5 alternative scenarios – aggregate demand shock only $$\widehat{PD}$$ both aggregate demand and sectoral supply shocks $$(\widehat{PD},{\hat{x}}_s)$$ both aggregate demand and sector-specific demand shocks $$\left(\widehat{PD}{\tilde{\xi }}_s^{\eta }\right)$$; total demand and supply shocks $$(\widehat{PD}{\tilde{\xi }}_s^{\eta },{\hat{x}}_s)$$; and the baseline $$(\widehat{PD}{\tilde{\xi }}_s^{\eta },{\hat{x}}_s,{\hat{A}}_s)$$. Sector changes in bankruptcy rates are first calculated at the 1-digit NACE level for each country and then aggregated across countries using (country x sector) gross value added from the OECD as weights. The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain. The last row is the sector GVA weighted average.

View Table

^{Table 5:}

∆ Bankruptcy Rate Comparison (Alternative Shock Combinations)

	(1) $$\widehat{PD}$$	(2) $$\widehat{PD},{\hat{x}}_s$$	(3) $$\widehat{PD}{\tilde{\xi }}_s^{\eta }$$	(4) $$(\widehat{PD}{\tilde{\xi }}_s^{\eta },{\hat{x}}_s)$$	(5) Baseline
Agriculture	0.82	2.89	1.01	3.82	4.08
Mining	0.61	19.05	1.17	19.73	23.54
Manufacturing	1.00	6.11	0.95	6.56	8.25
Electric, Gas & Air Con	1.98	1.98	1.94	1.94	1.96
Water & Waste	3.33	3.33	2.73	2.73	2.93
Construction	1.48	1.77	2.00	2.00	2.21
Wholesale & Retail	1.65	4.83	8.84	8.73	9.10
Transport & Storage	6.83	7.00	5.13	5.16	5.63
Accom. & Food Service	0.07	75.04	9.20	20.04	25.44
Info. & Comms	2.12	3.56	4.99	4.99	5.92
Real Estate	1.60	2.18	5.76	5.71	5.76
Prof., Sci., & Technical	3.25	3.88	7.35	7.48	8.60
Administration	3.67	5.87	10.61	10.72	11.06
Education	2.01	49.45	18.60	18.60	19.18
Health & Social Work	1.85	12.16	3.14	3.14	3.48
Arts, Ent., & Recreation	1.92	51.04	18.92	21.27	23.60
Other Services	0.13	47.27	16.90	17.62	18.62
Average	2.02	11.78	6.19	7.75	8.63

Notes: The table reports the change in bankruptcy rates (COVID-19 – non-COVID) under 5 alternative scenarios – aggregate demand shock only $$\widehat{PD}$$ both aggregate demand and sectoral supply shocks $$(\widehat{PD},{\hat{x}}_s)$$ both aggregate demand and sector-specific demand shocks $$\left(\widehat{PD}{\tilde{\xi }}_s^{\eta }\right)$$; total demand and supply shocks $$(\widehat{PD}{\tilde{\xi }}_s^{\eta },{\hat{x}}_s)$$; and the baseline $$(\widehat{PD}{\tilde{\xi }}_s^{\eta },{\hat{x}}_s,{\hat{A}}_s)$$. Sector changes in bankruptcy rates are first calculated at the 1-digit NACE level for each country and then aggregated across countries using (country x sector) gross value added from the OECD as weights. The aggregation is done over the high coverage group, which includes Belgium, Czech Republic, Finland, France, Greece, Hungary, Italy, Poland, Portugal, Romania, Slovakia, Slovenia, and Spain. The last row is the sector GVA weighted average.