The effects of COVID-19 on employment, labor markets, and gender equality in Central America

The database combined the data obtained from several sources. Firstly, World Bank Enterprise Surveys conducted before the pandemic – El Salvador (2016), Guatemala (2017), Honduras (2016), and Nicaragua (2016) – were combined with two follow-up COVID-19 surveys for the same sample of firms as the enterprise surveys. Round 1 of the COVID-19 follow-up surveys was conducted in June and July 2020 and Round 2 from November 2020 to early January 2021. For each country, data from two further sources at the country level (but varying on a daily basis) were added to the combined survey data for each country.

The first of these were data on the strength of government measures to contain the pandemic, taken from the containment and health index and its component indexes produced by the Blavatnik School of Government, Oxford COVID-19 Government Response Tracker | Blavatnik School of Government (ox.ac.uk) The second were morbidity and mortality data for the COVID-19 pandemic obtained from Coronavirus Pandemic (COVID-19) – Statistics and Research – Our World in Data Both sets of data are at the country level at daily intervals. Although they do not, of course, vary by firm, they are capable of significant variation over time. For each enterprise, they were matched with the precise dates of their Round 1 and Round 2 COVID-19 follow-up interviews. The final step was to combine the resulting data files for each country into a single “regional” file.

The sample of enterprises responding to the original enterprise survey comprised a total of 1,427 firms (excluding those who reported no sales or employees). Of these, 808 responded to the Round 1 COVID-19 follow-up survey, and 827 responded to the Round 2 survey. Further information on the World Bank Enterprise Surveys and the follow-up COVID surveys can be found here: Enterprise Surveys Indicators Data – World Bank Group.

Table 1 reports the changes in employment for the sample of enterprises (compared to December 2019).

Unsurprisingly, there is a substantial reduction in employment reported by firms in both the Round 1 and Round 2 surveys. Losses in female employment are almost invariably higher than for overall employment at the time of the Round 1 surveys and higher for all countries other than El Salvador at the time of Round 2. In El Salvador, small firms have suffered greater losses in employment than larger ones, but in all the other three countries, employment in medium-sized firms has been hit the hardest.

Table 2 reports a similar analysis but by sector and gender.

These data show a huge loss of employment for the hospitality sector at the time of the Round 1 survey, with an overall reduction of 41% overall and an almost complete loss of female employment. By Round 2, these employment reductions were high (26% overall and 30% of females) but less extreme. Other sectors experienced a much greater loss of employment by Round 2: fabricated metal products (which also reported large reductions in employment in Round 1) reported an overall reduction of 20% but 96% of females; nonmetallic mineral products reported a loss of 46% overall but 93.5% of females. A number of sectors reported increases in employment compared to the rate before the pandemic – at Round 2, these included textiles, chemicals, plastics and rubber, and electronics. Media reports have suggested that the effect of COVID-19 on firms is much more sector specific than would be typical for a more normal economic downturn. These data confirm this insight.

There are two important points with respect to the interpretation of Tables 1 and 2. Firstly, the data report changes in full-time permanent employment. Similar data were not collected for temporary, part-time, or casual labor. It is impossible to know, but one might expect job security to be weaker for these groups and, if so, for these data to underreport the full effect on employment.

A second key point is that the data only include those firms that survived long enough to report some employment at the time of the two follow-up COVID surveys. These surveys did ask firms who participated in the earlier enterprise survey whether they had permanently closed as a result of COVID-19. Very few (25 firms) responded accordingly, but only 827 firms (58% of the original enterprise sample) completed the Round 2 follow-up survey. Some of the approximately 600 “missing” firms may have chosen to not cooperate, some may have permanently closed as a result of circumstances unrelated to COVID-19, but it is likely that there were a number of firms that permanently closed as a result of COVID-19 and could not be contacted. The loss of employment from these permanent closures is not included.

Overall, the changes in full-time permanent employment for surviving firms since December 2019 would have been sufficient to generate serious concern had events been the result of a typical recession. That they are likely to understate the true picture shows that employment (including the impact on gender equality) is an important feature of the economic consequences of the pandemic for Central America.

The surveys allowed us to clearly identify the implications for gender equality in employment. The use of enterprise data to analyze the impact of COVID-19 has many advantages, but they do not otherwise identify many key characteristics of employees, which would allow a wider discussion of the impact on equality. The original enterprise surveys (but not the COVID-19 follow-ups) did contain information on skill categories and the proportion of workers at certain education levels. We calculated the “implied” effect on different categories of employees by, for instance, each firm, assuming that they employed workers in the same proportion at the time of the COVID-19 follow-up surveys. That is, firms that employed a high proportion of skilled production or university-educated workers were assumed to have continued to do so. The results are presented in Table A1. In general, they suggest that the least-skilled and least-educated workers faced marginally smaller reductions in employment than others.

As discussed earlier, the survey does not provide sufficient evidence to analyze permanent firm closures. In any case, the main impact of such closures would be on employment and not on other labor aspects. Temporary workplace closures are different because any loss of labor time is, by definition, not permanent. They do, however, impose substantial costs, which must be borne by one or more of employees, firms, or the government. Table 3 provides information on temporary closures and the implied loss of labor time (the number of weeks the firm was closed multiplied by the number of its full-time permanent employees). All of the four sample countries recorded workplace closures at the time of the first COVID-19 survey, but only El Salvador and Guatemala reported such closures at the time of the second survey.

At the time of the first COVID-19 survey, firms for all four countries reported an average temporary closure of just >5 weeks, implying an overall loss of just >322,000 labor weeks for the sample. Average firm closures were lowest for Nicaragua (1.7 weeks) and highest for Honduras (7.8 weeks), although the total number of lost labor weeks was the highest for El Salvador. The longest average temporary closures were recorded for the textiles and garment sectors (an average of just >8 weeks in both cases) and the lowest was for the food industry (just <2 weeks). For each country, temporary closures were, on average, the longest for small firms and the shortest mainly were for large ones.

At the time of the second survey, there were no recorded workplace closures for Honduras or Nicaragua, so the data in Table 3 are for El Salvador and Guatemala only. Across both these countries, an average of further temporary closures of 2.9 weeks was reported, with an implied loss of just <159,000 labor weeks. For El Salvador, further closures by the second COVID-19 survey were longer for large firms, and for Guatemala, the longest closures were for medium-sized firms. There were no reported further temporary closures for three sectors – wood, nonmetallic mineral products, and electronics. The highest reported temporary closure was for information technology (IT) services, but this may reflect the scope for remote working in that sector. Garments and construction were subject to the next highest average periods of closure (5.7 weeks and 4.8 weeks, respectively).

Table 4 provides details of furloughs, expressed as a percentage of full-time permanent employees, for both all employees and females. Again, all four countries reported furloughs in the first COVID-19 survey, but neither Honduras nor Nicaragua did in the second survey.

At the first COVID-19 survey, the proportion of females furloughed was lower in the sample (17%) than for all employees (21%). The highest proportion of furloughs (for all employees) was recorded for Guatemala (just <30%) and the lowest for Nicaragua (just <11%). With the exception of Guatemala, furloughs were more prevalent in small firms. For females, the highest proportion furloughed was again in Guatemala (21%) and the lowest in Nicaragua (10%). At the time of the first survey, there was considerable variation in the proportion of both all workers and females subject to furloughs. For all employees, the highest furlough rates were for textiles, garments, and hospitality and, for females, in the garments and hospitality industries.

At the second COVID-19 survey, furloughs were only recorded for El Salvador and Guatemala. El Salvador exhibited a higher proportion of female furloughs (11.7%) than for all employees (10.3%). For Guatemala, the proportion of females was very slightly lower than for all employees (5.4% against 5.5%). For the two countries combined, the proportion of additional furloughs at the Round 2 survey was higher for females than for all employees (8.8% compared to 8.4%). Sectors with high rates of furloughs at Round 2 included tobacco, machinery, and hospitality. For each of these, the proportion of females furloughed was significantly higher than for all employees.

Table 5 provides details on the proportion of firms that cut either wages or hours of work, or both.

Overall, about 26% of firms reduced the salaries or wages of their employees, and almost a third (32.4%) reduced the hours of work. Cuts in wages were most prevalent in Guatemala and least common in Nicaragua. Reductions in working hours were most prevalent in Honduras and least used in Guatemala. Wage cuts were most prevalent in some manufacturing sectors (transport, machinery, nonmetallic mineral products, and garments) and in construction. Reductions in working hours were also most common in manufacturing (leather, transport, machinery, and furniture).

Table 6 provides details of the average number of weeks that a firm would be expected to survive with no sales. By country, the average expected survival time varies from just <8 weeks to just >9 weeks. This should be compared to temporary firm closures that have already occurred (see Table 3). On average, the expected firm survival times under fresh closures are greater than the lengths of previous temporary closures but not markedly so. Being able to rely on external finance markedly improves country averages by somewhere between 2 weeks and 4 weeks. The expected times vary considerable by sector, with manufacturing (leather, machinery, and transport equipment) predicting some of the lowest expected survival times. Construction, despite being harder hit by past temporary closures, has relatively high projected survival times.

The key point here is that these figures are averages. Compared to past temporary closures, it is reasonable to suppose that there would plenty of survivors from another, similar round. However, it is also clear that there would also be a number of nonsurvivors for which temporary closures would become permanent. Minimum survival times vary by sector, from 0 weeks to 8 weeks with no sales and from 0 weeks to 15 weeks with no sales and external finance.

Table 7 examines the changes in the value of firm sales from the time of the second COVID-19 survey (mainly November and December 2019) and the same months in 2020. Overall, the (unweighted) average change in sales for firms was a reduction of just less than one quarter of their sales from 1 year previously. Several sectors had a majority of firms whose sales were 50% or lower than that 1 year previously – garments, wood, furniture, and hospitality. In general, small firms faced proportionately larger losses of sales than larger ones. At the country level, the average proportionate change in sales was comparable for three of the four countries but was markedly higher for Honduras.

Two sectors exhibited a modest average gain in sales – tobacco and electronics. All other sectors recorded an average reduction in sales. Particularly large average reductions (≥39%) were recorded for garments, wood, construction, and hospitality.

Across the full sample, just more than three quarters of the firms had neither received nor expected to receive government support. Firms with some foreign ownership were substantially less likely to have received or to be expecting to receive government support. A much higher proportion of firms (almost 40%) in El Salvador had either received or expected to receive government support. This compares to 15% in Guatemala, just >10% in Honduras and just > 6% in Nicaragua. A number of sectors that reported proportionately large falls in sales (Table 7) – garments, wood, construction, and hospitality – had varying degrees of support. Of the firms in the garment sector, fewer firms than the overall average (68% compared to 77%) had either received or expected to receive government support. Firms in the construction (82%) and wood (87.5%) sectors were more likely to not receive or expect support and firms in hospitality were only marginally less likely to be nonrecipients (72%) (Table 8).

The reasons for not receiving government support are set out in Table 9. A lack of awareness of the available support was modest (<20% of firms) for El Salvador and Guatemala but more substantial in Honduras (31%) and Nicaragua (24%). Nearly 30% of firms in El Salvador, but much lower proportions in Honduras and Nicaragua, were ineligible for support. An important reason in both El Salvador and Guatemala was that firms had no need for support.

Table 10 summarizes the enterprise responses to the most-needed support measures. Tax reductions or deferrals were by far the most preferred option in all countries other than in El Salvador (and still the second most popular option there). Access to new credit was the most preferred option in El Salvador but more of a minority choice in the other countries. In none of the countries were either wage subsidies or technical support popular choices by enterprises.

As the enterprises in the sample of Central American firms typically experienced significant reductions in sales revenues (Table 7), an inevitable consequence is that they faced a loss of liquidity, affecting the ability of firms to survive the COVID-19 crisis. Table 11 shows that nearly 60% of firms experienced a decrease in liquidity. A much lower proportion of firms with some foreign ownership (48%) experienced a decrease in liquidity. In each of the four countries, small firms were much more likely to face decreases in liquidity than larger ones. A very high proportion of firms in publishing and printing, transport machinery, wood, and paper experienced decreased liquidity. No firms in the tobacco or electronics sectors reported decreased liquidity. Of the service sectors, hospitality had by far the highest proportion of firms (70%) that experienced decreased liquidity.

Credit is an obvious response to a temporary lack of liquidity and, equally obviously, enables firms to survive, thereby sustaining higher levels of employment than would otherwise be the case. Table A2 provides details of the average changes in debt between January and September 2020. Across the full sample, the average effect was an increase in debt of 6.3%, but there were important variations. Firms with some foreign ownership reported no change in their debt. Smaller firms typically reduced their debt over this period. It is not immediately apparent why this was so. Although it is not a topic covered in this paper, it would be interesting to know whether this reflected decisions by small firms to reduce leverage or decisions by lenders based on perceptions of increased risk arising from the pandemic.

As Table 12 shows, a significant proportion of firms (32%) applied for a loan since the outbreak of the pandemic. In Guatemala and Nicaragua (and to some extent in Honduras), the proportion of small firms applying for a loan was much lower than that of larger firms. Sectors with a high proportion of firms applying for loans included the manufacturers of paper and transport machinery (60% in both cases).

Table A3 provides a summary of the main reasons for not applying for a loan, given by those firms who did not seek a loan, and the outcomes of the application for those that did. Unsurprisingly, the most common reason for nonapplication was that the firm had no need. Small but nontrivial proportions of firms did not apply because of complex procedures, a belief that the application would not succeed, or because interest rates were unfavorable. In three of the four countries, ≥75% of loan applications were approved in full or in part. The exception – El Salvador – had 45% of applications still in progress at the time of the second COVID-19 follow-up survey.

Table 13 presents the responses by firms to the question on potential obstacles to doing business in the original enterprise surveys. That is, they reflect the problems perceived by businesses before the outbreak of COVID-19. It is likely that such problems have not been since eliminated and coexist with those added by the pandemic.

Reading down the columns shows that, in each country, political instability and corruption were rated as among the most serious obstacles to business overall. In El Salvador and Guatemala, crime was relatively highly rated. In Guatemala and Honduras, the courts were ranked as one of the higher obstacles, and tax (rates and administration) was perceived as a problem in Honduras. Overall, these responses suggest that the capacity of local institutions to safely see firms through the COVID-19 crisis may be imperfect.

It is impossible to tell for certain from the available data, but a large number of firms were “missing” from the original enterprise surveys when both COVID-19 follow-ups were conducted. It is likely that COVID-19 did result in a significant number of firm closures, with a resulting loss of employment. For those firms that did survive, there were significant reductions in employment, and this reduction was greater for females than for all employees. Unsurprisingly, retail and hospitality recorded larger-than-average employment losses as did several manufacturing sectors.

The effects on labor markets were not restricted to employment. Workplace closures were associated with large numbers of lost labor weeks and a high number of workers furloughed. Presumably, furloughs were linked to temporary closures, but these would, at best, affect the sharing of the cost of lost labor weeks, not the total loss. A substantial minority of firms either cut wages or working hours or both.

Firms’ predictions of the number of weeks that they would survive without sales are, on average, longer than the average length of temporary closures. It is clear that many firms could survive further temporary closures, but it is also clear that this is far from the case for all firms. On average, firms have lost about 24% of their sales since December 2019, with the figure being much higher for sectors such as construction, garments, and hospitality. Unsurprisingly, close to 60% of firms report a decrease in liquidity as a result of COVID-19.

Ultimately, a loss of liquidity threatens the survival of firms. There are two main routes by which firms can survive without sufficient sales – government support or private finance. Nearly 77% of our sample had neither received nor expected to receive government support. Some sectors, of course, do not need support, but sectors such as construction and hospitality, which faced large losses in sales, have average or greater-than-average proportions of firms without support.

About one third of firms in the sample had submitted applications for loans, of which more than one half were approved in full in three of the four countries. The exception (El Salvador) had >40% of applications still under consideration. A majority of firms did not apply for loans because they did not need them. Overall, COVID-19 must have increased the risk faced by lenders, but there is evidence that finance has been available to a significant proportion of applicants. It is less clear that the same could be said of government support.

Firms, when asked in the Round 2 survey, most consistently ranked tax reductions or deferrals as a preferred means of government support. In the earlier enterprise surveys, they identified political instability and corruption as the most serious obstacles to business.

Overall, the COVID-19 crisis is still temporary even if deeper and more prolonged than any recent economic downturn. The focus of the paper on labor, and, in particular, employment, means that the hoarding of labor by firms is the key. This, in turn, requires firms to be able to survive in the presence of significantly reduced sales. The demand for labor is a derived demand, and a key strength of firm-level data is that they clarify the link between firm survival and labor demand.

Table 14 presents the results of OLS regressions for the four key dependent variables:

The estimation followed a general-to-specific approach, and it is the results of the specific model that are reported. The results include some variables that were not individually significant but were found to be jointly significant with other variables that were similarly insignificant on an individual basis.

The expectations of firms in the context of the length of their survival time without sales were found to be related to a number of variables. Expected survival time was found to be positively related to the change in sales, as one might expect, but this effect was only significant at the 90% confidence level and of a small magnitude. A much more powerful and statistically significant (at 99% confidence) was the negative effect of changes in debt. It suggests that increases in debt as a consequence of COVID-19 substantially reduce the length of time that firms believe they can survive. Two further variables that have a negative effect, but which are only statistically significant at 90% confidence, were the following: (a) developing online delivery; and (b) a female top manager. It is not clear whether the latter effect reflects either greater pessimism among female top managers or whether their firms face a tougher reality than those of males. Another strong and statistically significant effect on expected survival is the stringency of government workplace closures. Unsurprisingly, the tougher the government closures are, the lower is the expected survival time of firms.

As one might expect, the change in sales was statistically significantly and positively correlated (at 99% confidence) with both expected survival times and changes in employment. The causality, of course, is likely to be from sales changes to labor changes or survival expectations. Foreign ownership was found to have a modest but statistically significant (95%) positive effect on the change in sales. As one might expect, there are negative and statistically significant correlations between sales changes and (a) the number of weeks that the firm was closed and (b) furloughs. Both, almost by definition, should be associated with fewer sales. More quixotic is the relationship between different government containment measures and sales. In general, one would expect that the tougher the government restrictions, the worse are the firm sales. The results suggest this to be the case for stay-at-home restrictions and public transport closures, both of which have negative and statistically significant (at 99%) coefficients. A perverse result is that the coefficient for workplace closures is positive and significant.

As expected, the effect of a change in sales on the change in employment was found to be positive and statistically significant (at 99% confidence), but the magnitude was lower than expected. The coefficient for the share of workers furloughed was positive, statistically significant, and of some magnitude. This suggests that furloughs have been effective in helping to preserve employment.

The regression for the share of females in employment was statistically complex. The results include a number of variables that were individually statistically insignificant but which are jointly significant. The likely explanation is that many of these variables interact – e.g., stay-at-home restrictions, online sales, and delivery – and are multicollinear. This notwithstanding, the main variables affecting the share of females in employment are those one would expect – the change in overall employment and workplace closures. The effect of employment changes on the share of females was negative and statistically significant (at 95% confidence). This suggests that losses in employment have fallen more heavily on males and had the effect of modestly increasing the share of females. It is likely that this has been driven by gender segregation and the fact that a number of manufacturing sectors have been hard hit. Government workplace closures also had a negative and statistically significant effect on the share of females. Again, this is likely to be related to gender segregation, with services being more prone to closures.

The OLS regression results suggest that the change in firm sales is an important determinant of the change in employment. This is not a remarkable proposition. COVID-19 typically resulted in a decrease in sales (fall in product demand), which resulted in a loss of liquidity. This, in turn, resulted in a loss of employment (fall in labor demand). For these reasons, a two-equation SURE model was used to provide some thorough insights.

Table 15 reports the results of the SURE model linking the proportionate change in sales with the proportionate change in full-time permanent employees. As with the OLS regressions, the results reported are those after working from general to specific.

For the change in sales, the following variables exhibited a positive and statistically significant positive effect: developing deliveries; and foreign ownership. The following variables had a statistically significant negative effect: the share of online sales in the total sales, temporary closures, furloughs, stay-at-home restrictions and public transport closures. Most of these are as one might expect, but the negative effect of online sales presumably reflects the difficulties of the retail sector. The equation for the change in sales repeats the perverse result of the OLS equation of a positive effect of the level of stringency of government workplace closures on the change in sales.

For the change in employment, as with the OLS equation, the main statistically significant effects are positive – for the change in sales and for the proportion of workers furloughed. As before, they suggest that improved sales and furloughs save jobs.

Since the effects on gender equality are an important feature of this study, a second SURE model was estimated, linking the change in firm sales to the change in the share of females in the share of full-time permanent employees of the enterprise. As before, the underlying reasoning is the link between product demand and labor demand. The proportionate change in the share of females in employment was used rather than changes in the number of females, in order to identify any gender biases involved. Again, the results after working from general to specific are reported in Table 16.

Since the equation for the change in sales yields conclusions similar to those of both the OLS regression and the preceding SURE model, no further comment is offered. As with the OLS regression, numerous variables in the equation for the female share in firms are collinear and, hence, are individually insignificant but jointly significant. Surprisingly, the change on employment is not statistically significant at 90% or higher confidence level, but it is negative and only just outside (significant at 88% confidence). Both online sales and foreign ownership were found to be both positive and statistically significant at 95% confidence.