August 2020 · Number 9D The Impact of COVID-19 on Formal Firms: Evidence from Ethiopia Pierre Bachas, Anne Brockmeyer, Tom Harris, Camille Semelet1 S UMMARY almost all firms in the highly-impacted sectors register losses. This note uses administrative tax data on firms to measure The corporate income tax revenue loss is severe and in 2020 the direct impact of the COVID-19 containment and preven- would only collect 75.0% of its baseline.2 In addition, firms tion measures (referred to as ‘lockdown’ throughout this note) accumulate losses equivalent to 0.6% of GDP, suggesting that on firms’ profitability, employment and exit rates. It sepa- firms will need to substantially increase borrowing to survive. rates the economy in three sectors, which face different size Firms would cut 3.3% of total yearly payroll - wage subsidies shocks and considers two lockdown scenarios: one lasting can save a substantial share of payroll in the medium-impact three months and one lasting five months. The three-month sector, but will not be able to save employment in the high- lockdown scenario aligns, to some extent, with the strictest pe- impact sector (tourism, transport, personal services), where riod of Ethiopia’s containment measures - which were scaled- firms can’t pay their fixed costs. up in April 2020 (at the start of the five-month State of Emer- gency) and partially eased from June 2020. The five-month This note faces important limitations: (i) it does not in- lockdown scenario on the other hand could reflect the full pe- clude the indirect impacts of the shocks which operate through riod of containment and prevention measures. The simulations firms’ trade linkages, (ii) it only models a demand shock and estimate losses to corporate income tax revenue, increases in as such firms have no issues obtaining inputs (materials, labor), firms’ debt levels, cuts in payroll and their mitigation through (iii) Firms do not adapt to the crisis (for example by chang- wage subsidies, and aggregate output losses from firms’ exit. ing products, selling online etc.). Given these limitations, the numbers in this report should be considered as plausible lower Overall, the impact on the economy is severe, with large bounds arising from direct effects, in partial equilibrium. Dy- falls in tax revenue, increases in debt and decreases in wages namic general equilibrium models of the economy, with link- and/or employment. Under a three-month lockdown scenario, ages across sectors and firms, are needed to gauge longer term we estimate that only 53.9% of firms remain profitable and that effects. 1 Pierre Bachas: World Bank Research, pbachas@worldbank.org; Anne Brockmeyer: Institute for Fiscal Studies, University College London and World Bank, abrockmeyer@worldbank.org; Tom Harris: Institute for Fiscal Studies TaxDev Center, tom.harris@ifs.org.uk; Camille Semelet: World Bank Research, csemelet@worldbank.org. We thank Sarthak Agrawal and Edris Seid for excellent support. The findings and conclusions are those of the authors; they do not represent the views of the World Bank, its member countries or the countries mentioned in this study. We are grateful to the Ethiopian Ministry of Revenues for providing the data used in this study. We thankfully acknowledge funding by the World Bank through the Knowledge of Change Trust Fund and the Fiscal Policy and Sustainable Growth Unit, and by UKAID through the Centre for Tax Analysis in Developing Countries (TaxDev). 2 In this scenario, all revenue losses would affect collections for the 2019/20 fiscal year - which ran from 8 July 2019 to 7 July 2020. 1 L OCKDOWN S IMULATIONS AND C ATEGORIZATION tant to reduce their labor costs as re-contracting is costly and OF S ECTORS BY I MPACT cannot adjust their fixed costs. Finally, we assume that credit constraints prevent borrowing beyond existing loans used to The COVID19 (coronavirus) pandemic and associated cover predictable losses (i.e. losses unrelated to the shock). containment measures are expected to cause far-reaching damage to economies around the world. Firms are suffer- We classify sectors into three impact categories - high, ing from reduced demand due to movement restrictions, from medium and low – depending on their expected loss in rev- reduced labor supply and from constraints to sourcing mate- enue during the shutdown, displayed in Table 1. This classi- rial inputs. The breakup of otherwise healthy businesses in fication is based on a country-specific ad hoc assessment devel- response to a temporary shock implies large social costs. Gov- oped with input from government officials. In the high-impact ernments are therefore intent on designing emergency policies category are sectors which are severely impacted by the lock- to keep businesses afloat. down and lose 60% of their revenue. These include tourism, transportation, non-essential retail and entertainment. In the We present simulations using firm-level tax records medium impact categories are sectors which are moderately from Ethiopia, which vary the duration of the lockdown3 impacted and lose 40% of their revenue. These include man- and the relative impact across sectors. In these simulated ufacturing and education. Finally, the low impact sector only scenarios, demand shocks induce a loss in revenue which trig- loses 20% of its monthly revenue, in sectors such as essential gers a cut in profitability and possibly cuts in employment4 or retail, health, construction and agriculture. Naturally there is even firm closure. We compare these simulations to a baseline still a fair degree of heterogeneity of exposure within the cate- (pre-COVID) situation, which corresponds to the last year of gories, with some sub-sectors experiencing increased revenue. available administrative data. Our analysis relies on a few sim- Table 2 shows the number of firms and economic weight of ple assumptions about the structure of firms’ revenue and costs: each of the three impact sectors: the high-impact sector con- we assume that firms aim to weather the shock such that they tains 9% of the firms and 12% of the wage bill, the medium can scale their production capacity back up swiftly at the end impact sector contains 38% of the firms and 61% of the wage of the lockdown.5 In this stylized world, firms can reduce their bill, and the low-impact sector the remaining 54% of the firms material costs proportionally to the drop-in demand, are reluc- and 27% of the wage bill. Table 1: Sector Categories and Shocks Sectors Expected Monthly Categories (e.g., detailed list of sectors in Appendix Table 4) Revenue Loss Accommodation and Food Service Activities, Transport, High Impact and other highly affected sectors 60% Medium Impact Non-essential Retail, Education and other moderately affected sectors 40% Agriculture, Human Health and Social Work activities Low Impact and other mildly affected sectors 20% 3 We use the term lockdown for ease of reference and to maintain consistency with our other country notes on this topic. However, we do acknowledge that the containment and prevention measures implemented in Ethiopia have been more relaxed than those in other countries, and that in this context there has been no official ‘lockdown’. 4 While the Government of Ethiopia has made it illegal to fire anyone during the State of Emergency, we assume one could still see reductions in employment on the intensive margin (e.g. reductions in hours offered to shift workers, or employees forced to stay at home without pay). 5 We do, however, plan to extend this note in the future and present simulations in which the economy does not transition to 100% of its previous activities after coming out of lockdown. 2 Table 2: Statistics for High, Medium and Low Impact Sectors Aggregates Averages Wage Avg. size Avg. Labor Material Fixed Number Share Revenue Categories of firms of firms share bill (LCU, in Profit costs (% costs (% costs (% share millions) margin total cost) total cost) total cost) High impact 974 9% 7% 12% 37 13% 32% 39% 29% Medium impact 4236 38% 69% 61% 79 13% 25% 43% 31% Low impact 6011 54% 24% 27% 19 11% 31% 39% 30% E FFECT ON F IRMS ’ P ROFITABILITY main constant. The results are displayed in Figure 1, and show that in the high and medium impact sectors the vast majority In this section, we ask what share of firms would need of firms become unprofitable even under the three-month lock- government support to “stay afloat” under a three-month down scenario. In our simulations, as we use annual data, the and a five-month lockdown scenario. Assuming credit con- five-month lockdown scenario could represent: a partial lock- straints, a rough indication for firms’ ability to stay afloat is down lasting another two months, the re-imposition of a partial a non-negative profit rate. We start by simulating scenarios lockdown for two months later in the year, or the implementa- where firms lose a share of their revenue, while all costs re- tion of a shorter period of complete lockdown.6 Figure 1: Firm Profitability Under a Shock to Revenue, No Adjustment to Costs (a) 60% Revenue loss (b) 40% Revenue loss (c) 20% Revenue loss Note: These figures show the distribution of profitability, at baseline, and assuming that firms face a loss in revenue corresponding to either three or five months of loss in yearly revenue. They show the distributions holding all costs constant. In addition to a pure revenue shock, we simulate a more high impact sector. On aggregate, only 54% (44%) of all firms realistic scenario where firms adjust their material costs remain profitable under a three-month (five-month) lockdown. proportionally to their revenue loss. The results are dis- We also observe that the distribution becomes multi-modal for played in Figure 2: 75% of firms in the high-impact sector high impact firms: while firms using mainly material inputs are profitable at baseline, a number which drops to 42% for and little labor or capital inputs can adjust to some extent and the three-month lockdown scenario and to 27% under a five- limit their losses, firms with a small share of material inputs month lockdown. The impact is less severe in the medium and in total cost have little margin to adjust and suffer much larger low impact sectors, since the shock they face is less severe and losses. since these sectors rely more heavily on material inputs than the 6 Itis worth highlighting that in the five-month scenario, while the majority of the revenue losses would affect collections in the 2019/20 fiscal year, a portion of these might in fact affect the 2020/21 fiscal year - depending on the scenario one has in mind. 3 Figure 2: Firm Profitability Under a Shock to Revenue, Material Costs Adjust in Proportion (a) 60% Revenue loss (b) 40% Revenue loss (c) 20% Revenue loss E FFECT ON E MPLOYMENT AND S IMULATIONS OF firms which have to cut their wage bill proportionally to the WAGE S UBSIDIES shock in an attempt to stay afloat. In the middle of the distri- In this section, we study by how much employers would bution, a share of firms reduces their wage bill somewhat (but need to slash their yearly wage bill in the absence of gov- less than proportionally to the shock) and achieves zero profit ernment support. We continue to assume that material inputs (or retains to pre-shock projected losses): providing even mod- adjust first, and that firms only cut their wage bill if they are est wage subsidies to these firms has the potential to save jobs. still unprofitable after the material inputs adjustment. Figure 3 On aggregate, weighting by firms’ yearly wage bill, this would shows the resulting distributions of the reduction in the yearly lead to a cut in payroll of 3.3% (resp. 6.6%) of the formal wage bill for a three or five month lockdown scenario. The fig- economy’s total yearly wage bill in the three-month lockdown ure is bi-modal: the first spike corresponds to firms which are [resp. five-month]. The payroll loss is of course concentrated sufficiently profitable at baseline: they absorb the shock and in the high-impact sectors which would cut 9.2% (resp. 17.2%) keep paying their workers. The second spike corresponds to of payroll under the three-month lockdown (resp. five-month). Figure 3: Wage Bill Reduction from a Revenue Shock, Material Costs Adjust Proportionally (a) 60% Revenue loss (b) 40% Revenue loss (c) 20% Revenue loss To counteract these payroll losses, the government might consider offering wage subsidies to firms in order 7 As noted above, while the government has made it illegal to fire anyone during the State of Emergency, one could still see reductions in employment on the intensive margin (e.g. reductions in hours offered to shift workers, or employees forced to stay at home without pay). 8 It is worth noting that the four-month waiver of employment income tax for particular workers (those required to stay at home) may have assisted in this respect. 4 to protect wage levels and formal employment.78 Figure have to pay their fixed costs (e.g. rent) and a reduction in la- 4 shows each sector’s aggregate payroll losses when varying bor costs is not sufficient to counteract the revenue loss. On the the size of the wage subsidy, measured as the share of firms’ other hand, wage subsidies can save payroll for the low, and es- payroll paid by the government. In the case of a zero-wage pecially the medium-impact sector: in the latter sector, a 60% subsidy the loss in payroll corresponds to the numbers men- wage subsidy over the lockdown period would roughly halve tioned above. As the wage subsidy increases the loss in payroll the sector’s payroll loss. On aggregate, applying a 50% wage decreases, as some firms now return to zero profits (or to their subsidy across all sectors would reduce the yearly payroll loss baseline losses). The impact on payroll loss is however very from 3.3% to 2.7% (three-month lockdown) or from 6.6% to different across the three impact sectors: On the one hand, for 5.0% (five-month lockdown). It would take a substantial sub- the high impact sectors (Figure 3a), the loss in revenue is too sidy to save more payrolls: even with a 90% wage subsidy the severe to be compensated by wage subsidies and these firms loss in yearly payroll would be reduced only to 2.4% (three- are forced to cut employment, even for large wage subsidies. month lockdown) or to 4.2% (five-month lockdown). To understand this, note that we assume that these firms still Figure 4: Aggregate Sector Loss in Payroll as a function of the Size of the Wage Subsidy (a) 60% Revenue loss (b) 40% Revenue loss (c) 20% Revenue loss Note: These figures show to what extent a government wage subsidy for the retained labor force can absorb the aggregate loss in payroll, if the lockdown lasts three or five months. Firms readjust their decision after receiving a wage subsidy: they first adjust their material costs, and then their wage bill. It is still assumed that the drop-in wage bill can’t be more than proportional to the revenue fall and that due to re-contracting costs, firms keep paying wages as long as they remain profitable. F IRMS ’ E XIT R ATES I NDUCED BY THE R EVENUE estimated the share of firms which have negative profits due S HOCK to the crisis, for each impact sector. We thus combine these Here we predict the increase in firms’ exit under the results to measure the percentage increase in exits induced by different lockdown scenarios. We use the panel dimension the crisis, by multiplying the share of newly loss-making firm of the data to measure the excess exit rate in pre-crisis years with their excess exit rate. We show the results for the three separately for negative and positive profit firms (and in each of and five month lockdown scenario in 5 (b): under a three (five) the three impact sectors). Figure 5 (a) shows these exit rates month lockdown scenario, firms’ exits from the formal econ- in regular times: on average 13% of firms exit in any given omy increase by 47% (77%). This loss of firms is of course year; however firms which had losses in the previous year have particularly acute for the high impact sector where the percent- an exit rate which is at least 8 percentage points higher than age increase in firms’ exits is 115% (168%) compared to the firms which had positive profits. In our previous analysis, we average pre-crisis year. 5 Figure 5: Firms’ Exit Rate (a) Pre-Crisis Average Exit Rate (b) Crisis Induced Exits Note: Panel (a) shows the average exit probability for all firms, and then for loss-making and profit-making firms, using panel data before the crisis. Panel (b) shows the percentage increase of firms’ exit induced by a three or five month output loss, compared to baseline levels. AGGREGATE N UMBERS AND I MPACTS ON THE month shock], suggesting that firms will need to substantially E CONOMY increase borrowing. Payroll losses are also substantial, rang- The impact on the overall economy is severe, with large ing between 3.3% and 6.6% of the annual wage bill - wage falls in tax revenue, increases in debt, and reductions in subsidies can safeguard some employment, especially in the wages and employment.9 Table 3 summarizes the key num- medium-impact sectors: a 50% wage subsidy would reduce the bers for the 3 and 5 months lockdown scenarios and the ag- payroll losses from 3.3 to 2.7% [6.6 to 5.0%] in the three [five] gregate impact on the economy. 54% or less of firms remain month lockdown scenario. Increases in firm exit are relatively profitable after the shock, and almost all firms in the highly im- small, meaning that associated output and payroll losses are pacted sectors register losses. The Corporate income tax rev- also small, but this is likely an under-estimate: Our panel data enue loss is severe, reaching 24% overall in the three-month features only a smaller number of firms that experience large shock scenario and 38% in the five-month shock scenario. In revenue losses and hence allow us to estimate the effect, pre- the high-impact sectors, almost all CIT revenue is lost. This is sumably because most such firms exit the panel. Our estimates because, despite the temporary nature of the shock, the shock mean that the size of government rescue packages for firms and generates large losses which are counted against the profits workers needs to be large, and the budget support from donors made during the remainder of the year. The absolute increase to lower-income countries even larger, to compensate for the in losses is 0.6% [1.1%] with the three-month shock [five- massive loss in tax revenue. 9 As noted above, while it is illegal to fire anyone during the declared State of Emergency, there could still be reductions in employment on the intensive margin (e.g. reductions in hours offered to shift workers; or employees forced to stay at home without pay). 6 Table 3: Aggregate Impacts by Lockdown Duration and by Impact sectors High Impact Medium Impact Low Impact All Sectors 3 5 3 5 3 5 3 5 months months months months months months months months Share of firms 1 profitable at baseline 74.8 70.4 67.1 69.0 Share of firms still 2 profitable (material adj.) 42.3 27.4 52.1 39.0 57.1 50.3 53.9 44.0 CIT revenue loss 3 relative to baseline (%) 52.3 74.9 24.4 38.1 15.3 24.6 24.1 37.5 Absolute losses 4 increase (% GDP) 0.2 0.3 0.3 0.7 0.1 0.1 0.6 1.1 No wage subsidy 9.2 17.2 2.8 6.0 1.6 3.1 3.3 6.6 50% wage 5 Payroll Loss subsidy 8.5 15.7 2.5 4.7 0.5 0.9 2.7 5.0 90% wage subsidy 7.9 14.5 2.3 4.1 0.0 0.0 2.4 4.2 Percentage increase in 6 exit relative to baseline 115.0 167.6 60.7 104.1 29.7 49.3 46.9 77.4 Permanent output loss 7 from firm exit (% GDP) 0.0 0.1 0.1 0.2 0.0 0.1 0.2 0.4 Permanent payroll loss 8 from firm exit (% GDP) 0.2 0.3 1.0 1.7 0.2 0.4 1.4 2.3 7 A PPENDIX Table 4: Sectors and Impact Categories SECTORS (ISIC Rev 4 code) High - Medium - Low Impact A AGRICULTURE, FORESTRY AND Low Impact FISHING B MINING AND QUARRYING Low Impact C MANUFACTURING Low Impact Medium Impact Food products; Beverages; To- Textiles; Wearing apparel; Leather bacco products; Basic pharma- and related products; Wood and of ceutical products and pharma- products of wood and cork, except ceutical preparations furniture; manufacture of articles of straw and plaiting materials; Pa- per and paper products; Printing and reproduction of recorded me- dia; Coke and refined petroleum products; Chemicals and chemi- cal products; Rubber and plas- tic products; Other non-metallic mineral products; Basic metals; Fabricated metal products, except machinery and equipment; Com- puter, electronic and optical prod- ucts; Electrical equipment; Man- ufacture of machinery and equip- ment n.e.c.; Motor vehicles, trailers and semi-trailers; Other transport equipment; Furniture; Other man- ufacturing; Repair and installation of machinery and equipment D ELECTRICITY, GAS, STEAM AND Medium Impact AIR CONDITIONING SUPPLY E WATER SUPPLY; SEWERAGE, Medium Impact WASTE MANAGEMENT AND REME- DIATION ACTIVITIES F CONSTRUCTION Medium Impact 8 G WHOLESALE AND RETAIL TRADE High Impact Low Impact other than food, pharamacies, gas sta- tions Automobile Dealers; Other Mo- Remaining sub-categories tor Vehicle Dealers; Furni- ture Stores; Home Furnishings Stores; Clothing Stores; Shoe Stores; Jewelry, Luggage, and Leather Goods Stores; Sport- ing Goods, Hobby, and Mu- sical Instrument Stores; Book Stores and News Dealers; De- partment Stores; Florists; Of- fice Supplies, Stationery, and Gift Stores; Other Miscella- neous Store Retailers; Con- sumer Goods Rental; General Rental Centers; Apparel, Piece Goods, and Notions Merchant Wholesalers; Automotive Parts, Accessories, and Tire Stores; Direct Selling Establishments H TRANSPORTATION AND STOR- High Impact Medium Impact AGE Scheduled Air Transportation; Nonscheduled Air Transporta- tion; Taxi and Limousine Ser- vice; School and Employee Bus Transportation; Other Transit and Ground Passenger Trans- portation; Support Activities for Air Transportation; Support Ac- tivities for Water Transporta- tion; Traveler Accommodation I ACCOMMODATION AND FOOD High Impact Medium Impact SERVICE ACTIVITIES Special Food Services; Drink- Remaining sub-categories ing Places (Alcoholic Bever- ages); Restaurants and Other Eating Places 9 J INFORMATION AND COMMUNI- Low Impact CATION K FINANCIAL AND INSURANCE AC- Medium Impact TIVITIES L REAL ESTATE ACTIVITIES Medium Impact M PROFESSIONAL, SCIENTIFIC AND Low Impact TECHNICAL ACTIVITIES N ADMINISTRATIVE AND SUPPORT Low Impact SERVICE ACTIVITIES O PUBLIC ADMINISTRATION AND Low Impact DEFENCE; COMPULSORY SOCIAL SECURITY P EDUCATION Medium Impact Q HUMAN HEALTH AND SOCIAL Low Impact WORK ACTIVITIES R ARTS, ENTERTAINMENT AND High Impact Medium Impact RECREATION Performing Arts Companies; Remaining sub-categories Spectator Sports; Independent Artists, Writers, and Perform- ers; Amusement Parks and Arcades; Gambling Industries; Other Amusement and Recre- ation Industries S OTHER SERVICE ACTIVITIES High Impact Medium Impact Offices of Dentists; Personal Remaining sub-categories Care Services; Other Personal Services 10 C ALCULATION DETAILS FOR TABLE 3 divided by (3) GDP (current LCU of the same year), ex- pressed as percentage. Each figure is calculated for a specific Impact category (High, Medium, Low impact and All sectors) and for a specific lock- 5. Payroll Loss, at different wage subsidy rate: (1) sum of down scenario (three and five months): all firms’ new labor costs under lockdown, divided by (2) the sum of all firms’ labor costs at baseline, expressed as 1. Share of firms profitable at baseline: (1) number of firms percentage. with positive profit margin before output shock, divided by (2) total number of firms, expressed as percentage. 6. Percentage increase in exit rate relative to baseline: (1) exit rate of firms after lockdown minus (2) exit rate of 2. Share of firms still profitable (material adj.): (1) number firms at baseline, divided by (2) and expressed as per- of firms with positive profit margin, after material costs centage. adjustment proportional to the shock, divided by (2) total number of firms, expressed as percentage. 7. Permanent output loss from firm exit (% GDP): (1) ad- ditional exit rate relative to baseline multiplied by (2) 3. CIT revenue loss relative to baseline: (1) sum of all the sum of all firms’ turnover at baseline, divided by (3) firms’ profits at baseline multiplied by the corporate in- GDP (current LCU of the same year), expressed as per- come tax rate minus (2) sum of all firms’ profits after centage. lockdown multiplied by the corporate income tax rate, divided by (1) and expressed as percentage. 8. Permanent payroll loss from firm exit (% GDP): (1) ad- ditional exit rate relative to baseline multiplied by (2) the 4. Absolute losses increase (% GDP): (1) absolute value of sum of all firms’ labor costs at baseline, divided by (3) the sum of all firms’ losses after lockdown minus (2) ab- GDP (current LCU of the same year), expressed as per- solute value of the sum of all firms’ losses at baseline, centage. 11