The Impact of COVID-19 on Formal Firms: Lessons from Administrative Tax Data Pierre Bachas, Anne Brockmeyer, Pablo Garriga and Camille Semelet* January 15, 2025 Abstract Most low-income countries lack high-frequency firm-level data to monitor the ef- fect of economic shocks in real time. We examine whether administrative tax data can help fill this gap, in the context of the COVID-19 pandemic. In spring 2020, we used the full population of corporate tax returns for 2019 in six developing countries to predict the effect of COVID-induced shocks on formal firms’ activity. Comparing the predictions to the realized 2020 data, we find that firms were more resilient than predicted: the share of unprofitable firms increased by only 7 percentage points, while aggregate profits and taxes paid remained stable. The simulations failed to anticipate that labor and capital inputs would flexibly adjust and that large firms would be very resilient. Complementing our simulations with higher-frequency VAT data would have markedly improved predictions. JEL classification: H25, H32, H61, O12 Keywords: COVID-19, firms, corporate income tax, administrative tax data * Pierre Bachas: World Bank Research, pbachas@worldbank.org; Anne Brockmeyer: World Bank, Insti- tute for Fiscal Studies and University College London, abrockmeyer@worldbank.org; Pablo Garriga: World Bank, pgarriga@worldbank.org; Camille Semelet: ifo, LMU Munich, csemelet@worldbank.org. This work was funded by the World Bank through the Knowledge for Change Trust Fund and the Fiscal Policy and Sus- tainable Growth Unit, by UKAID through the Centre for Tax Analysis in Developing Countries (TaxDev), and by UK Research and Innovation through Brockmeyer’s Future Leaders Fellowship (grant reference MR/V025058/1). The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. We are grateful to the Ministries of Finance and Tax Authorities in Costa Rica, the Dominican Republic, Ethiopia, Guatemala, Honduras, and Uganda for their collaboration. We thank Rafael Vilarouca for excellent research assistance. We thank Tom Harris, Adrienne Lees, Vedanth Nair, Giulia Mascagni, Kyle McNabb, Giovanni Occhiali, Fabrizio Santoro, Edris Seid and Ben Waltmann for an excellent collaboration in the data analysis in select countries. We are grateful for helpful comments from Elwyn Davies, Leonardo Iacovone, and two anonymous referees. 1 1 Introduction The onset of the COVID-19 pandemic in 2020 forced governments to implement lock- downs and movement restrictions at the risk of curtailing economic activity. While high- income countries often compensated firms and households during the period of reduced activity, developing countries had less fiscal space to do so. It was hence unclear how the shock would impact firms in lower-income countries. The few ex-ante predictions varied across countries and studies and tended to be large (IMF, 2020; World Bank, 2020; Baker et al., 2020; Carletti et al., 2020), but ex-post data on the realized impact is now available. This paper presents ex-ante simulations of the impact of the COVID-19 shock on firm activity in six developing countries and compares them to the realization. The simula- tions were conducted in the spring of 2020 to inform governments at the height of the first COVID wave (Bachas et al., 2020). The realized data is reported a year later, in spring 2021. We find that formal firms’ profits were much more resilient than we had predicted. Our simulations were overly pessimistic about the size of the shock in terms of lost rev- enues, failed to anticipate that all input costs would flexibly adjust, and that large firms would fare better than smaller firms. The retrospective evaluation highlights both the chal- lenges and promises of using administrative tax data to address policymakers’ demands for real-time predictions. The simulations were conducted for six countries in which we were already working with firms’ corporate income tax (CIT) data: Costa Rica, Ethiopia, Dominican Republic, Honduras, Guatemala and Uganda.1 While high-income countries possess various indica- tors of firm activity, firm-level data in developing countries is scarce, and CIT returns often contain the most recent data on formal firms. CIT data include the largest firms, which are often missing in survey data, and constitute a panel.2 We harmonize the CIT data across countries by focusing on variables that are reported in the same way in all countries. Our simulations were based on firms’ total revenue, their cost breakdown and their sector of activity in 2019, and a simple set of assumptions. Specifically, we followed Vavra (2020) to divide sectors into three groups according to the severity of the assumed revenue shock during a lockdown. We assumed that the 2020 lockdown would depress firms’ yearly 1 In our earlier working paper, Bachas et al. (2020), we also included simulations for Albania, Eswatini, Ethiopia, Montenegro, Rwanda, and Senegal, but we do not evaluate these additional simulations, as we do not have the realized 2020 data on the outcomes for their firms. 2 Financial accounting data, such as Orbis or Compustat can be fairly representative of firm activity in high- income countries, yet it only captures a few very large firms in developing countries: e.g. in Honduras, Orbis includes 46 companies with positive sales in 2019, versus 23,390 firms in the corporate tax data for 2019. 2 sales by 25%, 12.5%, or 5% depending on their sector. In addition, we assumed that firms’ material costs would adjust one-for-one with revenue, labor costs would adjust only if necessary to prevent firms from making losses, and fixed (capital) costs would not adjust. Three factors hence governed the severity of the shock’s impact on firms in our simulations: the assumed loss in sales by sector, the sectoral composition of the economy, and firms’ input mix. Given the sectoral composition of the economies we study, our assumption implied a yearly revenue drop of 11% on average. We predicted that only 54% of firms would remain profitable in 2020, compared to a 75% average at baseline, and that aggregate profits would fall by over a third of baseline (or by around 2% of GDP), leading to a drop in corporate tax revenue of around 0.5% of GDP. We retrospectively evaluate the accuracy of these predictions, by comparing them to the realized outcomes reported on the CIT declaration for 2020, filed in 2021. We document that firms were resilient: the share of profitable firms fell in all six countries, but only by 7 percentage points on average, a third of the drop we had predicted. Despite a reduction in the share of profitable firms, aggregate profits and corporate tax collection hardly fell in 2020. These results are broadly consistent across the six countries. What explains firms’ resilience and the overly pessimistic simulation results? Quali- tatively, we see that the assumptions we made for our simulations were reasonable. The economic sectors predicted to face a larger reduction in sales did face a larger drop than other sectors, and the relative ranking of input costs by ease of adjustment held: material inputs adjusted more than labor inputs, which in turn adjusted more than capital inputs. Yet the magnitudes, measured in constant prices, were off.3 First, the size of the revenue shock was smaller than we predicted across all sectors and countries (except in Honduras): firms’ year-on-year revenue dropped by only 5% on average between 2019 and 2020, while our simulations assumed that the lockdowns would induce an annual average drop of 11%. Second, total input costs were adjusted flexibly to limit the loss in profits: contrary to our simulations which assumed some degree of rigidity in labor and capital costs, labor costs adjusted close to one-for-one with revenue, and capital costs also adjusted substantially. Third, larger firms fared much better than smaller firms, thus mitigating the aggregate loss in profits and tax revenue. Taken together, the moderate loss in revenue and the full in- put adjustment explain the modest reduction in the share of profitable firms; adding the resilience of large firms rationalizes the quasi-absence of a reduction in aggregate profits. 3 All outcomes in this paper are presented in real terms, adjusted to constant 2016 USD. This standardization enables comparisons across years by accounting for inflation. 3 In our retrospective evaluation, we examine how much the accuracy of our predictions would have improved if we had used “real-time” data on sales reported in monthly value- added tax (VAT) returns to measure the size of the revenue shock by sector and country. Although we did not have VAT data for all countries at the time, the governments had these data, and we obtained access to them in four of the six sample countries for the ex-post evaluation. The reported year-on-year changes to sales in the VAT data between March and June 2020 provide a real-time estimate of the sector-specific drop in sales. We thus augment our simulations by replacing the assumed sales drop with the observed sales drop. This halves the gap between our prediction and the realized effect of the shock on firms’ profitability in 2020. What might explain the remaining gap in firms’ profitability, once we control for the shock size with the VAT data? We previously discussed that inputs adjusted more flexibly than we assumed. Two further hypothesis concern changes to prices, and particularly to relative output versus input price, and support policies implemented by governments. Re- garding prices, we first document that inflation in 2020 remained moderate and in line with previous years, with limited differences in inflation across the three impact sectors. Yet, we show, that a moderate difference in inflation between output versus input prices could account for the residual gap, and that this hypothesis finds some support in the limited available data comparing consumer and producer prices. Finally, we discuss the role of governments’ support policies in explaining the cross- country heterogeneity and the resilience of firms. Unsurprisingly, the lockdowns’ strin- gency matters for the cross-country variation: Honduras experienced both the longest re- strictions to economic activity (over three months) and the largest drop to firms’ revenue, while Costa Rica had short lockdowns (two weeks) and saw the best firm performance. In terms of economic support, the policies enacted in the six countries were limited, repre- senting 1-3% of GDP, which is much smaller than the support packages provided in most high-income countries. The measures targeting firms (rather than households) seemed par- ticularly small. Stimulus policies appear unlikely to explain firms’ resilience. Our study makes two main contributions. First, we document the resilience of formal firms, in a comparable manner across six developing countries, and detail the channels that explain this resilience. Second, we examine a use case for administrative tax data to inform real economic outcomes, in a context where other data are scarce, and where just-in-time policy analysis is crucial for policymakers. High-frequency administrative data is often used for policy analysis in high-income countries. It could increasingly be used in lower- 4 income countries, as the digitization of tax data facilitates their extraction and processing. Specifically, databases that combine in close to real-time, yearly corporate tax returns and monthly value-added tax data, could be used, not only to inform the impact of COVID-19 on economic activity, but also for other shocks that impact firms’ sales, profits, employ- ment, and tax revenues. Our work relied on the assembly and harmonization of a novel dataset of corporate tax records for multiple countries, as part of a World Bank project to promote the use of administrative tax data for research and policy evaluation in developing countries.4 The paper is organized as follows. Section 1.1 reviews the literature on the impact of the COVID-19 pandemic on firms. Section 2 presents the data and the simulations. Section 3 compares the simulations to the realizations. Section 4 decomposes the channels explaining the differences between simulations and realizations. Section 5 discusses improvements to the simulations and government policies. Section 6 concludes. 1.1 Literature At the onset of the COVID-19 pandemic, firms worldwide experienced lost revenue, busi- ness closures, and worker layoffs.5 Yet, we find that most formal firms in developing countries were resilient, rebounding quickly and completely after the initial shock. This result is echoed in the literature: the most comparable papers also use VAT return data, but in a single country, and find that firms in China, Rwanda, and Zambia, respectively, recov- ered their sales back to pre-crisis levels fast (Chen et al., 2023; Mascagni and Lees, 2023; Hoy et al., 2022). In Uganda, one study shows that small and medium firms temporarily closed but quickly reopened after the lockdown (Alfonsi et al., 2021); and another finds that most employees returned to their original employer despite mass layoffs at the onset of the pandemic (Bassi et al., 2021). We also document the heterogeneity of the shock across firms and countries. First, we find that within the formal sector, firms in the top sales decile faced a lower revenue shock than smaller firms, in each of our six sample countries. This adds to the literature on the correlates of resilience, which also finds that younger firms and firms part of a business 4 See Pomeranz and Vila-Belda (2019) and Slemrod (2018) for a summary of recent studies using tax data. 5 See for example Apedo-Amah et al. (2020), Adams-Prassl et al. (2020), Bartik et al. (2020), Fairlie (2020), Humphries et al. (2020), Zou et al. (2020), Hatayama et al. (2022), Khamis et al. (2021), Aga and Maemir (2022), Guerrero-Amezaga et al. (2022), Kawaguchi et al. (2022), and Angelov and Waldenstr¨ om (2023). 5 group fared better (Jain and Kumar, 2023; Adian et al., 2020).6 Second, we show the importance of these heterogeneous effects in explaining the limited impact of the pandemic on aggregate revenue, profits and taxes, and suggest that the duration of lockdowns had a strong impact on the revenue shock. Our results on tax revenue align with James (2020), who shows that tax collection shortfalls due to the pandemic were lowest in Sub-Saharan Africa and Latin America. Finally, we add to the literature that predicts and estimates in a timely manner the impact of sudden shocks (such as the COVID-19 pandemic). Our exercise is most closely related to Carletti et al. (2020), who simulated the impact of a three-month lockdown with sector- specific shocks for Italian firms.7 More broadly, an expanding body of work uses high- frequency data—on a daily, weekly or monthly basis—to estimate the impact of shocks. The data sources are numerous, including real-time survey data (Adams-Prassl et al., 2020); stock market returns (Alfaro et al., 2020; Baker et al., 2020); electricity and housing prices (Bricongne et al., 2023; Fezzi and Fanghella, 2020); credit card usage (Chetty et al., 2020; Horvath et al., 2023); job adds (Forsythe et al., 2020); and working hours (Bartik et al., 2020; Kurmann et al., 2020). These studies are conducted for the United States and Europe, where such data are available.8 In lower-income countries, where the aforementioned data sources are typically of lower quality, less relevant or entirely unavailable, we argue that monthly tax data can help generate timely estimates in lower-income countries, and help inform government policies to support the economy. 2 Data and Simulations 2.1 Data Corporate Income Tax Returns. Our data contains all CIT returns for 2019 and 2020, for six countries: Costa Rica, the Dominican Republic, Ethiopia, Honduras, Guatemala, and Uganda. We arrived at this sample by selecting, among the countries for which we 6 Formal firms were also more resilient than informal firms, after controlling for government support; the adoption of technologies before the pandemic is associated with a higher resilience in Egypt (El-Haddad and Zaki, 2023), and in El Salvador, Guatemala, Honduras, and Nicaragua (Olvera et al., 2022). 7 Carletti et al. (2020) use Orbis data and assume a drop in firm revenues in each sector that is proportional to the fraction of value added forgone in the corresponding industrial sector as a result of lockdown. They assumed that labor costs would go down and material and fixed costs would remain constant. We instead allowed material costs to adjust, which turned out to be consistent with observed firms’ responses. 8 Administrative data on payroll and benefit claims has also been used in richer countries, see for example Cajner et al. (2020) in the US, Alstadsæter et al. (2020) in Norway, and Cui et al. (2022) in China. 6 simulated the impact of COVID-19 on firm activity at the onset of the pandemic, those countries that shared their 2020 data to compare simulations to realizations. CIT data face two main limitations: they cover only formally registered firms and thus do not include the informal sector;9 and the reported values could differ from true values. Despite these drawbacks, the CIT data present several advantages over other data on firms in developing countries and are arguably the most accurate data on the formal sector: they contain the most recent information on firms’ activity; they include the largest firms which might not respond to surveys; and they are often the only panel on firms. While the structure of corporate income tax returns differs across countries, we were able to harmonize the data to obtain measures of revenues, costs and profits that are as comparable as possible. We break down the costs into material inputs, labor costs and fixed costs.10 We use data on firms’ economic sector to assign them to one of three impact groups, low, medium and high, following a revenue-shock-severity taxonomy for the pan- demic designed by the World Bank (Vavra, 2020). We will use the terms sales and revenue interchangeably. In the CIT data, revenue includes sales and other sources of income, but the latter are very small and hence treated as sales in the simulations. We run our simulations on a balanced panel of firms that appear in both 2019 and 2020. This allows us to have a simulated and a realized outcome for each firm. To be in the sample, firms must have filed a CIT return in 2019 and 2020, and the return must contain information on the variables used in our analysis: revenue, gross tax base, costs (either labor or material costs must be reported, or both), and industry group. Firms reporting zero sales in either 2019 or 2020 are dropped. The net change in the number of firms between 2019 and 2020 is summarized in Table A1. Table 2 shows descriptive statistics for each country’s firms in 2019 and the distribution of firms across the three impact sectors.11 Value-Added Tax Returns. To measure the monthly realization of the revenue shock we use monthly VAT returns for 2018-2020, obtained ex-post in 2021. We define a semi- balanced panel of firms by including firms that: a) filed VAT returns at least once every 9 Unincorporated firms and firms filing under simplified regimes are also excluded from this analysis because their tax treatment differs across countries (thus the data available on these firms also differs): some firms for example only pay presumptive taxes on sales or assets and do not report profits and costs. 10 In Bachas et al. (2020), we validate the cost breakdown in the administrative data by comparing it to the World Bank Enterprise Surveys (see Figure A3 of the paper), and show that cost shares vary with firm size, allowing larger firms to more easily adjust to demand shocks (Figures A1 and A2). 11 In the countries where we can trace firm entry and exit over a longer period, we do not observe a substantial change in the number of firms in 2020 compared to previous years (Figure A1). Our analysis hence focuses on intensive margin changes. 7 quarter of 2018 and 2019, and b) filed at least once in the second or third quarter of 2020. This sample allows us to capture the intensive margin change in revenue in the first half of 2020.12 While we do not limit ourselves to a matched CIT-VAT sample of firms, the overlap between the two datasets is high: An average of 65% of firms appearing in the CIT data also appear in the VAT data. Other Data. To capture the nature of lockdowns, we use the data assembled by Hale et al. (2021) on the timing and stringency of firm closures and movement restriction policies. This data also contains information on economic support policies, which we complement with qualitative data on the list of policies implemented from the ILO (2020). 2.2 A Simple Framework to Simulate the Impact of Lockdowns Lockdown Shock. We simulate a scenario in which a transitory demand shock generates a drop in firms’ sales over three months, after which firms’ activities return to their pre- shock level. The three-month window can be interpreted either as the length of lockdowns under perfect compliance with movement restrictions or as a way to proxy the reduction in activity over a longer period under imperfect compliance. The severity of the revenue shock is modeled as the percentage drop in monthly sales which, in our simulation, depends on the economic sector: firms in the high, medium, and low impact sectors face a 100%, 50%, and 20% drop in demand respectively during the lockdown period, which translates into a yearly drop in firms’ revenues of 25%, 12.5% or 5%. Table 1 lists the sectors and their severity assignment. Sectors with low impact include industries that were expected to be less disrupted during the pandemic due to their essential nature and the possibility for remote work (e.g. agriculture, I.T.). Sectors with medium impact were expected to face noticeable disruptions related to changes in demand (e.g. manufacturing). Finally, high- impact sectors were significantly affected by the pandemic and often had to close entirely due to governmental restrictions (e.g. restaurants, transportation). Production Function and Adjustment Costs. Firms produce a unit of output with a Leontief production function which requires material, labor, and capital in fixed propor- 12 We also use alternative definitions of the panel and obtain similar results. In addition, we compute the revenue shocks for a fully balanced panel of firms. As expected, the shocks are slightly smaller but not significantly different, except for Costa Rica. The divergence in results in Costa Rica is due to a change in filing requirements that occurred in the middle of our sample period. 8 tions.13 These proportions are estimated from firms’ 2019 tax declarations. We assume that each input adjusts differentially to the revenue shock, given the nature of adjustment costs: • Material inputs fully adjust in proportion to the revenue shock: firms should be able to adjust inventory and raw materials quickly to match expected demand drops. • Labor costs only adjust if necessary to avoid making losses, because we assume that re-contracting workers is costly and hence preferably avoided if firms expect the demand shock to be temporary. Thus, firms that can absorb the demand shock without losses prefer not to lay off workers, even if profits temporarily decrease. • Fixed costs are assumed to be non-adjustable as firms continue to honor longer-term contracts (rental agreements, debt payments, etc). Hence, we conjecture that firms face substantial adjustment costs for labor and capital, and that they aim to weather this transitory shock to be in a position to scale back their production fast, without changing their production technologies. Predictions from Simulations. In the spring of 2020, using the above-stated assump- tions, we generated predictions for how the COVID-19-induced lockdowns would impact firm activity, to engage with policymakers in partner countries as the crisis was unfolding. These predictions relied on the 2019 corporate income tax returns which had just been sub- mitted. The analysis was published in the form of country-specific notes, accompanied by a replication package, a blog and a synthesis paper (available here).14 The results of our simulations are shown in Table 3. For firm-level outcomes, we report the average year-on-year change (relative to 2019). Aggregate outcomes are in levels and can be compared to the 2019 baseline in the first row. The main predictions are similar across countries and can be summarized as follows: • We predicted a drop in yearly aggregate sales of 11% on average. This follows from the above-stated assumption about the sales drop for each impact sector, and the sectoral composition of the countries’ economies. 13 We assume that firms are price takers and that all prices are fixed at their pre-pandemic level, such that the margin of adjustment is quantity. 14 The original predictions covered ten countries and we were able to update the data to 2020 for six of them. The predictions also included simulations on the potential effects of employment support programs (which were rarely implemented in practice in lower-income countries) and on firm exit from the formal sector. 9 • We expected firms to reduce their costs on average by 55% of the size of the revenue shock. This follows from the assumption we made on the adjustment of material, labor and capital costs, and the observed composition of firms’ costs in the 2019 data: Materials constituted on average 45% of total costs, labor 20% and capital 35%. • The sales shock was expected to reduce the share of profitable (non-loss-making) firms by 21 percentage points. This is because, although an average of 75% of firms are profitable at baseline, the median baseline profitability of firms is low (2-3% of revenue). Aggregate (positive) profits were expected to fall by 36% compared to baseline, and total losses were expected to increase by 62%. The drop in aggregate profits would in turn lead to a fall in corporate tax revenue of the same proportion. 3 Realization Versus Simulations How did firms’ revenue and profits actually fare during the pandemic year 2020? And how do the realized outcomes compare to our predictions? Figure 1 compares our simulations to the realizations for the year 2020, and to the baseline year of 2019, in each of the six countries. Panel (a) displays the size of the shock, measured as the aggregate reported revenue (sales plus other income), as a percentage of the 2019 baseline. In 2019, the total revenue reported by firms in the six countries ranged from USD 14 Billion in Ethiopia to USD 89 Billion in the Dominican Republic. Our simulations predicted a drop in total revenue of around 11%, varying slightly across countries depending on their sectoral mix. The realized data shows that aggregate revenue indeed fell in five out of six countries, but the average aggregate drop across countries is less than half of the prediction. The results are fairly heterogeneous: in Costa Rica aggregate revenue rose, in Honduras revenue fell by more than we predicted, and in the other four sample countries realized revenue lies somewhere between the prediction and the baseline 2019 revenue. Given the revenue loss, firms’ profits are expected to decrease. Figure 1, Panel (b) shows the share of profitable firms (non-loss-making): at baseline, 75% of firms were profitable on average across countries. We predicted that the share of profitable firms would drop by 21 percentage points on average, with variations across countries depending on the sector mix, the cost structure of firms, and the initial distribution of profits. In practice, firms’ profitability was resilient: although the share of profitable firms fell in five of the six countries (it stayed stable in Ethiopia), it only dropped by an average of 6.7 percentage 10 points, a third of the predicted drop. Figure 1, Panel (c), plots the aggregate taxable profits. At baseline, profits represented 5-10% of aggregate revenues depending on the country, and ranged from USD 1.2 billion in Uganda, to USD 6.2 billion in the Dominican Republic. Reported losses are not counted as negative values in this measure. In our simulations, the combination of a large reduc- tion in revenue and high adjustment costs for labor and capital inputs implied very large drops in aggregate profits, of the order of 35% of the baseline on average. In practice, the aggregate profits hardly fell on average. Aggregate profits dropped in three countries (Do- minican Republic, Ethiopia, Honduras) and rose in the other three countries (Costa Rica, Guatemala, Uganda).15 Table 3 reports on the same outcomes as Figure 1, and adds several others, including the average profit margin, aggregate taxes paid, and aggregate losses.1617 To summarize, Figure 1 shows that the average revenue shock was smaller than pre- dicted and that the share of profitable firms fell by only 7 percentage points, about a third of our prediction. Despite drops in aggregate revenues and in the share of profitable firms, aggregate profits remained stable. This is contrary to our pessimistic prediction of a 35% drop in profits, which would have shaved a third of the corporate income tax base. To reconcile how aggregate profits could remain stable over time while the share of profitable firms dropped and aggregate losses increased, we turn to examine the distribution of firms’ profit margins, before and after the shock. Figure 2 plots the distribution of profit margins, defined as the ratio of profits over revenue, for each country, comparing the 2019 baseline to the 2020 realization. The left and right tails are winsorized at -25% and 25% profitability. The right tails of the distribution are comparable, meaning that there are as many very profitable firms, with profit margins above 25%, in 2020 as in 2019. The key changes in the distribution are (1) a compression around the modal profitability of 2-3%, and (2) a rising number of firms reporting large losses in the left tail. These distributional results help rationalize the previous finding: the same number of firms remain very profitable, but fewer firms report moderate profits, and more report large losses. The distributional analysis highlights the heterogeneity in responses across firms, which we 15 Note that in our simulation all firms lost revenues and none could grow, so lockdown-induced losses for some firms could not be offset by growth of other firms or during the post-lockdown period. 16 We saw in Figure 1 Panel (c) that aggregate profits remained fairly stable, implying stable corporate income taxes paid (Table 3, column (9)). However, aggregate losses increased slightly (Table 3, column (10)), a result which has negative long-term implications for corporate tax revenues and firms’ health. 17 The quality of the 2019 data—reported in early 2020—could be affected by the pandemic. To address this concern, Table A2 shows the robustness of the simulations to changing the baseline year to 2018. The results are qualitatively similar, indicating that the reporting conditions of 2019 do not drive the results. 11 return to below when studying effects by firm size. 4 Where Did Predictions Go Wrong? Decomposing the Key Channels The realized 2020 data shows that formal firms were more resilient than predicted during the pandemic. We now examine the factors that explain the discrepancies between simula- tions and realizations. Drop in Revenues. We assumed that the lockdown shock would lead to an annual drop in reported revenues of 25% for the high-impact sectors, 12.5% for the medium-impact sectors, and 5% for the low-impact sectors. Figure 3, Panel (a), displays for each country and sector, the realized drop in sales compared to the simulation. The figure reveals two key patterns. First, the qualitative ranking of the size of the revenue drop across sectors was as expected: in all countries, the high-impact sectors (e.g. restaurants, transport) were more impacted than the medium-impact sectors (e.g. manufacturing), which in turn were more impacted than the low-impact sectors (e.g. agriculture). Second, revenue fell much less than expected in all sectors, across countries: the average realized revenue drop for the high-impact sectors was 18% (compared to a 25% prediction), for the medium-impact sectors 7% (compared to a 12.5% prediction), and for the low-impact sectors 1% (compared to a 5% prediction) (see Table A3). We will examine in Section 5 how much better our simulations could have fared if we had used monthly sales data from VAT declarations to calibrate the size of the revenue shock for the simulations. We will also see that the size of the shock is associated with the duration of the lockdown. Cost Adjustments. Although firms’ revenues did not fall by as much as we predicted, they still dropped in most sectors and countries. How did firms’ input costs adjust? Fig- ure 3, Panel (b), combines all sectors to show the average change in material, labor and fixed costs, respectively, and compares them to the changes in revenues. In our simula- tions, we had assumed a one-for-one adjustment of material inputs to revenue changes, a partial labor adjustment, and no adjustments to fixed costs. The figure shows that the rel- ative ranking of cost adjustments was generally as expected: material costs adjusted the most, followed by labor costs, and fixed costs adjusted least. Yet, the magnitudes we had assumed were off. In four countries, material costs adjusted slightly more than one for one with revenue. More importantly, labor costs adjusted close to one for one with revenues, 12 and fixed costs also adjusted some. Our assumption of labor and fixed cost rigidity was in- correct. Overall, firms reduced their total costs approximately in proportion to the revenue drop. This in turn implies that firms’ profitability was less impacted than we predicted. We note that, had we modeled firms’ production function as Cobb-Douglas instead of Leontief, firms could have reduced the easy-to-adjust input (material) more than propor- tionally to the revenue shock, and substituted it with harder-to-adjust inputs (labor, capital). Thus, with a Cobb-Douglas production function, our simulations would have predicted a slightly smaller drop in profits.18 Heterogeneity by Firm Size. Given the skewness of the firm size distribution, a large share of profits and taxes are accounted for by the largest firms: across the sample countries, the 10% largest firms in each country (ranked by total sales) account for 83% of total sales and 86% of total profits. To understand differences in aggregate values of revenues and profits between the simulations and the realization, we examine the heterogeneity of the shock across the firm size distribution. Figure 4 ranks firms by deciles of revenue within their country, and shows the year-on-year changes in revenues by decile. The dotted lines plot the year-on-year revenue changes between 2018 and 2019, which could approximate the pre-pandemic ‘equilibrium’ growth. We observe that most firms were growing in the year before the pandemic, and small firms were growing faster than large firms. The crossed lines show firms’ revenue change between 2020 and 2019, thus including the effect of the pandemic. The slope of revenue growth by firm size is now inverted: larger firms fared substantially better than (or, put differently, not as badly as) smaller firms. Graphically, this relative reversal of firm growth by size can be seen by comparing the area between the two lines for the top deciles versus the bottom deciles. Between 2020 and 2019 the gap in firm growth rapidly falls across deciles, where red intervals denote a year-on-year reduction in the growth of that firm size decile and green intervals a year-on-year increase. A few hypotheses that might explain the resilience of larger firms relative to small and medium firms, are higher inventories, better access to credit, and more government support. We investigate empirically the first two hypotheses in the administrative data by measuring changes between 2019 and 2020, in inventory and in interest payments. Figure A4 shows that drawing on inventory (as a share of revenue) was not size-related (panels a and b). 18 While we can calculate how much better our simulations would have fared if we had made different as- sumptions on the cost adjustments within the Leontief framework, conducting the simulations with a Cobb- Douglas production function would require additional assumptions on the elasticity of substitution of dif- ferent inputs and a parametrization of the fixed cost of adjustment. 13 However, panel (c) supports the hypothesis that larger firms could increase interest deduc- tions by more than medium-sized firms. Further, the literature finds that larger firms, on average, benefited more from government support than smaller ones, although in our sam- ple of countries, stimulus policies were small (see Section 5.3).19 To summarize, our simulations predicted a bleak outcome for formal firms in develop- ing countries following the COVID-19 pandemic shock. In practice, firms’ activity was quite resilient, and only in one country, Honduras, was the shock to revenues as large as anticipated. On average, aggregate profits and corporate taxes hardly suffered in 2020. Al- though our assumptions about which sectors would be most impacted and about the relative adjustment of inputs held qualitatively, the assumed magnitudes were off. Revenues fell by only half of our predictions, and we did not anticipate that large firms would fare better than small ones. Further, while we assumed that labor and fixed costs would be difficult to adjust, in practice firms were able to adjust all inputs, such that total costs fell propor- tionally to revenue. Taken together, the moderate loss in revenue and the strong adjustment of inputs help explain the limited fall in the share of profitable firms. These facts, together with the resilience of large firms, rationalize the stability of aggregate profits. 5 Improved Predictions, Inflation, and Government Policies We now examine the extent to which timely data on sales, reported in monthly VAT declara- tions, could have improved our predictions. We then discuss inflation, and how differential changes to output versus input prices might explain the results. We finish by reviewing the role of government policies in terms of lockdowns and of economic support enacted during the pandemic, and their heterogeneity across countries. 5.1 Monthly Value-Added Tax Data to Measure the Size of the Shock When performing the simulations, the most recent data we had access to was the latest set of annual CIT returns, typically filed in the first quarter of a year. Firms also remit the VAT at a monthly frequency, which is the main tax on consumption in the six sample countries. As part of our retrospective evaluation, we obtained access to VAT data for four of six 19 the Covid 19 Business Pulse survey by Cirera et al., 2021 results by firm size across 60 countries: on average, large firms (100+ employees) received more wage subsidies than small firms (5-19 employees (20.2% vs 15.8%), more tax support (9.5% vs 7.58%), more payments deferral (6.2% vs 5.9%), and more access to credits (6.7% vs 5.6%). Yet, the fraction of business receiving any support is low. 14 sample countries, allowing us to study the year-on-year change in monthly sales variations for VAT-registered firms. Figure 5 plots the year-on-year change in monthly sales from the VAT data. The hori- zontal line corresponds to the onset of the pandemic, which we mark as starting in March 2020. The low-impact sectors are shown in green, the medium-impact in yellow, and the high-impact in red. First, we observe that in all four countries, all sectors saw a drop in sales starting in March 2020. The lowest point in sales was attained by April 2020 already, and thereafter sales started recovering. Second, the qualitative ranking of impact categories is appropriate (a result we also observed in the corporate tax data): the high-risk sectors were the most impacted and their firms’ sales remained at levels lower than pre-pandemic levels for all of 2020. The low and medium-impact sectors, however, quickly rebounded and recovered to their pre-pandemic sales levels by mid-2020. The low-impact sectors were the least affected in all countries. How much better could our simulations have fared if, instead of assuming a shock size, we had used the observed revenue shock reported in the VAT data contemporane- ously? This requires us to assume a time window of observation relative to the onset of the pandemic in late February 2020, which we (arbitrarily) set to four months: in other words, we suppose that we would have observed the VAT sales data as of June of 2020, and assigned the year-on-year sales deviation in the first half of 2020 relative to 2019 as the yearly country-sector size of the shock. We display these shock size numbers in Fig- ure 5. We then make our predictions about firms’ profitability and tax liabilities, keeping assumptions about cost adjustments constant. Note that this approach implicitly assumes that from July 2020 onward the shock is over, which is a simplification but not far from reality as shown in Figure 5. Figure 6 shows the augmented predictions using the monthly VAT sales data, and com- pares them to the original predictions, the realization, and the baseline. Using the VAT country-sector-specific estimates of the shock size (as of June 2020) would have produced substantially more accurate predictions. Panel (a) shows that these adjusted simulations predict accurately the size of the aggregate revenue shock, on average across sample coun- tries. This is particularly the case for Costa Rica and Guatemala, the countries where the gap between our baseline simulation and the realization was the largest. The (more) accu- rate revenue shock prediction then helps improve the predictions for the share of profitable firms and aggregate profits (Panels (b) and (c)): half of the misprediction gap for these outcomes has been closed by using monthly VAT returns data to predict revenue losses. 15 5.2 Price Changes Consumer Price Inflation in 2020. All variables are expressed in constant 2016 prices, thus controlling for inflation. Yet, one might wonder how prices evolved in 2020, and if there were differences in inflation between impact sectors, and between output versus input prices. We rely on widely published monthly consumer prices (instead of rarely available producer prices), and create a concordance between products and sectors.20 Figure A2, Panel (a) shows yearly aggregate inflation between 2012-2022. Inflation stayed moderate in 2020 (the median country’s inflation was 3%) and followed prior years’ trend.21 Yet, a concern is that the pandemic could have caused large sectoral variations. Panel (b) shows however that this concern seems second order: median (mean) inflation for the low, medium, and high impact groups were relatively similar at respectively, 4%(5%), 4%(5%) and 5%(7%). Equipped with a sector-level monthly price index, we can adjust our measure of the revenue shock for each lockdown impact group based on the VAT data (as in Section 5.1), by deflating prices at the impact group-month level. Figure A3, shows that these more precise price adjustments only slightly changed monthly sales by impact-sector, and our simulated aggregate revenue. Differential output versus input price inflation. Although aggregate and sectoral con- sumer price inflation remained moderate in 2020, one explanation for the resilience of firms’ profitability, is a faster increase in output prices relative to input prices.22 Although we lack data on producer prices (with one exception), we can estimate the inflation differ- ential needed in our simulations to match the realized profitability of firms. We are able to re-run our simulations, assuming that output prices grew faster than input prices, for three of the six countries (Costa Rica, Honduras, and Guatemala). In Table A5, columns (1) to (4) display by country, respectively, the share of profitable firms at baseline in 2019, the realized share of profitable firms in 2020, the simulated share of profitable firms in 20 Table A4 summarizes the process through which we match products to sectors. Each consumption group is matched to industry sections which, in turn, are associated with a lockdown impact group. In order to construct a measure of monthly price indices for each of these lockdown impact groups, we calculate a weighted average of the CPIs of the corresponding consumption groups, weighted by their contribution to the general CPI of each country. 21 Price levels increased in subsequent years, outside of the scope of our study, with median (average) inflation reaching 4%(8%) in 2021, and 9%(12%) in 2022. 22 Ex-ante this seems plausible: for inputs such as labor and capital, prices could have stagnated over the period, since, wage increases were unlikely, governments might have kept interest rates low, and anecdotal evidence points to re-negotiation of leases. 16 2020 (ex-ante), and the VAT-augmented simulated share of profitable firms in 2020 (ex- post) from Section 5.1. Our goal is to set an output price growth relative to the input price growth, such that our simulations match the realized profitability share (Column 2). Column (5) reports the results: the differential inflation needed to match the realized profitability is large, ranging from 3.7% in Guatemala to 7% in Honduras. We note that in all three countries, this differential is higher than their aggregate 2020 inflation. Column (6) reports the results of the more modest exercise which aims to close the profitability pre- diction gap in the VAT-augmented simulation. This augmented simulation already closes a good part of the gap (see Figure 6 (b)). Thus, for Costa Rica and Guatemala, inflation differences between output and input prices of 1% and 0.7% would be sufficient to close the residual gap, yet in Honduras, this differential would need to be large (5.1%). Costa Rica is the only country for which we have producer prices for some sectors. Fig- ure A5 (a), reports yearly consumer and producer prices for “food and beverages” where the concordance between production and consumption appears close. A gap appears between output and input prices in 2020 (3.7% vs 1.6%), which is sufficient to close the residual profitability gap in the augmented simulations for Costa Rica. Yet, we note that during the peak of the pandemic in March-June 2020, monthly producer price inflation was higher than consumer price inflation (Figure A5 (b)). 5.3 Lockdown Duration, Government Policies and Heterogeneity Across Countries Lockdown Duration. The resilience of firms’ revenue and of aggregate profits is not ho- mogeneous in magnitude across the six sample countries. In particular, firms in Honduras fared worse than our predictions, in sharp contrast to firms in Costa Rica that grew.23 To better understand these patterns, we consider the policies enacted during the pandemic in the six sample countries. We use the comprehensive global database assembled by Hale et al. (2021) to detail the restrictions to economic activity and movement. Figure 7, Panel (a), shows the restrictions to activity that were in place over each day of 2020. We focus on two key restrictions: forced business closures, which directly im- pact firms; and stay-at-home mandates which impact the consumers (and workers) of these firms. We observe large variation in the stringency of restrictions across countries: the least restrictive countries were Ethiopia and Costa Rica, followed by Uganda, while at the other end, the Dominican Republic, Guatemala, and especially Honduras imposed long 23 In Table 2, we see that the sectoral mix does not vary substantially across countries, so differential sectoral exposure cannot explain the large differences in outcomes observed across countries. 17 lockdowns. We obtain similar country rankings with different definitions of lockdowns and using the index of overall restriction stringency of Hale et al. (2021), which combines school closures with business and movement restrictions (Table A6, columns 1-4). Figure 7, Panel (b), plots the restriction stringency index against the average year-on- year change in realized revenues in each of the six countries. The stringency of restrictions correlates negatively with the change to firms’ economic activity. At the extremes, Costa Rica faced short lockdowns and firms fared well, while Honduras faced the longest lock- down and was the country where firms fared worst. Table A6 also compares the six sample countries to the average of their regions (Latin America and Africa). Compared to other countries in their respective regions, the stringency of restrictions is on the higher side in our sample of countries, which could imply that non-sample countries might have faced even smaller revenue shocks. Economic Support Policies. To counteract the restrictions, governments enacted poli- cies to support economic activity. To capture the importance of these policies we use a measure of the size of the announced stimulus (as a % of GDP) from the IMF fiscal mon- itor database (IMF, 2024) and qualitative data on the list of policies specifically designed to support firm activity from the ILO (2020). Table A6, column 5, shows the size of the announced stimulus packages in our sample countries, and compares them to averages for Latin America, Africa, and high-income countries. High-income countries spent 10% of their GDP to support the economy, three times as much as countries in Latin America and four times as much as countries in Africa. In four of six countries in our sample, the support packages covered less than 2% of GDP. The support policies targeted workers and households more than firms. According to our calculations based on the information and legal documents referred by Hale et al. (2021) and the IMF fiscal monitor database (IMF, 2024), Costa Rica, the Dominican Republic, and Guatemala allocated approximately 93%, 95%, and 75% of their respective support pack- ages to households. In Ethiopia, approximately 63% of the support funding was directed towards households. To zoom in on the nature of support to firms, Table A7 lists the poli- cies implemented in our countries. The first column lists relief to tax and social security payments. While there were tax measures in most countries, they were typically not out- right tax reductions (except in Ethiopia), but rather tax deferrals. Most deferrals applied to social security contributions, the relevant margin for limiting layoffs. Deferrals would have reduced the real tax obligation only by the rate of inflation, in addition to averting late 18 fees or the requirement to take out short-term loans to meet tax payment deadlines.24 Other policies to support firms include the provision of lines of credit to fund working capital and employment flexibilization policies. We did not find evidence of direct transfers to firms. Given the small size of government support packages in our sample countries, the lim- ited targeting to firms, and the near-absence tax relief measures, it seems unlikely that government policies played a key role in firms’ resilience.25 The exception is Guatemala: it had a larger stimulus program (5% of GDP) and fared relatively well, despite stringent restrictions (second longest lockdown after Honduras). 6 Conclusion Using micro tax return data in six low and middle-income countries, this paper documents the resilience of formal firms to the shock induced by the COVID-19 pandemic. The num- ber of firms that became unprofitable only increased by 7 percentage points, while aggre- gate revenue and profits fell even less. The realized outcomes are far more positive than those we predicted to inform governments early on during the pandemic. Our simulations missed the mark because (1) the drop to revenue was smaller than we assumed; (2) firms managed to reduce their input costs flexibly, and (3) large firms fared much better than small firms which explains the limited aggregate impacts. In addition, it is possible that output prices grew faster than input prices, limiting firms’ profit losses. The realized outcomes are heterogeneous across countries and correlate with the stringency of the lockdown. This paper highlights the value of administrative tax data to inform about real economic outcomes in a timely manner. The use of such data is becoming common in high-income countries (Chetty et al., 2020) but remains limited elsewhere. We show that combining the corporate income tax with monthly VAT sales data could have improved the nowcast- ing of firm activity compared to only relying on the yearly corporate income tax data. In the future, more detailed administrative microdata–such as daily electronic transaction receipts–could permit even more granular insights in close to real time. 24 We are implicitly assuming that deferred tax liabilities were ultimately collected, but enforcement might also have changed during the pandemic. 25 Surveying firms across 60 countries, Cirera et al. (2021) show that policy support has been especially limited for the most vulnerable firms and countries. In China, more than half of tax-registered firms did not benefit at all from payroll subsidies due to labor informality (Cui et al., 2022; Chen et al., 2023). Surveys in eight Latin American countries also showed that despite some available assistance to firms, awareness and applications to these programs were low and the impact was short-lived (Guerrero-Amezaga et al., 2022). 19 References Adams-Prassl, A., T. Boneva, M. Golin, and C. Rauh (2020). “Inequality in the Impact of the Coronavirus Shock: Evidence from Real Time Surveys”. Journal of Public Eco- nomics 189, p. 104245. DOI: 10.1016/j.jpubeco.2020.104245. Adian, I., D. Doumbia, N. Gregory, A. Ragoussis, A. Reddy, and J. Timmis (2020). Small and Medium Enterprises in the Pandemic: Impact, Responses and the Role of Develop- ment Finance. The World Bank. Aga, G. and H. Maemir (2022). “COVID-19 and Sub-Saharan Africa Firms: Impact and Coping Strategies”. The Journal of Development Studies 58 (12), pp. 2415–2443. Alfaro, L., A. Chari, A. N. Greenland, and P. K. Schott (2020). Aggregate and Firm-level Stock Returns during Pandemics, in Real Time. Tech. rep. National Bureau of Economic Research. DOI: 10.3386/w26950. Alfonsi, L., O. Bandiera, V. Bassi, R. Burgess, I. Rasul, O. A. Veroux, and A. Vitali (2021). COVID-19 and Ugandan SMEs: Impacts and Speed of Recovery. Tech. rep. Alstadsæter, A., B. Bratsberg, G. Eielsen, W. Kopczuk, S. Markussen, O. Raaum, and K. Røed (May 2020). The First Weeks of the Coronavirus Crisis: Who Got Hit, When and Why? Evidence from Norway. Working Paper 27131. National Bureau of Economic Research. DOI: 10.3386/w27131. Angelov, N. and D. Waldenstr¨ om (2023). “The Impact of COVID-19 on Economic Ac- tivity: Evidence from Administrative Tax Registers”. International Tax and Public Fi- nance, pp. 1–29. Apedo-Amah, M. C., B. Avdiu, X. Cirera, M. Cruz, E. Davies, A. Grover, L. Iacovone, U. Kilinc, D. Medvedev, F. O. Maduko, et al. (2020). Unmasking the Impact of COVID-19 on Businesses: Firm Level Evidence from across the World. The World Bank. Bachas, P. J., A. Brockmeyer, and C. M. Semelet (2020). The Impact of COVID-19 on Formal Firms: Micro Tax Data Simulations across Countries. Policy Research Working Paper Series 9437. The World Bank. Baker, S. R., N. Bloom, S. J. Davis, K. J. Kost, M. C. Sammon, and T. Viratyosin (Apr. 2020). The Unprecedented Stock Market Impact of COVID-19. Working Paper 26945. National Bureau of Economic Research. DOI: 10.3386/w26945. Baker, S. R., N. Bloom, S. J. Davis, and S. Terry (Apr. 2020). Covid-Induced Economic Uncertainty. Working Paper w26983. National Bureau of Economic Research. 20 Bartik, A. W., M. Bertrand, Z. B. Cullen, E. L. Glaeser, M. Luca, and C. T. Stanton (Apr. 2020). How Are Small Businesses Adjusting to COVID-19? Early Evidence from a Sur- vey. NBER Working Papers 26989. National Bureau of Economic Research, Inc. Bassi, V., T. Porzio, R. Sen, and E. Tugume (2021). “The Impact of the COVID-19 Lock- down on SMEs and Employment Relationships in Uganda”. Bricongne, J.-C., B. Meunier, and S. Pouget (2023). “Web-scraping housing prices in real- time: The Covid-19 crisis in the UK”. Journal of Housing Economics 59 (PB). DOI: 10.1016/j.jhe.2022.101906. Cajner, T., L. Crane, R. Decker, J. Grigsby, A. Hamins-Puertolas, E. Hurst, C. Kurz, and A. Yildirmaz (2020). The U.S. Labor Market during the Beginning of the Pandemic Recession. NBER Working Papers 27159. National Bureau of Economic Research, Inc. Carletti, E., T. Oliviero, M. Pagano, L. Pelizzon, and M. G. Subrahmanyam (2020). “The COVID-19 Shock and Equity Shortfall: Firm-level Evidence from Italy”. The Review of Corporate Finance Studies 9 (3), pp. 534–568. Chen, Z., P. Li, L. Liao, L. Liu, and Z. Wang (Nov. 2023). Assessing and Addressing the Coronavirus-Induced Economic Crisis: Evidence From 1.5 Billion Sales Invoices. Tech. rep. PBCSF-NIFR Research Paper, 2020. DOI: 10.2139/ssrn.3661014. Chetty, R., J. N. Friedman, N. Hendren, M. Stepner, and T. O. I. Team (June 2020). How Did COVID-19 and Stabilization Policies Affect Spending and Employment? A New Real-Time Economic Tracker Based on Private Sector Data. Working Paper 27431. National Bureau of Economic Research. DOI: 10.3386/w27431. Cirera, X., M. Cruz, E. Davies, A. Grover, L. Iacovone, J. E. L. Cordova, D. Medvedev, F. O. Maduko, G. Nayyar, S. Reyes Ortega, et al. (2021). “Policies to Support Busi- nesses through the COVID-19 shock: A Firm Level Perspective”. The World Bank Re- search Observer 36 (1), pp. 41–66. Cui, W., J. Hicks, and M. Norton (2022). “How Well-Targeted are Payroll Tax Cuts as a Response to COVID-19? Evidence from China”. International Tax and Public Finance 29 (5), pp. 1321–1347. Fairlie, R. (2020). “The Impact of COVID-19 on Small Business Owners: Evidence from the First Three Months after Widespread Social-Distancing Restrictions”. Journal of Economics & Management Strategy 29 (4), pp. 727–740. Fezzi, C. and V. Fanghella (2020). “Real-Time Estimation of the Short-run Impact of COVID-19 on Economic Activity using Electricity Market Data”. Environmental and Resource Economics 76, pp. 885–900. 21 Forsythe, E., L. B. Kahn, F. Lange, and D. Wiczer (2020). “Labor Demand in the Time of COVID-19: Evidence from Vacancy Postings and UI claims”. Journal of Public Economics 189, p. 104238. Guerrero-Amezaga, M. E., J. E. Humphries, C. A. Neilson, N. Shimberg, and G. Ulyssea (2022). “Small Firms and the Pandemic: Evidence from Latin America”. Journal of Development Economics 155, p. 102775. El-Haddad, A. and C. Zaki (2023). “Storm Survivors: Evidence from Firms in Times of Pandemic”. The Journal of International Trade & Economic Development, pp. 1–34. Hale, T., N. Angrist, R. Goldszmidt, B. Kira, A. Petherick, T. Phillips, S. Webster, E. Cameron-Blake, L. Hallas, S. Majumdar, and H. Tatlow (Apr. 2021). “A Global Panel Database of Pandemic Policies (Oxford COVID-19 Government Response Tracker)”. Nature Human Behaviour 5 (4), pp. 529–538. ISSN: 2397-3374. DOI: 10 . 1038 / s41562-021-01079-8. Hatayama, M., Y. Li, and T. K. Osborne (Feb. 2022). Understanding and Predicting Job Losses due to COVID-19 : Empirical Evidence from Middle Income Countries. Policy Research Working Paper Series 9933. The World Bank. Horvath, A., B. Kay, and C. Wix (2023). “The Covid-19 Shock and Consumer Credit: Evidence from Credit Card Data”. Journal of Banking & Finance 152, p. 106854. Hoy, C. A., L. Simbeye, M. A. A. Malik, A. Koivisto, and M. Maboshe (Aug. 2022). How Have Formal Firms Recovered from the Pandemic ? Insights from Survey and TaxAd- ministrative Data in Zambia. Policy Research Working Paper Series 10139. The World Bank. Humphries, J. E., C. Neilson, and G. Ulyssea (Apr. 2020). The Evolving Impacts of COVID- 19 on Small Businesses Since the CARES Act. Cowles Foundation Discussion Paper No. 2230; NYU Stern School of Business. ILO (2020). COVID-19 and the world of work: country policy responses. Accessed on 30 Nov 2023. IMF (2020). World Economic Outlook: The Great Lockdown. Tech. rep. International Mon- etary Fund, Washington, DC. IMF (2024). Database of Country Fiscal Measures in Response to the COVID-19 Pan- demic. Jain, R. and R. Kumar (2023). “Effect of COVID-19 Lockdown on the Profitability of Firms in India”. Economic Papers 42 (1), pp. 54–71. 22 James, S. (Sept. 2020). Revenue Effects of COVID-19 - September 2020 Update. Working Paper 3698956. SSRN Electronic Journal. DOI: 10.2139/ssrn.3698956. Kawaguchi, D., S. Kitao, and M. Nose (2022). “The Impact of COVID-19 on Japanese Firms: Mobility and Resilience via Remote Work”. International Tax and Public Fi- nance 29 (6), pp. 1419–1449. Khamis, M., D. Prinz, D. L. Newhouse, A. Palacios-Lopez, U. J. Pape, and M. Weber (Jan. 2021). Early Labor Market Impacts of COVID-19 in Developing Countries. Jobs Group Papers, Notes, and Guides 32769257. The World Bank. Kurmann, A., E. Lale, and L. Ta (May 2020). The Impact of COVID-19 on Small Business Employment and Hours: Real-Time Estimates With Homebase Data. Working Papers 20-02. Chair in macroeconomics and forecasting, University of Quebec in Montreal’s School of Management. Mascagni, G. and A. Lees (2023). “The Economic Impact of the Pandemic in Rwanda: An Analysis of Firm-Level VAT Data”. Journal of African Economies 32 (3), pp. 209–236. Olvera, B. C., M. Gonzalez-Sauri, D.-A. H. Moya, and F. Louvin (2022). “COVID-19 in Central America: Firm Resilience and Policy Responses on Employment”. Journal of Policy Modeling 44 (6), pp. 1280–1295. Pomeranz, D. and J. Vila-Belda (2019). “Taking State-Capacity Research to the Field: In- sights from Collaborations with Tax Authorities”. Annual Review of Economics 11, pp. 755–781. Slemrod, J. (2018). Tax Compliance and Enforcement. NBER Working Paper 24799, Vavra, J. (2020). Shutdown Sectors Represent Large Share of All US Employment. Tech. rep. Becker Friedman Institute for Economics at the University of Chicago. WBG (2023). GDP (current US$) indicator. URL: https://data.worldbank.org/ indicator/NY.GDP.MKTP.CD. World Bank (2020). Global Economic Prospects, June 2020. Tech. rep. DOI: 10.1596/ 978-1-4648-1553-9. eprint: https://elibrary.worldbank.org/doi/ pdf/10.1596/978-1-4648-1553-9. Zou, P., D. Huo, and M. Li (2020). “The Impact of the COVID-19 Pandemic on Firms: a Survey in Guangdong Province, China”. Global Health Research and Policy 5 (1), pp. 1–10. 23 Tables and Figures Table 1: Shocks to Firms’ Revenue from COVID-19 Assumed shock Observed shock (VAT sales) Impact sector Industries Y-o-y revenue drop (%) CRI DOM GTM HND Low A: Agriculture, forestry, fishing; 5.0 3.0 3.6 3.7 4.9 B: Mining and quarrying; C: Manufacturing; J: Information and communication; M: Professional, scientific and technical activities; N: Administrative and support service activities; O: Public administration and defence; compulsory social security; Q: Human health and social work activities; T: Activities of households as employers; undifferentiated goods- and services-producing activities of households for own use; U: Activities of extraterritorial organizations and bodies; Z: Other Medium D: Electricity, gas, steam and air conditioning 12.5 3.6 6.0 6.6 11.5 supply; E: Water supply; sewerage, waste management and remediation activities; F: Construction; G: Wholesale and retail trade; repair of motor vehicles and motorcycles; K: Financial and insurance activities; L: Real estate activities; P: Education High H: Transportation and storage; 25.0 14.1 15.8 6.3 15.4 I: Accommodation and food service activities; R: Arts, entertainment and recreation; S: Other service activities Note: This table summarizes the COVID-19-induced revenue shocks for each of the three impact cate- gories: low, medium and high impact. The “industries” column shows the ISIC sections that correspond to each of the three impact categories as used in Vavra (2020). The column labeled “Assumed shock” displays the shock sizes in terms of the year-on-year reduction in sales, assumed for our initial simulation exercise based on Vavra (2020). The columns under “Observed shock” display the sales shocks calculated using monthly VAT data for each country and impact sector. This table is discussed in Sections 2.2 and 5.1. Section 5.1 discusses the details of how the observed sales shocks are calculated. 24 Table 2: Baseline Data on Formal Firms (Fiscal Year 2019) Costa Rica Dominican Republic Guatemala Honduras Ethiopia Uganda (CRI) (DOM) (GTM) (HND) (ETH) (UGA) All Low Med. High All Low Med. High All Low Med. High All Low Med. High All Low Med. High All Low Med. High Impact sectors: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) N firms 27056 6965 15246 4845 50734 26129 19323 5282 16189 3734 10907 1548 21077 6700 9055 5322 12985 4455 7048 1482 14020 4271 7438 2311 Profitable (%) 82.7 77.1 84.7 84.4 66 68.8 63.4 61.4 75.8 73.2 76.7 75.5 79.6 79.3 82 76 77.5 75.9 79.6 72.8 68.9 63.7 73.9 62.2 Profit Margin (%) Profit margin (mean) 3.5 1 3.9 5.6 -4.3 -3.6 -4.8 -6.1 -0.2 -1.4 0.2 0.6 1.5 1.2 2.2 0.5 6.5 5.1 7.5 5.9 -3.3 -5.2 -1.1 -6.5 Profit margin (median) 3.6 3.2 3.4 4.7 2.4 2.3 2.7 1.7 3.3 3.4 3.2 3.4 1.8 1.8 1.9 1.6 7.7 7.8 7.6 7.5 1.3 1.4 1.3 1.2 25 Costs (% Total Costs) Material costs 42.7 37.3 52.1 21 41.7 50.3 32.4 33 46.8 50.6 48.7 24.2 31.6 29.5 38.7 22.1 58.9 55.8 63.9 44 51.1 44.2 61.1 31.8 Labor costs 19.8 20.4 19.1 21 44.2 38 50.4 52.8 24.1 22.8 23.7 30.1 25.5 28.8 22.1 27.2 15.3 16.8 13.1 21.3 11.5 13.5 9.4 14.8 Fixed costs 37.5 42.3 28.7 57.9 13.6 11.3 16.6 13.9 29 26.6 27.5 45.7 41.4 40.2 37.9 48.7 25.2 27 22.2 34.1 37.1 42 29.4 53.1 GDP Per Capta (USD) 12669 8173 4647 2519 840 823 Note: This table shows the baseline characteristics of the firms in the six countries in this study: Costa Rica, Dominican Republic, Guatemala, Honduras, Ethiopia, and Uganda. In each country, the data comprises firms belonging to a balanced panel of firms between 2019 and 2020. The table shows the share of profitable firms, their mean and median profit margin, and the breakdown of costs into labor, material and fixed costs. There are four columns for each country. The first column summarizes the information for all firms, while the following three columns split the full sample of firms into low, medium, and high impact categories. The data set used to construct this table is presented in Section 2.1. Table 3: Simulated vs Realized Outcomes Firm-level outcomes Aggregate outcomes (Means) (As a share of GDP) Revenue Material Labor Fixed Profitable (%) Profit margin Revenue Profit Tax paid Loss (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) CRI Baseline — — — — 82.7 3.5 56.6 3.7 0.7 1.3 Simulation -12.2 -11.4 -7.3 0 54.9 -3.3 52.1 2 0.4 2.6 Realization 10.3 13.3 10.3 6.2 72.9 1.3 65.5 4.3 0.9 1.7 DOM Baseline — — — — 66 -4.3 97.7 6.8 1.8 2.1 Simulation -10.1 -9.4 -7.3 0 61 -7.2 93.2 4.5 1.2 3 Realization -7.9 -9.3 -12.2 6 59.3 -8.9 95.6 6.3 1.7 2.7 GTM Baseline — — — — 75.8 -0.2 87.2 5 1.2 0.8 Simulation -10.9 -10.1 -7.7 0 54.7 -5.2 79.4 2.7 0.7 1.5 Realization -6 -7.1 -3.8 -6.5 69.3 -3.2 83.8 5.4 1.3 1 HND Baseline — — — — 79.6 1.5 147.4 10 2.5 3 Simulation -10.1 -9.6 -7.2 0 40.8 -5.9 143.9 6.4 1.6 5.4 Realization -16.3 -20.2 -11.6 -8.7 69.4 -3.1 134 9.1 2.3 3.8 ETH Baseline — — — — 77.5 6.5 16.8 2.8 0.9 0.5 Simulation -10.5 -10.7 -3.6 0 70.8 2.8 14.4 2.3 0.7 0.5 Realization -4.5 -3.8 -3.3 -4.4 77.9 6 15.4 2.6 0.8 0.5 UGA Baseline — — — — 68.9 -3.3 64.9 3.4 1 1.9 Simulation -10.5 -10.5 -6.7 0 44.4 -9 56.8 2.1 0.6 2.9 Realization -7.2 -6.4 -9.6 -8.8 61.9 -5.7 58.9 3.6 1.1 2.2 AVG Baseline — — — — 75.1 0.6 78.4 5.3 1.4 1.6 Simulation -10.7 -10.3 -6.6 0 54.4 -4.6 73.3 3.3 0.9 2.6 Realization -5.3 -5.6 -5 -2.7 68.4 -2.3 75.5 5.2 1.3 2 Note: This table compares the results from the simulations to the realizations for the year 2020 in each of the six countries analyzed. Firm-level outcomes (columns 1 to 6) are averages relative to the baseline (2019), except for columns 5 and 6, which are levels. Aggregate outcomes (columns 7 to 10) are relative to GDP. See Sections 2.2 and 3 for further discussion of these results and Figure 1 for a graphical illustration. Table A2 shows that the results are very similar when using as a baseline the data from 2018 instead of 2019. 26 Figure 1: Impact of COVID-19 on Firm Activity: Simulation vs Realization (a) Aggregate revenue (percentage change from baseline) (b) Share of profitable firms (percentage points) (c) Aggregate profits (percentage change from baseline) Note: This figure shows the simulated and realized impact of the 2020 lockdowns on firms’ reported revenues and profits, for the six countries in our study. Panel (a) shows the aggregate revenues reported relative to the baseline, Panel (b) the share of profitable firms, and Panel (c) the aggregate (positive) profits reported relative to the baseline. The baseline corresponds to the reported outcome in the year 2019 (red dots), the realization concerns the reported outcome in the year 2020 (blue dots), and the simulations correspond to predictions made in the spring of 2020 based on 2019 data and our assumptions (green triangles). Panels (a) and (c) also include numerical values that represent aggregate revenue and aggregate profits, respectively, both measured at baseline and expressed in billions of 2016 US dollars. The three panels are related to outcomes shown in Table 3. Panels (a) and (c) are related to columns (7) and (8), respectively, with the difference that the values in the table are expressed in levels and as a ratio of GDP. Panel (b) is analogous to column (5). See Section 3 for further discussion of these results. 27 Figure 2: Distribution of Firms’ Profit Margins Note: This figure plots the distribution of firms’ profit margin, defined as the ratio of profits over revenues, for each country, comparing the 2019 baseline to the 2020 realization. The left and right tails are winsorized at -25% and 25% profitability. See Section 3 for further discussion on these results. 28 Figure 3: Revenue Shocks and Input Adjustment (a) Revenue shock relative to baseline by impact sector (b) Revenue shock and input adjustment Note: This figure compares the results from the simulations to the realizations for the year 2020. Panel (a), displays for each country and for the three impact sectors, the realized drop in revenue compared to the simulation. Panel (b), combines all three impact sectors, to show the average change to revenue and the change to each of material, labor and fixed costs. These results are displayed in more detail in Table A3. See Section 4 for further discussion on these comparisons. 29 Figure 4: Year-on-year Changes in Revenue by Firm size Note: This figure ranks firms by deciles of revenue within their country in the base year. For each decile, it displays the median of the distribution of year-on-year changes in revenue for firms in the decile. The dotted lines represent the change in revenues between 2019 and 2018, while the crossed lines show the evolution of revenues between 2020 and 2019. The area between the two lines is colored red for intervals where there is a reduction in the growth of that firm-size decile, and green for intervals with an increase in year-on-year growth. We omit the first decile from the figure, as the year-on-year changes in revenues are very volatile for these firms. See Section 4 for further discussion on these results. 30 Figure 5: VAT Sales Over Time by Impact Sectors Note: This figure plots the year-on-year change in monthly sales from the VAT data for the four (out of six) sample countries for which VAT data is available. The gray horizontal line corresponds to the onset of the pandemic, which we mark as starting in March 2020. The low-impact sector is shown in green, the medium impact in yellow, and the high-impact in red. We use these year-on-year changes to calculate the observed revenue shocks (i.e., shocks calculated with observed VAT data), summarized in Table 1. To do so, we arbitrarily assume a time window of observation of four months since the onset of the pandemic—in other words, we assume that we would have observed the VAT sales data as of June of 2020—and assign the year-on-year sales deviation in the first half of 2020 relative to 2019 as the yearly country-sector size of the shock. See Section 5.1 for further discussion on these results. 31 Figure 6: Impact of COVID-19 on Firm Activity: Augmented VAT Simulations (a) Aggregate revenue (b) Share of profitable firms (c) Aggregate profits Note: This figure shows the simulated and realized impact of the 2020 lockdowns on firms’ reported revenues and profits, for the four countries for which monthly VAT data was available. The simulations that include the VAT data use sector-by-firm-size-decile shocks, except for the Dominican Republic, where the shocks are not size-specific. Panel (a) reports the aggregate revenues reported, Panel (b) the share of profitable firms, and Panel (c) the aggregate profits reported. The dots correspond to observed data from 2019 and 2020, and the triangles to the simulations. These results are discussed in Section 5.1. 32 Figure 7: COVID-19 Restrictions (a) Timing of COVID-19 restrictions in 2020 (b) Restriction stringency and aggregate revenue losses Note: Panel (a) shows the timing and duration of lockdowns in the six sample countries. The red solid line denotes the dates of forced business closure. The blue dashed line marks the dates of strict stay-at-home orders. Data on the timing of lockdowns are from Hale et al. (2021). Panel (b) correlates the index of restriction stringency constructed by Hale et al. (2021) with the average drop in firms’ revenue realized in the six sample countries, using the administrative microdata. This figure is discussed in Section 5.3. 33 Additional Figures and Tables Table A1: Number of Firms Country CIT: Balanced CIT: Unbalanced VAT Tax rates (%) 2019 2020 2019 2020 2019 2020 CIT VAT CRI 27,056 27,056 31,939 28,090 31,374 29,606 10/20/30 13 DOM 50,734 50,734 67,513 69,450 47,722 43,589 27 18 GTM 16,189 16,189 19,331 16,795 88,635 83,618 25 12 HND 21,077 21,077 23,390 21,795 14,140 12,784 25 15 ETH 12,985 12,985 18,473 19,109 – – 30 15 UGA 14,020 14,020 23,015 24,302 – – 30 18 Note: This table presents the firm count for each sample. For the CIT data, we use a balanced sample of firms present in 2019 and 2020 and show the difference with an unbalanced sample. For the VAT data, the sample is semi-balanced: we retain firms that filed at least once every quarter in 2018 and 2019 and appeared at least once in the second and third quarters of 2020, coinciding with the onset of COVID. Disparities in firm numbers between the CIT and VAT samples can be attributed to the fact that not all firms are mandated to file both types of taxes. The last two columns indicate the CIT rate and the general VAT rate for each country in 2020. This table is discussed in Section 2.1. 34 Table A2: Simulated vs Realized Outcomes - Robustness to Changing the Baseline Firm-level outcomes Aggregate outcomes (Means) (As a share of GDP) Revenue Material Labor Fixed Profitable (%) Profit margin Revenue Profit Tax paid Loss (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Panel A: Baseline 2018 CRI Baseline — — — — 85.4 3.5 55.8 3.3 0.7 1.3 Simulation -12.3 -11.4 -7.6 0 55.1 -3.3 49.9 1.6 0.3 2.6 Realization 15.2 15.6 15.2 12.4 72.9 1.3 65.5 4.3 0.9 1.7 GTM Baseline — — — — 74.3 -1 87.5 4.7 1.2 1.1 Simulation -10.7 -10.1 -7.9 0 53.5 -6 76.9 2.5 0.6 1.7 Realization -2.8 -5 -1.7 -1.9 69.3 -3.2 83.8 5.4 1.3 1 HND Baseline — — — — 79.4 1 150.1 10.6 2.6 3.5 Simulation -10.1 -9.5 -7 0 42 -6.7 143.1 6.8 1.7 6.4 Realization -15.8 -16 -8.7 -16 69.4 -3.1 134 9.1 2.3 3.8 Panel B: Baseline 2019 CRI Baseline — — — — 82.7 3.5 56.6 3.7 0.7 1.3 Simulation -12.2 -11.4 -7.3 0 54.9 -3.3 52.1 2 0.4 2.6 Realization 10.3 13.3 10.3 6.2 72.9 1.3 65.5 4.3 0.9 1.7 GTM Baseline — — — — 75.8 -0.2 87.2 5 1.2 0.8 Simulation -10.9 -10.1 -7.7 0 54.7 -5.2 79.4 2.7 0.7 1.5 Realization -6 -7.1 -3.8 -6.5 69.3 -3.2 83.8 5.4 1.3 1 HND Baseline — — — — 79.6 1.5 147.4 10 2.5 3 Simulation -10.1 -9.6 -7.2 0 40.8 -5.9 143.9 6.4 1.6 5.4 Realization -16.3 -20.2 -11.6 -8.7 69.4 -3.1 134 9.1 2.3 3.8 Note: This table analyzes the robustness of the simulation results to changing the baseline year. Panel (a) is analogous to Table 3, where the baseline year is 2019. Panel (b) uses 2018 as the baseline period instead. Firm-level outcomes (columns 1 to 6) are averages relative to the baseline (except columns 5 and 6, which are levels). Aggregate outcomes (columns 7 to 10) are relative to GDP. See Section 3 for further discussion on these results. 35 Table A3: Details of Simulation Outcomes All Low Medium High Revenue Material Labor Fixed Revenue Material Labor Fixed Revenue Material Labor Fixed Revenue Material Labor Fixed (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) CRI Simulation -12.2 -11.4 -7.3 0 -5 -5 -1.8 0 -12.5 -12.5 -6.6 0 -25 -25 -21.7 0 Simulation (VAT) -3.4 -3.2 -2.2 0 -1.7 -1.2 -1.3 0 -2.2 -3.3 -0.8 0 -13.6 -10.3 -12.1 0 Realization 10.3 13.3 10.3 6.2 13.4 12.7 13.7 11.6 13.5 15.2 13.4 10.3 -13.6 -7.9 -13.5 -9 DOM Simulation -10.1 -9.4 -7.3 0 -5 -5 -1.6 0 -12.5 -12.5 -7.1 0 -25 -25 -18.5 0 Simulation (VAT) -5.9 -5.4 -3.4 0 -3.6 -3.6 -1.1 0 -6 -6 -2.8 0 -15.8 -15.8 -9.1 0 Realization -7.9 -9.3 -12.2 6 -3.4 -6.9 -5.4 11 -7.6 -8.9 -11.5 7.4 -28.8 -33.3 -26.6 -17.1 GTM Simulation -10.9 -10.1 -7.7 0 -5 -5 -1.7 0 -12.5 -12.5 -7.6 0 -25 -25 -21 0 Simulation (VAT) -5.5 -5.4 -2.5 0 -3.6 -3.4 -2.1 0 -6.4 -6.6 -2.3 0 -7.2 -7.7 -5 0 Realization -6 -7.1 -3.8 -6.5 -0.8 -1.7 -2.1 -4 -7.9 -10 -2.8 -6.3 -15.4 -16.9 -12.7 -11.4 HND Simulation -10.1 -9.6 -7.2 0 -5 -5 -2.3 0 -12.5 -12.5 -7.3 0 -25 -25 -22.3 0 Simulation (VAT) -8.7 -8.6 -5.8 0 -4.5 -4.3 -3 0 -11.4 -11.7 -6.5 0 -15.2 -15.2 -12.4 0 Realization -16.3 -20.2 -11.6 -8.7 -12.1 -13.4 -6.9 -9.7 -18.1 -25.6 -12.1 -2.6 -29.4 -21.5 -25 -28.4 ETH Simulation -10.5 -10.7 -3.6 0 -5 -5 -1.3 0 -12.5 -12.5 -3.9 0 -25 -25 -13.1 0 Realization -4.5 -3.8 -3.3 -4.4 -3.7 -2.9 3 -3.4 -4.6 -3.2 -6.9 -1.7 -8.2 -14.7 -15.5 -18.5 UGA Simulation -10.5 -10.5 -6.7 0 -5 -5 -1.5 0 -12.5 -12.5 -8.2 0 -25 -25 -20.8 0 Realization -7.2 -6.4 -9.6 -8.8 -0.3 -5.1 1.9 3.9 -10.6 -7.3 -17.2 -16.7 -13 2.2 -12.9 -14.1 AVG Simulation -10.7 -10.3 -6.6 0 -5 -5 -1.7 0 -12.5 -12.5 -6.8 0 -25 -25 -19.6 0 Simulation (VAT) -7.1 -6.9 -3.6 0 -5.2 -5 -2.4 0 -7.5 -7.9 -3.2 0 -13 -12.4 -8.7 0 Realization -5.3 -5.6 -5 -2.7 -1.1 -2.9 0.7 1.6 -5.9 -6.6 -6.2 -1.6 -18.1 -15.3 -17.7 -16.4 Note: This table extends the simulation results presented in Table 3, columns 1-4. Columns 1-4 of this table reproduce columns 1-4 of Table 3 – firm-level outcomes for firms in all sectors – for ease of comparison. Columns 5-16 show these results for each of the three impact sectors. In addition, for each country, a row is added to present simulation results using monthly VAT data to calibrate the size of the sector-firm-size- specific revenue shock. All outcomes are firm-level averages relative to the baseline. See Section 4 for further discussion on these comparisons. 36 Table A4: Correspondence Between Consumption Groups and Industry Sections Weights (%) Consumption Group Industry Section Lockdown Impact CRI DOM GTM HND ETH UGA 01: Food and non-alcoholic G (A, I, T, C) Medium 24.3 25.1 28.8 31.8 53.7 27.1 beverages 02: Alcoholic beverages, G (A, I, R, C) Medium 0.6 2.3 0.3 0.4 4.9 3.9 tobacco and narcotics 03: Clothing and footwear G (C) Medium* 4.0 4.6 7.4 8.2 5.7 7.0 04: Housing, water, D, E (L, B, I, T) Medium 12.5 11.6 12.6 19.2 16.7 10.4 electricity, gas and other fuels 05: Furnishings, household D, E, G, F (T) Medium 5.9 6.5 5.4 6.7 4.7 4.8 equipment and routine household maintenance 06: Health Q (G, S) Low 6.0 5.2 4.2 3.6 1.5 4.7 07: Transport H (D) High 14.8 18.0 10.4 9.1 2.5 10.5 08: Information and J Low 7.8 3.0 5.2 1.7 2.0 4.4 communication 09: Recreation, sport and R High 5.1 4.1 5.6 4.0 0.4 5.0 culture 10: Education services P Medium 5.4 3.7 3.7 3.0 0.2 5.8 11: Restaurants and I High 6.0 8.5 9.2 7.2 5.3 8.7 accommodation services 12: Insurance and financial K Low 0.7 0.0 0.0 0.0 0.0 2.3 services 13: Personal care, social S (G, Z) High 6.8 7.5 7.2 5.2 2.5 5.4 protection and miscellaneous goods and services Note: Each consumption group is matched to industry sections which, in turn, are associated with a lock- down impact group. In order to construct a measure of monthly price indices for each of these lockdown impact groups, we calculate a weighted average of the CPIs of the corresponding consumption groups, weighted by their contribution in the general CPI of each country. Product groups correspond to the di- visions in the 2018 Classification of Individual Consumption According to Purpose (COICOP). *: The lockdown impact for “Clothing and footwear” should ideally be classified as “High risk”. However, for consistency with other sector definitions throughout the paper, it is kept as medium risk. Simulations were also run with this reclassification, and the results showed no significant changes. This table is discussed in Section 5.2. 37 Table A5: Differential Price Change Between Output and Input Prices Needed to Explain the Gap Between Realized and Simulated Profitability Profitability Share (%) Output vs Input Price Growth (%) Baseline Realization Simulation Sim. VAT Simulation Sim. VAT Country (1) (2) (3) (4) (5) (6) Costa Rica 82.7 72.9 54.9 68.3 6.1 1.0 Guatemala 75.8 69.3 54.7 63.7 3.7 0.7 Honduras 79.6 69.4 40.8 44.1 7.0 5.1 Note: This table shows the share of profitable firms and the differential in inflation between output vs input prices needed to match the realized profitability share. Differential price change between output and input prices. Columns (1) through (4) show the share of profitable firms at baseline in 2019, the realized share of profitable firms in 2020, the simulated share of profitable firms in 2020, and the VAT-augmented simulated share of profitable firms in 2020, respectively. Columns (5) and (6) show the differential in inflation between output and input prices needed to match the realized profitability share, respectively. In ordet to obtain the values in columns (5) and (6), we re-run the simulations, adjusting the output price growth (relative to the input price growth) such that our simulations match the realized profitability share in Column (2). This table is discussed in Section 5.2. 38 Table A6: COVID-19 Restrictions and Support Policies Country Stay at home Stay at home Firm close Overall Announced (Strict) (Light) requirement stringency stimulus Share of Days in 2020 (%) Index 0-100 (%) GDP CRI 0.00 0.00 10.66 54.39 1.09 DOM 5.46 64.48 16.94 63.24 0.90 ETH 0.00 0.00 0.00 56.48 1.67 GTM 0.00 52.73 36.07 64.05 3.11 HND 43.44 79.51 21.86 71.77 3.19 UGA 0.00 75.68 11.48 62.57 0.85 Latin America 7.32 51.96 20.81 56.86 3.47 Africa 2.48 32.77 8.07 47.74 2.7 High Income 2.37 23.69 10.37 49.25 10.10 Note: Columns 1-4 of this table use data assembled by Hale et al. (2021) to measure the share of days in 2020 with restrictions enforced (columns 1-3) and the overall restriction stringency index (column 4). Each of these variables is shown for the six sample countries. For comparison, we add the simple average across all of Latin America, all of Africa, and for high-income countries. We define a “light stay-at-home policy” as individuals being required not to leave their homes, with exceptions for daily exercise, grocery shopping, and ‘essential’ trips. A “strict policy” is defined as individuals being required not to leave their homes with minimal exceptions. The closing requirement variable forces the closure of businesses. Additionally, the table displays the average stringency index for each country and region, as calculated by Hale et al. (2021). The stringency index is based on ordinary variables categorizing different kinds of restrictions such as school, workplace, and gatherings. Column 5 uses data from the IMF fiscal monitor database (IMF, 2024) to assess the size of the government’s economic stimulus package. To represent the economic stimulus package as a share of GDP, we use 2019 GDP data from the World Bank (WBG, 2023). See Section 5.3 for further discussion. 39 Table A7: Support Policies Targeting Firms Country Tax remittance and Monetary policies Loans and credit Employment flexibiliza- social contributions tion CRI 1- Elimination of advance 1 - Reduction of the gross in- 1 - Policies for the reduction 1 - Regulation for the pro- payments of profit tax; 2 - terest rate; 2 - Adjusted pro- and extension of credit; 2 cedure of temporary suspen- VAT exemption for commer- visions’ minimum accumu- - Temporary moratorium on sion of employment con- cial leases; 3- Tax exemp- lation. leases; 3 - Renegotiation of tracts. 2 - Regulation for re- tions for the nationalization loans, without affecting the ductions of work hours; 3 of products. risk rating of the debtors. - Recommendation to grant vacations in advance. DOM 1 - Extensions for the pay- 1 - Reduction of interest 1 - Authorized freeze of 1 - Recommendation for re- ment of Income Tax and in- rates 2 - Provisions of liq- debtor ratings and provi- duced working hours and stallment agreements for the uidity for banking entities 3- sions, unchanged risk rat- increased telecommuting; 2 VAT; 2 - Deferral of Social Relaxation of regulations in ing in credit restructurings; - Employers of firms that Security contributions. the financial sector 2 - 60-day grace period for should remain closed should credit line loans; 3 - 90- grant paid vacations to eligi- day extension for updating ble employees. appraisal-related guarantees. GTM 1 - Deferral fo the Impuesto - 1 - Loans to finance working 1 - Creation of a procedure de Solidaridad (ISO); 2 - capital 2 - Creation of funds for work contract suspen- Authorized deferral of em- to support the establishment sions; 2 - Measures to al- ployer social security contri- of credit lines for micro and low and promote telecom- butions; small businesses muting. HND 1 - Income and value-added - 1 - Facilitation for refinanc- 1 - Authorization for the tax extension for small and ing and credit restructuring. implementation of telecom- medium-sized taxpayers; 2 - muting Authorized deferral of em- ployer social security contri- butions; 3 - Additional spe- cial deduction from income tax for those who retain all their employees. ETH 1 - Tax exemptions; 2 - Pro- - 1 - Establishment of a fund 1 - Workers can take unused cessing of Value Added Tax to assist private banks in ad- annual leaves and anticipate refunds is accelerated. dressing debt and relief chal- leaves from the next bud- lenges; getary period. UGA 1 - Authorized deferral of 1 - Reduction of the gross - - employer social security interest rate of 1%; 2 - The contributions; Central Bank has committed to providing liquidity assis- tance to commercial banks facing liquidity distress. Note: This table was constructed using the information on country policy responses to COVID-19 from ILO, 2020. The table shows a summary of the main economic policies announced by the governments in support of firms. We do not know to what extent these policies were actually implemented. See Section 5.3 for further discussion. 40 Figure A1: Stability in the Administrative Data Over Time (a) Log number of firms Guatemala Honduras 10.6 Log of Firms 10.4 10.2 10.0 2010 2015 2020 2014 2016 2018 2020 Year (b) Share of profitable firms GTM HND 1.00 0.75 Share 0.50 0.25 0.00 2010 2015 2020 2014 2016 2018 2020 Year (c) Median profit rate GTM HND 0.2 Profit Rate 0.1 0.0 −0.1 2010 2015 2020 2014 2016 2018 2020 Year Note: This figure shows time series information from the administrative data in Guatemala and Honduras, where we can observe firms for an extended period before 2020. The share of profitable firms is calculated for firms with non-zero total revenue. We can also produce these figures for other countries but would need some additional time for this as we do not have direct access to the long panel. This figure is referred to in Section 2.1. 41 Figure A2: Price indices (a) CPI by country (b) Price index by lockdown impact category and country Note: Panel (a) illustrates the evolution of the Consumer Price Index (CPI) over time for each country, with CPI values normalized to 1 in 2019. The solid black line represents the observed CPI values, and the labels indicate the CPI for 2020, 2021, and 2022. The dashed line in red projects the CPI growth rate from 2012 to 2019 in each series over the entire sample period until 2022. In the cases where the CPI in 2020–2022 followed the same trend as in the previous years, the observed values closely align with the projected inflation rate. Panel (b) repeats the same structure but shows instead the CPI and the lockdown- impact price indices. The black line corresponds to the CPI, while the green, yellow, and red lines represent the price indices for low, medium, and high lockdown impact groups, respectively. The time series for the lockdown impact-specific price indices are available for 2016 through 2022 (except Uganda, starting in 2017). This table is discussed in Section 5.2. 42 Figure A3: Impact of COVID-19 on Firm Activity: Augmented VAT Simulations with Price Change Note: This figure reproduces Figure 6 (a) for Costa Rica, Guatemala, and Honduras, and adds an addi- tional simulation result: The black triangles correspond to the VAT-augmented simulations after accounting for price changes by lockdown impact group. This figure is discussed in Section 5.2. 43 Figure A4: Inventory and Debt by Firm Size (a) Baseline Inventory, Percentage of Turnover (2019) Honduras Costa Rica Guatemala 22 13 10 20 Inventory (% Turnover) Inventory (% Turnover) Inventory (% Turnover) 18 12 9 16 8 11 14 12 7 2.5 5.0 7.5 10.0 2.5 5.0 7.5 10.0 2.5 5.0 7.5 10.0 Firm Size Quantiles Firm Size Quantiles Firm Size Quantiles (b) Inventory Change, Percentage of Baseline (2019-2020) Honduras Costa Rica Guatemala 0.50 1.5 5 Inventory change (% Baseline) Inventory change (% Baseline) Inventory change (% Baseline) 1.0 0.25 4 0.5 0.00 3 0.0 2 −0.25 −0.5 2.5 5.0 7.5 10.0 2.5 5.0 7.5 10.0 2.5 5.0 7.5 10.0 Firm Size Quantiles Firm Size Quantiles Firm Size Quantiles (c) Interest Payments, Percentage of Baseline (2019-2020) Honduras Costa Rica Guatemala 1 6 Interest Payments Change (% Baseline) Interest Payments Change (% Baseline) Interest Payments Change (% Baseline) −2 0 4 −3 −1 −4 2 −2 2.5 5.0 7.5 10.0 2.5 5.0 7.5 10.0 2.5 5.0 7.5 10.0 Firm Size Quantiles Firm Size Quantiles Firm Size Quantiles Note: This figure shows the average 2019 baseline inventory level as a share of turnover (panel a), and the average growth rate in inventory (panel b) and interest payment (panel c) at the firm level, between 2019 and 2020, across size deciles. Ratios are capped at -100 and 100%, to limit large outliers. Deciles are constructed based on 2019 turnover. Variables on inventory are directly available from the tax form for Costa Rica, Honduras and Guatemala. Interest payments information is available for Honduras. For Costa Rica and Guatemala, interest payments is part of a larger category which includes other financial charges. This table is discussed in Section 4. 44 Figure A5: Consumer vs Producer Price Evolution (a) Yearly Inflation (2012-2022) (b) Monthly Inflation (Jan2019-Dec2020)) Note: The figure shows the evolution of consumer and producer prices in the food and beverages sec- tors in Costa Rica. Producer prices are disaggregated into industry sections according to the International Standard Industrial Classification of All Economic Activities (ISIC Rev. 4). The disaggregated producer prices are only available for a subset of sectors in the economy, mainly Energy and Manufacturing. For the purpose of this exercise, we select the producer price index for “manufacturing of food, beverages, and tobacco” (ISIC section C, divisions 10, 11, and 12). Consumer prices, on the other hand, are disaggregated into product groups according to the 2018 Classification of Individual Consumption According to Purpose (COICOP). To match the producer price index, we focus on the corresponding product groups: “Food and non-alcoholic beverages” and “Alcoholic beverages and tobacco”. We then construct the consumer price index as a weighted average of these groups, with weights reflecting their relative importance in the overall consumption basket. Panel (a) shows the yearly price index rates for consumer and producer prices from 2012 to 2022, normalized to 1 in 2019. Panel (b) shows the monthly inflation rates for consumer and pro- ducer prices from January 2019 to December 2020, normalized to 1 in December 2019. Source: Consumer Price Index by consumption group and Producer Price Index by economic activity, Banco Central de Costa Rica. 45