3.1 Aggregate picture
This section presents quarterly labor market indices in Belgium compiled from the NSSO firm-level data. Figure 1 shows job turnover rates covering the first quarter of 2004 through the last quarter of 2020. The indices plot year-on-year net employment growth, job creation, and job destruction, and gross job reallocation rates computed using equations (2) to (4), respectively. First, the average job creation rate and job destruction rate fluctuate at around 6.4 and 5.4 percent, respectively. This gives a slight positive average net employment growth of 1 percent per year.
Second, we observe that during the first and second quarters of 2020, job destruction rates increased sharply by 9 and 23 percent, respectively. Job creation rates declined, however to a lesser degree compared to job destruction rates. We also note that the job destruction rate moves counter-cyclically so that we observe higher job destruction in a recession than in a boom as expected. This is also evident with the financial crisis of 2008 where the job destruction rate was 9 percent and the job creation rate was 5 percent (2009q2).
Third, the results indicate that there is a V-shaped recovery[13] of the pandemic crisis compared to the U-shaped recovery of the Great Recession. During the Great Recession, the net employment growth was down by about 3 percent for four consecutive quarters. During the first three months of 2020, the volume of employment (FTE) in Belgium fell by 3.6 percent compared to the first quarter of 2019. This negative employment effect is even stronger in the second quarter of 2020. Between April and June, aggregate employment dropped by 20.8 percent compared to the second quarter of 2019. The impact on growth is less severe during the last two quarters where the net creation rate was down by around 5 percent. Nevertheless, the impact of the Covid-19 outbreak on employment is unprecedented and even more drastic than the financial crisis of 2008.
A sharp drop in FTE is expected since many firms placed their workers on temporary layoffs. The fact that the number of FTE has fallen sharply indicates that there is considerable overcapacity in Belgian companies and that they have consequently experienced a strong drop in GDP growth. Eurostat (2021)[14] reports that the GDP growth in Belgium declined by almost 14 percent in the second quarter of 2020, and by 2 percent in the first quarter of 2020 compared to the same quarter of the previous year.[15]
These findings provide a broad picture of the impact of the pandemic on the labor market for the year 2020. Other papers have shown similar negative short-run impacts mostly using survey data on sales (Bloom et al., 2021) and firm activity (Fairlie, 2020). Bloom et al. (2021) show the negative impact of Covid-19 on sales and employment peaked in the second quarter of 2020 in the US. These findings suggest that the 2020 shock is much more severe than the 2008 shock. Furthermore, the spikes in job destruction rates are comparable to the oil shocks in the UK (Konings, 1995) and the US (Davis and Haltiwanger, 1992). For instance, job destruction rates rose sharply than job creation rates following oil and monetary shocks. In other work, Davis et al. (1996) show that since 1972, oil shocks affected major employment and unemployment fluctuations in the US as the unemployment rose sharply in the aftermath of each major oil shock. We also observe that gross job reallocation (sum of job creation and job destruction rates) moves counter-cyclically which is evidence of creative destruction. The cleansing mechanism would suggest that gross job reallocation rises due to joib destruction that is more procyclical than job creation (Caballero and Hammour, 1994). Davis and Haltiwanger (1992) and Konings (1995) also found evidence that during the recessions gross job reallocation moves counter-cyclically.
The aggregate picture of job creation and destruction masks important heterogeneity in the channels driving aggregate employment growth. In particular, it is the heterogeneity in firm growth that we will focus on next. The evidence from previous crises suggests that not only small firms but also young firms are vulnerable to external shocks and thus are affected the most (Haltiwanger et al., 2013). This size and age characteristic of the firms suggests that financial constraints, as well as uncertainty and the lack of confidence, hit young and small firms particularly hard. In the next sections, we further investigate how firms of different size and age are affected by recessions.
3.2 Job flows by firm size
Figure 2 depicts the evolution of net employment growth by firm size. To construct the size classes, we use the average firm size at t and t-4.[16] We adopt four size classes to indicate micro firms (0-9), small firms (10-49), medium firms (50-249), and large firms (>250). We observe that during the Great Recession employment dynamics of large firms were affected the most. During the peak of the Great Recession (2009q1), net employment growth within large firms went down by 6 percent, whereas micro firms saw a 0.4 percent decline in net growth. However, during the peak of the pandemic crisis (2020q2) micro firms with less than 10 workers were affected the most with a drop in employment of around 33 percent, while large firms with more than 250 employees saw around a 16 percent decline in FTE employment.
Figure 3 depicts the relative job creation rates (Panel A) and job destruction rates (Panel B) for each size category. We find that large firms destroyed more jobs during the Great Recession, whereas during the year 2020 micro firms' employment dynamics were more affected than large firms both on the job creation and on the job destruction margin.
The differential response by firm size during the pandemic appears to be in line with the recent findings from the US (Bartik et al., 2020; Cajner et al., 2020; Bloom et al., 2021). Cajner et al. (2020) report that during the first few months after the pandemic shock hit the US economy, small firms with less than 50 workers were hit the most with a 25 percent decline in employment over the first quarter of 2020. These results are broadly consistent with the previous studies which state that small firms are more responsive to business cycles as they tend to be more adversely affected by credit constraints (Gertler and Gilchrist, 1994; Sharpe, 1994). This contrasts the view of Moscarini and Postel-Vinay (2012) who document that large firms have a disproportionate response to business cycles relative to small firms due to a poaching effect. In particular, during an economic expansion when the unemployment rate is low and the workers’ pool is limited, large firms have a greater ability to poach workers away from smaller firms. During an economic downturn, large firms shed more workers and consequently, are more cyclically sensitive than small firms. Nevertheless, these results are based on the downturns only, a true test would test the cyclicality of large versus small firms over time not merely focusing on the major shocks. Thus, in section 5 we provide a simple regression based approach to look over time the correlation between net aggregate employment growth in small and large firms and business cycle measures, such as unemployment and growth rate in GDP.
3.3 Job flows by firm age
Figure 4 plots the evolution of net employment growth by age category. Following Lawless (2014), we adopt four age categories: 0-5, 6-10, 11-20, and more than 21 years of operation firms.[17] We define young firms as those with less than or equal to 5 years of operation. Old firms are those older than 20 years of operation. Firms that are in between 6 and 20 years are considered mature firms. First, the average net employment growth for young and old firms fluctuates at around 15 and 1 percent, respectively. In Table 1 we show that young firms account for only about 5 percent of employment, yet their annual average net creation rate is around 15 percent. In contrast, older firms account for 67 percent of workers, yet only contribute 1 percent to net growth. This indicates that young firms disproportionately contribute to aggregate employment growth.
Second, both during the Great Recession and the pandemic crisis, the growth rate of young firms was more affected than that of older firms. Figure 4 shows that the differential net employment growth between young and old firms just before the pandemic was about 15 percentage points and dropped by more than 40 percentage points during the second quarter of 2020, emphasizing the stronger effect of the pandemic on the employment growth for young firms.
Figure 5 depicts the relative job creation rates (Panel A) and job destruction rates (Panel B) for the young and old firm category. We find that the employment dynamics, in terms of creation and destruction rates, of young firms were more affected during both crisis periods. Young firms created fewer jobs and destroyed more jobs during the Great Recession and during the year 2020 compared to old firms.
The importance of young firms on firm growth has been well documented recently (Haltiwanger et al., 2013; Criscuolo et al., 2014; Lawless, 2014). Haltiwanger et al. (2013) show that young firms, in particular, are very heterogeneous. They are typically small due to their lack of reputation in product and credit markets, constraining them to scale up. Hence they are often characterized by an ‘up or out’ dynamics with high job creation and job destruction rates. Fort et al. (2013) find that small and young firms are more sensitive to cyclical conditions than mature firms and provide evidence that shocks in house prices and restricted access to credit due to home equity financing by small and young firms might explain these results. We show that small firms (10-49) account for about 22 percent of all jobs (Table 1), yet they only contribute 7 percent to job creation and 5 percent to job destruction, on average (Figure 3), so they contribute disproportionately less to overall job creation and job destruction (likewise for micro firms). In contrast, young firms account for only about 5 percent of aggregate employment, yet they contribute more than proportionate to job creation and job destruction, 24 and 8 percent respectively, on average. Furthermore, they are also more responsive to crisis years, which is especially clear from the pandemic shock. These findings indicate that young firms are more important than small firms for net employment growth and for overall job creation and destruction.
3.4 Job flows by wage quintiles
As an alternative outcome variable of interest, we further explore information on wages to shed some light on heterogeneity in the type of jobs, i.e. high versus low-paying jobs that are generated or destroyed over the business cycle. This also allows to shed some light on the creative destruction as we can use wages as proxy for productivity of the firm.
We classify firms based on the different quintiles of the wage distribution. Specifically, we use the average wages per worker (in FTE) per firm to define five quintiles for our analysis.[18] As seen from figure 6, during the Covid crisis, drop in employment is disproportionately concentrated among firms in the bottom quintile of the wage distribution. In particular, these firms saw more than a 35 percent decline in employment compared to a 13 percent drop in employment in firms at the top quintile of the wage distribution during the second quarter of 2020. Employment for this group of firms has partially recovered through the third quarter, but still saw a drop in the last quarter of 2020. Hence, we observe that the loss in employment during the Covid crisis is disproportionately concentrated among low-wage firms hinting to the cleansing effect of the pandemic. The cleansing effect, thus, may trigger productivity improvements where resources are reallocated from low-wage, typically unproductive firms, to high wage, productive firms. This is because recessions prompt productivity enhancing job reallocation by driving out unproductive investments and freeing up resources for more productive uses (Schumpeter, 1939). Hence, productivity growth at the country level may arise when resources are allocated towards high productivity firms. These results differ when we look at the response of firms to the 2008 crisis. As can be seen from figure 6, firms in the third quintile of the wage distribution were affected the most with an 11 percent drop in employment in the second quarter of 2009. Figure 7 also shows that the rise in job destruction rates accounts for the most part of the drop in net growth.
The differential response by wage quintiles seems to reflect the industry-specific effects of the shocks. In particular, the pandemic crisis affected the hospitality industry the most, which is most likely to cover low-level wage workers and firms. These are typically small firms, which we also confirm in section 4.2 to be affected the most. In the next section, we further distinguish the sectoral impact of the crises.
3.5 Job flows by sector
Figure 8 shows the evolution of net employment growth by broad sectors, manufacturing, services, and trade. The largest decline in employment during the 2008 crisis was in manufacturing whereas, during the pandemic crisis, the services sector that requires substantive interpersonal interactions was hit the most. For instance, through March, FTE employment in the services sector fell by 24 percent, whereas employment in the manufacturing sector dropped by 16 percent.
Figure 9 shows that both a steep change in employment dynamics in terms of job creation and destruction rates for the manufacturing sector during the Great Recession. During 2020, the services sector saw a steep change in job creation and destruction rates. The volume of workers declined by more than 24 percent in the services sector during the peak (2020q2) compared to the 16 percent decline in the manufacturing sector.
These findings suggest that the difference in responses to crises by small and large firms stems from the industry-specific effects of the shocks. In particular, a manufacturing sector containing, on average, larger firms, was hit harder by the Great Recession whereas the services sector, comprised primarily of small firms, has been disproportionately affected by the pandemic crisis. These findings reflect the underlying cause and the nature of the recessions. The Great Recession was a result of disruptions in credit markets that started in the housing sector and resulted in lower demand for consumer and capital goods, produced in manufacturing. On the other hand, the Covid-19 disease is entirely an exogenous shock that require containment measures and precautionary behavior thus affecting non-tradable service sectors that requires substantive interpersonal interactions, such as the leisure and hospitality industries.
[13] Please note that the current dataset is up to and includes the last quarter of 2020, whereas the pandemic crisis is still ongoing. Hence, as more data become available the shape of the recovery might change.
[14] https://ec.europa.eu/eurostat/documents/portlet_file_entry/2995521/2-02022021-AP-EN.pdf/0e84de9c-0462-6868-df3e-dbacaad9f49f.
[15] On the other hand, according to the National Bank of Belgium, over the second quarter of 2020, the average number of workers decreased by only 0.8 percent. However, this measure of employment is related to temporary unemployment and includes those part-time workers on temporary layoff. The less outspoken drop in the number of workers reflects the fact that there are furlough schemes set up by the government to cushion the crisis. This could have two implications for the study: one regards the actual comparability of the data across periods, which we will partially address in Table 4; and the other regards the possibility that the features of the 2020 furlough scheme may favor certain firms over others. The fact that the scheme was simplified in March 2020 may have facilitated its adoption by those firms that, otherwise, would have had limited access to it. The latter would require data on other firm performance indicators as well as indicators on whether firms actually received any support measures. Unfortunately, we cannot answer this question due to the data limitations, but leave it for future research. Nevertheless, the challenge will be once the system runs out whether companies will take all these people back on a full-time basis. After the global and financial crisis of 2008, furlough schemes were also in place. However, 20 percent of the workers on furlough schemes were not back to work two years later (Struyven et al., 2016).
[16] To avoid the statistical pitfall, i.e. regression to the mean bias, we opt to use average size as described in Davis et al. (1996) and Haltiwanger et al. (2013). Alternatively, we use the initial firm size that is fixed once and for all with longitudinal data. The results are robust and can be found in figures A4 and A5 and table A1.
[17] Results yield very similar conclusions when we exclude entry (age 0), i.e. focus on surviving and exiting firms only (see Appendix, Figures A2 and A3).
[18] In the Appendix, we also show the results where we fix the wage per worker at the beginning of our sample period. See Figures A6 and A7.