How Have Formal Firms Recovered from the Pandemic? Insights from Survey and Tax Administrative Data in Zambia

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


Policy Research Working Paper 10139
This paper examines how formal firms have been impacted by and recovered from the pandemic, by drawing on two distinct but complementary data sources. This is the first attempt to use both survey and tax administrative data to measure the initial decline and subsequent recovery of firm sales and employment in a low-or lower-middle-income country. The findings of three rounds of follow-up surveys to a standard World Bank Enterprise Survey completed immediately prior to the pandemic are compared to information contained in the universe of value-added tax and personal income tax returns filled by firms during 2020 and the first half of 2021 in Zambia. Despite substantial differences in terms of the breadth and depth of these data sources, they show a very similar pattern. The sales of formal firms recovered from the pandemic far more strongly than their employment levels. By July 2021, both the survey and tax administrative data show that most firms experienced a complete recovery in sales, while levels of employment worsened over the course of the pandemic for many firms.
Two key insights emerge from this analysis. First, formal firms appear to have adjusted their operations in a way that reduced their need for as much labor to achieve the same (or higher) level of sales. Second, if formal firms' reduced reliance on labor persists, lower levels of formal employment in low-and middle-income countries may be a concerning consequence of the COVID-19 pandemic that lingers for years to come. This paper is a product of the Poverty and Equity Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at choy@worldbank.org.

Introduction
Policies introduced to control the spread of COVID-19 have caused unprecedented economic upheaval throughout the world, particularly for firms in low-and lower-middle-income countries that typically received little to no government support. Mobility and economic activity restrictions, alongside higher transaction costs and disruptions in cross-border trade, have hindered the allocation of resources within countries and across industries, lowering aggregate productivity growth (Apedo-Amah et al. 2020). Additionally, the pandemic has substantially shifted consumer demand, which has forced fundamental changes to business practices that are likely to persist in the medium term (Barrero et al. 2020). Therefore, it is imperative to develop a greater understanding of how firms have coped in this extremely challenging environment, especially in lower-income settings.
We examine this issue by drawing on a combination of survey and tax administrative data to analyze how formal firms in Zambia have recovered from  In general, the composition of the formal sector in Zambia is similar to many other medium-sized Sub-Saharan African countries, except there is a greater prevalence of firms in the mining sector (World Bank 2022). As is the case in most countries around the world, at the outset of the pandemic the Government of Zambia introduced restrictions to control COVID-19, which considerably curtailed normal economic activity. We investigate how firms have fared since this initial shock by using three follow-up surveys to a standard World Bank Enterprise Survey (ES) covering a representative sample of 601 formal firms that was conducted immediately prior to the pandemic and by drawing on the universe of monthly value-added tax (VAT) and pay-as-you-earn (PAYE) returns filed by around 20,000 firms throughout 2020 and the first half of 2021. 2 Both data sources have strengths and weaknesses. A key advantage of the survey data is that firms provide extensive details about their operations when completing the question-1 For the purposes of this study, formal firms are defined as firms that are registered with the Zambia Revenue Authority.
2 In Zambia, personal income tax is recorded in PAYE returns.
naire, but a limitation is that respondents only make up a relatively small share of all formal firms and some sectors are excluded (e.g. the mining sector). In contrast, tax administrative data captures all firms in all sectors that file VAT and PAYE returns. However, not all firms file returns each month, and they are only required to provide limited information about their activities when they do (e.g. total monthly sales and number of employees paid each month). By comparing the findings across these different data sources, it is possible to have far greater confidence in what actually took place regarding the recovery (or lack thereof) of formal firms.
We show, in both the survey and tax administrative data, that the sales of formal firms in Zambia recovered from the pandemic far more strongly than their employment levels. The order of magnitude of the initial decline and subsequent recovery of firms varied between the survey and tax administrative data, with the latter presenting a more positive picture.
However, the overall trend was the same. In June 2020, on average, formal firms reported a large decline in sales and a moderate fall in employment. In contrast, by July 2021, most firms had experienced a complete recovery in their sales, while the share of firms reporting that they had decreased their number of employees doubled. Econometric analysis examining the factors associated with an initial decline and subsequent recovery illustrates structural factors (e.g., the type of business activity a firm was involved in) were more closely correlated with changes in sales than changes in employment, whereas the latter was more closely correlated with firm-specific factors (e.g. the level of experience of the top manager).
These results provide several insights about how formal firms have recovered from the COVID-19 crisis. First, there has been a 'decoupling' between the levels of sales and employment among some formal firms in Zambia. Formal firms appear to have adjusted their operations in a way that reduced their need for labor to achieve the same (or higher) level of sales. As such, one way firms were able to be resilient and rebound from the crisis was to reduce their labor inputs, which can often be challenging to lower in more stable settings.
Second, if formal firms' reduced reliance on labor persists, lower levels of formal employment in low-and middle-income countries may be a concerning consequence of the COVID-19 pandemic that lingers for years to come. Consequently, there may be a need for government support to expand opportunities for workers to join formal firms. Third, firm-specific factors, more than structural factors, appear to be associated with whether firms experienced a recovery in employment. This implies that there may be further scope for some firms to fully return their number of employees to pre-pandemic levels by taking actions within their control.
This paper makes two contributions to the existing literature on the impact of COVID-19 on formal firms. First, we draw on two distinct but complementary data sources to track how firms were impacted by and have recovered from the pandemic. Previous studies have either relied on surveys of firms (e.g., Davies et al. 2021;Apedo-Amah et al. 2020;Cirera et al. 2021;Karalashvili and Viganola 2021) or tax administrative data (Angelov and Waldenström 2021;Bachas et al. 2020Bachas et al. , 2021Mascagni and Lees 2022). To the best of our knowledge, this is the first effort to combine these data sources, which allows us to present a far richer picture of how COVID-19 has impacted firms. Extensive efforts are made to identify whether differences (and/or similarities) in the findings across survey and tax administrative data can partly be explained by differences between the samples of firms.
Second, this is one of the first in-depth studies to specifically focus on the recovery of formal firms from the pandemic in a Sub-Saharan African country. The recovery in the region is likely to be distinct from what has occurred elsewhere in the world for several reasons (Aga and Maemir 2021). First, in general, firms in Sub-Saharan Africa received very low levels of government support and it has been argued that this led firms to be more likely to lower wages, lay off workers, and face bankruptcy . Second, at the outset of the pandemic, the operations of firms in the region relied heavily on face-to-face interactions, which were greatly disrupted by COVID-19 lock downs (Bachas et al. 2021;Davies et al. 2021). Lastly, small firms make up the bulk of economic activity in Sub-Saharan African countries, and due to having more limited financial, technological, and human resources, alongside higher dependence on supply chains, were more likely to fare worse than larger firms (Bachas et al. 2021;Davies et al. 2021;Muzi et al 2021).
This paper is structured as follows. Section 2 provides background about the related literature, the private sector prior to the pandemic in Zambia, and the government policies implemented to address COVID-19 in Zambia. Section 3 outlines the data sources that we draw on and the analysis that we conducted to produce the paper's findings. Section 4 presents the findings and Section 5 discusses the implications, as well as areas for future research. Another strand of literature has used tax administrative data to estimate the impact of COVID-19 on the private sector. For example, Bachas et al. (2020Bachas et al. ( , 2021, Mascagni and Lees (2022), and Angelov and Waldenström (2021) Jolevski et al. (2021) suggest that firms that survived the COVID-19

Related Literature
crisis were older and more productive; they also tend to be innovators, use digital technology, and operate in less burdensome business environments. For example, more than half of (reporting) surviving firms in Mongolia adjusted their production or services; 31 percent of these businesses started online business activity, and 40 percent started or increased contact less delivery. were wholesale and retail, mining, construction, manufacturing, and agriculture (which contributed approximately 21, 14.5, 10, 7, and 6.5 percent of GDP, respectively). The five largest sector contributors to GDP growth over the pre-pandemic period were wholesale and retail, information and communication technology (ICT), construction, manufacturing, and finance, which contributed 25, 15, 10, 10, and 6 percent to GDP growth, respectively.

Setting
While real growth has been positive, it has been declining in most major sectors due presumably due to declining consumer demand as inflation and budget deficits increased.
The ambitious infrastructure development program that the government embarked on around 2014-18 significantly contributed to construction sector growth. However, most construction projects had to be stopped due to the rising budget deficits and increased debt service obligations just before the onset of the pandemic. To get a sense of the firm performance just before the COVID-19 pandemic in Zambia, we use employment and sales data from the World Bank ES conducted from September 2019 to March 2020. While the survey excluded some sectors, such as mining and agriculture, most of Zambia's largest economic sectors (wholesale and retail, manufacturing, and construction) were covered, as were rapidly growing sectors such as tourism and ICT. According to the World Bank (2020), Zambia's annual private sector employment grew by an average of 3 percent, which was lower than the Sub-Saharan African average employment growth rate of 7 percent. Real sales in Zambia declined by an average of 2 percent per annum, while the average Sub-Saharan African growth rate was 3 percent. The below-average employment growth may have been due to the negative impact of the 2015-18 electricity outages that affected economic output and thus employment demand in the manufacturing, hotel and restaurant, and construction sectors. The negative growth in real sales is likely due to both reductions in actual economic output due to the electricity crisis and relatively high inflation faced in Zambia just before the COVID-19 pandemic. Other factors, such as poor access to adequate finance, poor access to reliable electricity, and competition from informal sector players, are also thought to have impacted Zambian formal firms' ability to increase sales and employment (World Bank 2020).

COVID-19 in Zambia and government responses
Since the confirmation of the first case on March 8, 2020, Zambia has seen a significant rise in COVID-19 cases, manifesting in four waves. The first wave surged around June/July 2020,

Tax administrative data
There are three main types of tax in Zambia; corporate income tax (CIT), personal income tax (PIT), and VAT. CIT in Zambia is taxed in accordance with the principal activity of the business. The CIT rates range from 10 percent for incomes earned from agriculture to 40 percent for incomes in the banking and telecommunications sectors. All businesses with annual turnover exceeding ZAR800,000 (approximately US$45,000) are expected to register, file returns, and pay income tax. It is compulsory for some sectors, such as activities involving scientific and professional services, to register for CIT irrespective of the size of their annual sales/turnover.
There are few self-reported PIT taxpayers in Zambia, which means the responsibility to pay PIT largely falls on employers. In this form, PIT is a PAYE tax and is withheld at Firms that filed VAT and PAYE returns in 2020 were largely based in urban areas and were mainly involved in wholesale/retail trade and manufacturing. Almost two-thirds of firms were in Lusaka (the capital and by far the largest city). Almost half of firms were involved in wholesale and retail trade, and another 14 percent were involved in manufacturing (these were by far the two most common types of business activities). This is a somewhat similar composition of firms to the survey sample. A noteworthy difference is that the bulk of firms that file VAT and PAYE returns have fewer than five employees (and these firms were disproportionately involved in retail trade). Table 1 summarizes the key differences between the tax administrative and survey data.

Differences between the two data sources
It is important to keep these differences in mind when considering what analysis will be completed and comparing the findings across the data sources.

Analysis
There are two main parts to the analysis. The first focuses on measuring descriptive trends in sales and employment in the tax administrative and survey data over time, and the second examines what characteristics are associated with changes in firm sales and employment throughout the pandemic.

Descriptive trends in the tax administrative and survey data
The first and most straightforward part of the descriptive analysis is to calculate the overall trends in terms of sales and employment from pre-pandemic levels until mid-2021 according to the tax administrative and survey data. The general patterns across the different data sources are compared to see whether they suggest a similar trajectory in terms of the impact of COVID-19 on formal firms in Zambia. The tax administrative data is adjusted for inflation, drawing on monthly inflation rates released by the central bank of Zambia, so that comparisons over time can be made in real terms. The survey data is weighted to ensure it is representative of formal firms in Zambia that share the key characteristics that the original ES sampling frame is based on (e.g., having more than five employees, operating in specific sectors, located in urban areas). The survey weights also factor in the small amount of attrition between the baseline and follow-up surveys; however, this has negligible impact because there was a very high response rate across all three rounds. In addition, both data sources are winsorized at 2.5 and 97.5 percent to ensure that the average trends are not entirely driven by outliers.
The next part of the descriptive analysis is to solely use the tax administrative data for the subset of firms that share similar characteristics to the firms that participate in the survey. The first step is to only focus on firms that file each month (i.e. they are 'perfectly compliant'). This is necessary as the overall trends in the tax administrative data are partly influenced by which firms choose to file in a given month and there is seasonality in when firms choose to file. A drawback from restricting the focus to these firms is that they are likely to be quite distinct from other formal firms. For example, they may be quite profitable and/or large so each month they always exceed the threshold for filing a tax return, whereas other firms may have sales that fluctuate around the threshold for filing. Another difference between the 'perfectly compliant' firms and other formal firms is that they may have a lower risk tolerance and/or greater exposure to the tax authority, which means they are less willing to avoid paying taxes. These characteristics could also be associated with other aspects of business operations (e.g. investment in R&D).
The second step in aligning the two data sources involves solely focusing on firms for which there are five or more employees and dis-aggregating the analysis by firm size. Specifically, we look at the trends over time across both data sources for 'perfectly compliant' firms with 5-19 employees (small), 20-99 employees (medium), and 100+ employees (large). These types of firms are also quite distinct from those with fewer than five employees (micro); in particular, it is expected they do not experience the same extent of fluctuations in sales and employment levels. A shortcoming of this approach is that the VAT returns do not include information about the number of employees of each firm. As such we merge the VAT and PAYE return data and can only report on the subset of firms that provided both types of returns in a given month. Once again this is likely to mean the results may not be generalizable beyond firms that share these characteristics.
The final part of the descriptive analysis we conduct is to examine the trends from both data sources based on characteristics other than firm size. The three most notable characteristics are in terms of sector, location, and firm age, as this information is available for all firms in both data sources. The sector analysis is decomposed into three categories that are identical across the survey and tax data (manufacturing, retail and wholesale services, and other). The location analysis in both data sources focuses on a simple binary split between firms in Lusaka and firms everywhere else.

Determining the covariates of recovery
Both data sources, but particularly the survey data, provide the opportunity to explore what firm characteristics are associated with experiencing an initial reduction in sales and employment, as well as a recovery by 2021. This analysis complements exploring the descriptive trends because through conducting econometric analysis we can identify what firm characteristics appear to be most closely correlated with firm performance. It is important to emphasize that this regression analysis is not causal as there is no exogenous variation.
We cannot rule out, among other things, reverse causality and omitted variable bias. For example, on the former, if we find a positive relationship between receiving government support and recovering from the pandemic, we cannot be sure whether this is because firms that received support were better placed to recover anyway or whether government support actually aided their recovery. An example of the latter is that both data sources (partic-ularly the tax administrative data) only have a limited number of variables and there are likely to be other factors, such as the determination of the firm manager/owner, that may be impacting firm performance. These caveats need to be kept in mind when examining the findings of the econometric analysis. For both data sources we conduct an ordinary least squares (OLS) regression in the form of a linear probability model. 4 Specifically, we have four main outcomes (Y) that are in the form of a dummy variable. These are: (1) whether a firm experienced a decline in sales compared to pre-pandemic levels in June 2020; (2) whether a firm experienced a decline in employment compared to pre-pandemic levels in June 2020; (3) whether a firm experienced a complete recovery in sales compared to pre-pandemic levels in July 2021; and (4) whether a firm experienced a complete recovery in employment compared to pre-pandemic levels in July 2021. The econometric model can be expressed formally as follows: whereby, β 1 captures the extent to which the size of a firm (small, medium or large) is associated with an outcome, β 2 captures the extent to which the age of a firm (in years) is associated with an outcome, β 3 captures the extent to which the activity of a firm (manufacturing, retail and wholesale services or other) is associated with an outcome, β 4 captures the extent to which the location of a firm (Lusaka or elsewhere) is associated with an outcome, β 0 is the intercept and ϵ is the model error term.
For the survey data we conduct further analysis to fully utilize the richer information available on respondents. First, we rerun the regression above and include additional baseline variables captured in the baseline ES (top manager's years of experience, >5 competitors entering the market in the last two years, new products introduced over the last three years, access to finance is a very severe/major obstacle, training is offered to permanent employees). Second, we rerun the regression above with the variables captured in the baseline ES and include further variables captured in the follow-up survey rounds reflecting how firm circumstances changed during the pandemic (changed business practices, received any cash support from government). A noteworthy limitation of this analysis is that there is a nontrivial number of missing values for several of these survey questions. As such, as a robustness check we present in Appendix B the analysis excluding all respondents with missing values.

Survey findings
According to the survey data, on average, formal firms in Zambia experienced a very large decline in sales in 2020 and a substantial recovery in the first half of 2021. Figure 3 shows the average decline in sales (compared to the same month in 2019) was 42.7 percent in June 2020, 33.9 percent in December 2020, and 16.4 percent in July 2021. The results dis-aggregated by firm size are striking. By July 2021, the average sales of large firms had fully recovered to 2019 levels; in contrast, small and medium-sized firms reported a larger initial decline and a slower recovery.
These average changes in sales somewhat mask variation between firms. Figure 4 shows that most firms reported a decline in sales (compared to the same month in 2019) in June and December 2020, but by July 2021 less than half of the firms reported a decline in sales. As was the case for the average change, large firms fared much better than small and mediumsized firms. In July 2021 only around one-third of large firms reported a decline in sales (compared to the same month in 2019), and almost half reported an increase in sales.

Tax administrative data findings
According to the tax administrative data, the sales of formal firms in Zambia followed a consistent seasonal pattern across the calendar year and were relatively stable in real terms in the years prior to 2020. This was followed by unseasonably low levels of sales during the strictest lockdown period in April and May 2020, followed by a strong recovery to near record high levels by the end of 2020. Figure 5 shows the log of reported monthly sales since January 2016 according to all VAT returns that were filed by firms, with the sales amounts indexed to January 2020. Sales during the COVID-19 lock downs were around 10 percent lower than January 2020 levels, whereas in the same months in 2019 sales were around 10 percent higher than January 2020 levels. The recovery by December 2020 is notable as sales were over 20 percent higher than January 2020 levels, whereas in December 2019 sales were around 15 percent higher than January 2020 levels.
As was the case for the survey data, examining the overall trend in sales based on VAT returns of all filing firms masks significant heterogeneity between firms. Figure 6 shows the trends in nominal sales dis-aggregated by firm size and solely focusing on perfectly compliant firms (examining changes in sales in real terms shows a similar pattern). At least two key patterns emerge. First, these results indicate an initial decline as almost half of firms reported a decline in sales in June 2020, but by July 2021 there were signs of recovery as almost twothirds of firms reported an increase in sales. Second, consistent with the survey data, the sales of a higher share of large firms have recovered more strongly than was the case for small and medium-sized firms.  Source: authors' calculations using the tax data

Summarizing the findings across the survey and tax administrative data
Both data sources illustrate that there has been a recovery in sales for most formal firms.
Initially there was a large decline in sales during the period of the strictest lockdowns as almost half of firms in the tax administrative data, and close to 90 percent in the survey data, reported lower levels of sales than in the same month in 2019. However, since this time there has been a dramatic improvement in the sales of firms. These trends are likely to be quite robust as there are consistent patterns across data sources that are rather distinct in terms of their breadth and depth.

Survey findings
The survey data suggests that there was an initial moderate decline in the number of permanent employees and no persistent signs of recovery since this time. Figure 7 shows the average decline in employment compared to February 2020 was 10.8 percent in June 2020, 2.7 percent in December 2020, and 24.7 percent in July 2021. Any indication of recovery in employment by the end of 2020 clearly dissipated by mid-2021, when the reported fall in permanent employment was even larger than the initial fall. There was some variation by firm size whereby medium-sized firms were the most likely to report improvements in December 2020, but by July 2021 there were no differences by firm size. These average changes in employment in the survey data are largely consistent with analysis of the share of firms that experienced a change in employment since February 2020. Figure 8 shows that most firms reported a similar number of permanent employees in June 2020 as in February 2020; however, over time more firms reduced their number of employees.
By July 2021, more than half of firms reported a decline in employment, with less than one in five increasing their number of employees. As was the case for the average change, there was little variation between firms by their size (determined using the baseline ES from 2019).

Tax administrative data findings
According to the tax administrative data, the average number of employees of all firms filing PAYE returns in Zambia declined abruptly at the start of the pandemic and slightly worsened over the next 12 months. Figure 9 shows that on average there were just over 24 employees at firms filing PAYE returns in January and February 2020, but this fell to around 22 employees in March 2020, which represents a decline in employment of almost 10 percent.
By February 2021, the average number of employees had fallen to 20, which suggests that around one in five jobs in formal firms no longer existed 12 months into the pandemic. There has been a slight recovery since this low point, but employment levels are still far from what they were pre-pandemic. Source: authors' calculations using the tax data These average changes in employment in the tax administrative data are largely consistent with analysis of the share of firms that experienced a change in employment since February 2020. Figure 10 shows the trends in employment dis-aggregated by firm size and solely focusing on perfectly compliant firms. This indicates a worsening of the employment situation throughout 2020, with no signs of recovery in 2021. In June 2020, around twothirds of firms reported no change in levels of employment since February 2020, but this fell to around one-third of firms by July 2021. Over this period the share of firms reporting a decline in employment almost doubled. There were few substantive differences between firms by size. Source: authors' calculations using the tax data 4.2.3 Summarizing the findings across the survey and tax administrative data Both data sources paint a sobering picture about how the pandemic has impacted the employment levels of formal firms. There was an initial moderate decline in employment and there is evidence to suggest that over time the situation has only worsened. Regardless, there are clearly no meaningful signs of recovery, with a non-trivial share of people who were employed in formal firms in Zambia prior to the COVID-19 lockdowns no longer having a job even almost 18 months later. As discussed at length in Section 3, these data sources are quite different, but complementary. In the case of the survey data, em-ployment refers solely to full-time employment, which may be less volatile than all types of employment captured in the tax administrative data. The fact that we see the same pattern in both the survey and tax administrative data illustrates how robust these findings are likely to be.

Covariates of recovery
We conduct econometric analysis to understand what factors are associated with firms experiencing an initial decline in sales and employment as well as a recovery. Table 2 reports on the factors associated with a decline in sales and employment levels in June 2020 from pre-pandemic levels, while Table 3 reports on the factors associated with a recovery in sales and employment to pre-pandemic levels by July 2021. As discussed at length in Section 3.2, we are not claiming that there is a causal relationship between the factors that we examine and firm sales and employment. Rather, the findings that follow show which factors are most closely correlated with firms experiencing a decline and/or recovery. In the discussion that follows we make an important distinction between structural (e.g. location of firms) and firm-specific (e.g. management experience) factors as it is likely that policy makers can more easily influence the latter.  (1) in Section 3 of the paper. Columns (1) and (5) present the results for the tax administrative data. Columns (2)-(4) and (6)-(8) present the results for the survey data. * p < 0.1, * * p < 0.05, * * * p < 0.01.  Table 2 shows that there were key structural factors associated with firms experiencing a decline in sales, but more firm-specific factors emerge regarding declines in employment. A decline in sales was closely correlated with business activities being outside of manufacturing and wholesale/retail trade. These firms were largely involved in services (other than retail), which were probably substantially impacted by social distancing restrictions introduced to reduce the spread of COVID-19. Larger firms were far less likely to report a decline in sales by June 2020, which may be because they have greater stability in operations. In addition, firms that reported changing their business practices in response to the pandemic were less likely to experience a decline in sales. In terms of employment, firms that faced large or very severe obstacles to access to finance immediately prior to the pandemic (according to the baseline ES) were more likely to reduce their number of employees. Firms that had reported introducing new products in the three years prior to the pandemic were far less likely to report reducing their number of employees. Table 3 shows that a combination of structural and firm-specific factors were associated with firms ex-periencing a recovery in sales and/or employment by July 2021. Recovery in sales was more common among larger firms and those in manufacturing (these factors were also associated with less likelihood of a decline in sales initially). In addition, the top manager's level of experience was positively associated with a recovery in sales. A lack of recovery was more common among firms that were in a partic-ularly competitive market (immediately prior to the pandemic they stated more than five competitors had entered their market in the previous two years). In regard to a recovery in employment, this was positively associated with top managers having more experience and receiving government support. A lack of recovery in employment was more common among older firms according to the survey data (the opposite was the case according to the tax administrative data) and those outside of manufacturing and retail trade.  (1) in Section 3 of the paper. Columns (1) and (5) present the results for the tax administrative data. Columns (2)-(4) and (6)-(8) present the results for the survey data. * p < 0.1, * * p < 0.05, * * * p < 0.01.

Explanations for the findings of the study
This study has illustrated that, on average, the sales of formal firms in Zambia have recovered from the pandemic far more strongly than their employment levels. These findings suggest that formal firms have adjusted their operations in a way that reduces their need for labor to achieve the same level of sales. In other words, there has been a shift in the capital-labor mix that formal firms use, away from labor. If this is a persistent change, lower levels of formal employment in low-and lower-middle-income countries may be a concerning consequence of the COVID-19 pandemic that lingers for years to come. This pattern of a weak recovery of formal employment has also been observed in several emerging-market economies (Andaloussi et al. 2022). Ultimately, this means more people are in vulnerable employment situations that lack a formal safety net and there is less income tax being paid to fund the provision of public goods and services. Detailed investigation of the survey and tax administrative data does not suggest that the overall findings are due to improvements in formal sector labor productivity. There is no evidence of existing firms reporting higher average wages for the employees they retained or large shifts between relying on full-time compared to casual employees. In addition, the tax administrative data does not suggest there have been a large number of new firms entering the formal sector. 5 Rather, a more compelling explanation for what we observe is that many firms have tried to rebound from the crisis by reducing (in some instances potentially unproductive) labor inputs, which can often be challenging in a more stable setting. An unavoidable limitation of this study is that the focus is on only formal sector employment, which means that it is entirely possible that people who were previously in formal employment have moved to the informal sector and maintained (or even increased) their productivity.

Implications for policy makers
The key implication for policy makers from this study is that there is a clear need to foster greater recovery of employment among formal firms in Zambia. Non-trivial numbers of workers who were previously employed by formal firms no longer had formal jobs even over a year after the pandemic began. This means that across the economy as a whole, it is likely that formal employment has not rebounded to anywhere near pre-pandemic levels. There may be a need for government support to expand opportunities for workers to join formal firms. A potentially promising sign that emerges from the econometric analysis is that firmspecific factors, more than structural factors, appear to be associated with whether firms experienced a decline or recovery in employment. This may imply that there is greater scope for specific actions to be taken by firms to increase their number of employees. If structural factors were more closely correlated with changes in employment (e.g. the sector that firms operate in), there would potentially be far less scope for policy makers to have an influence.

Contribution to knowledge
We draw on two distinct but complementary data sources that illustrate this same pattern.
To the best of our knowledge, this is the first attempt to use both survey and tax administrative data to track how firms have recovered from the pandemic. We have carefully examined the strengths and weaknesses of both data sets to try to ensure as much comparability as possible. We hope that this effort will be replicated and improved upon elsewhere as triangulating findings across different data sources can provide both researchers and policy makers far more confidence in the robustness of the results. Particularly, the use of tax administrative data as a low-cost way to monitor economic activity is rather novel, and this study shows how it can complement the largest existing effort to collect survey data from firms across countries, which is undertaken by the World Bank Enterprise Analysis Unit.   (1) in Section 3 of the paper. * p < 0.1, * * p < 0.05, * * * p < 0.01. 1 The omitted category is manufacturing. 2 The omitted category is Small. Manager experience: Defined as the number of years experience of the top manager in the firm. Competitive market: Dummy variable for whether a firm has had more than five competitors entering their market in the previous two years. Introduced new products: Dummy variable for whether a firm introduced new products in the previous three years. Limited finance: Dummy variable for whether a firm stated that access to finance was a very severe or major obstacle for their business. Trained employees: Dummy variable for whether a firm stated that they offered their permanent employees training. Changed practices: Dummy variable for whether a firm changed their business operations in response to the COVID-19 pandemic. Receive gov support: Dummy variable for whether a firm received some kind of cash support from the government to help deal with impact of the pandemic.  (1) in Section 3 of the paper. * p < 0.1, * * p < 0.05, * * * p < 0.01. 1 The omitted category is manufacturing. 2 The omitted category is Small. Manager experience: Defined as the number of years experience of the top manager in the firm. Competitive market: Dummy variable for whether a firm has had more than five competitors entering their market in the previous two years. Introduced new products: Dummy variable for whether a firm introduced new products in the previous three years. Limited finance: Dummy variable for whether a firm stated that access to finance was a very severe or major obstacle for their business. Trained employees: Dummy variable for whether a firm stated that they offered their permanent employees training. Changed practices: Dummy variable for whether a firm changed their business operations in response to the COVID-19 pandemic. Receive gov support: Dummy variable for whether a firm received some kind of cash support from the government to help deal with impact of the pandemic.