Labor market effects of COVID‐19 in Sweden and its neighbors: Evidence from administrative data

Abstract

This paper studies the labor market effects of non‐pharmaceutical interventions (NPIs) to combat the COVID‐19 pandemic. We focus on the Nordic countries which showed one of the highest variations in NPIs despite having similar community spread of COVID‐19 at the onset of the pandemic: While Denmark, Finland and Norway imposed strict measures (‘lockdowns’), Sweden decided for much lighter restrictions. Empirically, we use novel administrative data on weekly new unemployment and furlough spells from all 56 regions of the Nordic countries to compare the labor market outcomes of Sweden with the ones of its neighbors. Our evidence suggests that the labor markets of all countries were severely hit by the pandemic, although Sweden performed slightly better than its neighbors. Specifically, we find the worsening of the Swedish labor market to occur around 2 to 3 weeks later than in the other Nordic countries, and that its cumulative sum of new unemployment and furlough spells remained significantly lower (about 20–25%) during the time period of our study (up to week 21 of 2020).

1. INTRODUCTION

The vast majority of countries have implemented strong non‐pharmaceutical interventions (NPIs) to slow the spread of COVID‐19. A number of studies provide evidence that NPIs are effective at reducing adverse health outcomes (see, e.g., Cho, 2020; Flaxman et al., 2020; Glogowsky et al., 2020; Juranek & Zoutman, 2020; Born et al., 2021). However, there are important concerns about the potential damage NPIs cause to the economy and labor markets (Cajner et al., 2020; Kong & Prinz, 2020; Sheridan et al., 2020). Specifically, the mandated restrictions many countries have enforced (‘lockdowns’) are assumed to inflict stark economic pain (Baldwin & Weder di Mauro, 2020). Thus, the decision problem governments are facing may be portrayed as a trade‐off between public health and the health of the economy (Lin & Meissner, 2020). However, the virus itself also deters economic activity, since households reduce consumption and working to avoid own contagion and personal health risks. Our study aims to shed light on the question how much of the tremendous decline in economic activity resulted from government‐imposed restrictions versus people voluntarily choosing to stay home to avoid infection. A better understanding of the existence and size of this trade‐off is of high policy relevance, given the ongoing debates over the right stringency of measures to combat COVID‐19.

We use novel high‐frequency (weekly) regional unemployment and furlough spells from all four Nordic countries to evaluate the economic effects of NPIs during the first wave of COVID‐19. We employ this data to study the differential labor market effects of one of the most prominent policy variations observed during the COVID‐19 pandemic: Sweden departed substantially from its neighbors in the response to the spread of the disease, kept kindergarten and elementary schools but also businesses and shops open. Our estimation strategy is similar to Sheridan et al. (2020) and draws on this large policy variation in the Nordics, comparing countries which were equally exposed to the COVID‐19 pandemic but responded in very different ways.

The Nordic countries represent an ideal laboratory to study the differential impact of NPIs on labor market outcomes. First, the Nordics are very similar with regard to the general economic environment (e.g., GDP per capita, trade openness), their health care sectors and the general institutional background. Furthermore, employment levels and trends were very similar across the four countries prior to the Covid‐19 crisis (OECD, 2018). Second, due to geographical proximity and their economic interrelations these countries experienced highly similar trajectories of the COVID‐19 pandemic: The 100th case of a confirmed infection occurred in Norway on the 4th, in Sweden on the 6th, in Denmark on the 9th and in Finland on the 12th of March. The measures to slow the spread of COVID‐19, however, differed substantially between the four countries. Starting in week 11 of 2020, Denmark, Norway and Finland responded with strong NPIs to limit social interaction, while Sweden imposed much lighter restrictions. Table 1 depicts the dates of the introduction of various measures along with an overall government stringency index, developed by Hale et al. (2020). The index shows that Norway and Denmark imposed the toughest restrictions followed by Finland, and the much weaker response of the Swedish government to the pandemic.

TABLE 1.

Timing of closures and containment in Nordic countries

Measure	Denmark	FinlandNorway		Sweden
		Day in March 2020
School Closing	13	16	12	–
Workplace closing	13	16	12	–
Cancel public events	18	12	12	12
Close public transport	–	–	–	–
Restrictions on internal movements	–	28	16	–
International travel controls	11	19	15	19
Stringency index (maximum in week 11–13)	72.2	67.3	75.9	32.4

Notes: Dates in italics indicate that a measure was general in scope. The stringency index is a compound of eight closing measures and is ranged between 0 and 100, where a higher index represents stronger overall restrictions; see Hale et al. (2020).

The measures had direct implications for many types of economic activity: In Norway, Finland and Denmark, the hospitality industry (such as bars, restaurants or hotels) was largely shut down, personal services (e.g., hair dressers, masseurs or dentists) were closed, shopping centers had to stop operating, and public transport was limited. In contrast, Sweden decided for much less strict measures, keeping restaurants and bars open (under certain proximity restrictions), and allowing private businesses and shops to continue to operate. In fact, Google's COVID‐19 Community Mobility Reports show very different mobility patterns for Sweden than for the other three countries during the period of the lockdowns (see Figure 1). This suggests that there are behavioral responses caused by the lockdown measures in addition to the threat of the virus (see also Chan et al., 2020). In other words, the NPIs seem to have constrained the choices of the population.

FIGURE 1.

Economic activity in Nordic countries. Notes: The figures show how visits and length of stay at different places changed compared to the median weekly value, using the 5 week period from January 3 to February 6, 2020 as comparison. The blue shaded vertical line indicates the date of the lockdowns from Table 1, which is around March 13 (week 11). The vertical line indicates Easter holidays (week 16). The black solid graph represents how the mobility patterns changed in Sweden, and the dashed ones show the corresponding changes in its neighboring countries Denmark, Finland and Norway. Source: Google LLC “Google COVID‐19 Community Mobility Reports.” https://www.google.com/COVID19/mobility/ [July 15, 2020] [Colour figure can be viewed at wileyonlinelibrary.com]

At the same time, all Nordic countries introduced government programs to soften the impact of the pandemic on the economy and labor markets. Denmark and Sweden almost simultaneously introduced a novel short‐time work furlough program on the 9th and 16th of March, respectively. 1 Both programs guarantee between 75% and 80% of the salary of workers which are currently not needed but kept on payroll of their companies. Furthermore, the salary cap for furloughed workers are almost identical in both countries (EUR 4,150 vs. 4,000 per month). In a similar vein, Finland made its existing furlough program more generous due to the crisis, with replacement rates varying between 80 and 100% for workers reducing their working hours during the pandemic. Norway also made its existing furlough program more accessible and more generous when the COVID‐19 crisis struck. While program generosity may vary in details, the incentives of affected businesses to participate in the respective furlough program were very large (OECD, 2020).

For our study, we collected novel administrative data on weekly new unemployment and furlough spells from the Nordic countries at the regional level. It is a key strength of our work to not only cover the effect of the crisis on unemployment, but also on the number of people filing for one of the national furlough programs. In some aspects, furlough spells act as a form of hidden unemployment. From the employer's perspective, placing workers on furlough rather than laying them off may be beneficial if the furlough arrangement is generous enough, and the downturn is expected to be temporary. Therefore, studying unemployment without considering data from furloughs gives an incomplete picture of the development in the labor market.

Empirically, we compare labor market outcomes between Sweden and its Nordic neighbors in an event‐study framework. Our comparison focuses on the regional number of new weekly unemployment and furlough spells between week 1 and week 21 of 2020 with the corresponding figures in 2019. Week 11 serves as the event date, when the lockdowns of Denmark, Finland and Norway were implemented. To adjust for the general business cycle and seasonal effects we include a set of region‐year and country‐week fixed effects.

Overall, our results show that the labor markets of all Nordic countries were hit hard by the pandemic. Starting in week 11 of 2020, we observe a sharp increase in newly unemployment and furlough spells especially for Norway and Denmark, but also for Finland. Sweden shows a similar but much less pronounced peak in new unemployment and furlough spells. When using the cumulative (total) number of new weekly unemployment and furlough spells, we again find the labor markets of Denmark and Norway to have suffered the most, followed by Finland and Sweden. To quantify the differences in unemployment and furlough spells, we employ difference‐in‐differences (DID) regressions. We find the DID coefficient of the cumulative sum of unemployment and furlough spells to be around 1,360 spells higher per 100,000 of population for Denmark in week 21 compared to Sweden. This suggests that Denmark would have accumulated around 22% less unemployment and furlough spells if lighter restrictions similar to Sweden would have been implemented. Employing weekly regional stock data of furloughs (which is only available for Denmark and Sweden), we do not find significantly different patterns of outflow from furloughs between the two countries. In sum, the results from the unemployment and furlough data mirror the pattern from the Google mobility data shown in Figure 1: The lockdowns of Norway and Denmark seem to have had the largest impact, followed by Finland. In contrast, the Swedish labor market appeared to have suffered the least from the pandemic.

Evidence on the economic consequences of social distancing laws to combat COVID‐19 is still scarce but increasing (see, e.g., Sheridan et al., 2020: for a comparison between Sweden and Denmark in consumer spending). The papers most closely related to our work on labor market effects of NPIs are all using U.S. data, comparing the differential strictness of stay‐at‐home policies across U.S. states with subsequent unemployment insurance claims. The evidence is mixed. Baek et al. (2020) as well as Gupta et al. (2020) find sizable increases in unemployment insurance (UI) claims due to stay‐at‐home policies. In contrast, other studies find rather similar increases in UI claims across U.S. states, regardless of the timing of local policy responses (see, e.g., Kong & Prinz, 2020; Lin & Meissner, 2020; Forsythe et al., 2020). We believe the Nordic countries offer an important policy experiment to shed new light on this question due to their i) very similar exposure (regarding time and space) to the initial spread of COVID‐19, and (ii) very large as well as long‐lasting variation in NPI strictness. 2 To the best of our knowledge, our study is the first one to explore the labor market effects of Sweden's COVID‐19 strategy compared to its Nordic neighbors. It is important to note that our data and analysis ends in week 21, 2020. Nevertheless, we think the policy variation around the first wave of the pandemic is highly suited to study the labor market effects of the COVID‐19 crisis and its related policy measures. Specifically, with the arrival of the second wave in Europe in autumn 2020, the policy responses of Sweden and its neighbors were much more similar, making the identification of effects challenging. 3

The paper proceeds as follows. The next section introduces the institutional background and in particular the unemployment and furlough programs implemented in the Nordic countries. Section 3 presents the data and provides some descriptive statistics. Section 4 elaborates the empirical specification to identify the impact of NPIs on labor markets and presents the empirical results. Section 5 concludes.

2. INSTITUTIONAL BACKGROUND

Many countries around the world have created short‐term worker programs to avoid large mass layoffs of workers. In the following, we briefly describe the different programs of the Nordic countries. 4

2.1. Denmark

Denmark introduced its new short‐time work compensation program on March 9th 2020. This new program allows partaking companies to receive a government refund of 75% of the salaries paid to their retained workers. The requirement for a company to be eligible is that it otherwise would have laid off a minimum of 30% of its workforce (Bennedsen et al., 2020). Furloughed workers keep their jobs and salaries but are not allowed to work, meaning that their working time is reduced by 100%. There is a salary cap on the maximum level of support at 30,000 DKK (around 4,000 EUR) per month for full‐time employees (Rothwell & Drie, 2020). 5

2.2. Finland

In Finland, there exists no short‐time work compensation program as such. However, companies can temporarily layoff employees due to financial or production‐related reasons (so called furloughs). This furlough system already existed before but was made more generous and accessible due to COVID‐19. A furloughed worker continues to have a valid employment contract with the employer, but the employer stops wage payments temporarily due to the lack of work. Furloughed workers are entitled to the same UI benefits as unemployed workers. All workers, including furloughed workers who work reduced hours (i.e., part‐time furloughed), may be entitled to partial UI benefits on top of wage income. Especially the partial UI benefit scheme is generous in Finland, with replacement rates varying between 80 to 100% (Kyyrä et al., 2017). There is no cap to the (partial) UI benefit in Finland, but the replacement rate declines with the previous (full‐time) wage.

2.3. Norway

Similar to Finland, Norway already had a short‐time work and unemployment program in place prior to the pandemic. Originally, a furloughed employee reduced working hours by at least 50%, with the state paying 62.4% of the lost income, up to approximately 31,000 NOK (around EUR 2,900) per month for a full‐time unemployed. The government strengthened the program with effect on March 20 by granting 100% pay, capped at 31,000 NOK per month, for the first 20 days. From day 21 on, the part of the income below 25,000 (around 2,300) is replaced at 80%, whereas the coverage remains unchanged for the other parts of the income (Alstadsæter et al., 2020a). Furthermore, the minimum required reduction in working hours decreased to 40%.

2.4. Sweden

Sweden, similar to Denmark, created a novel short‐time work compensation program coming into effect on March 16th 2020 (Hensvik & Skans, 2020). The new program can be used when companies are faced with temporary financial or production challenges as a consequence of the COVID‐19 pandemic. The most important distinction between Sweden's program, and that of its Nordic neighbors is that a company's employees can reduce their working hours only up to a maximum of 60% (up to 80% after 1st of May) while the government provides financial support in the form of a short‐time work allowance. In our analysis we deal with this difference, by comparing full‐time equivalent (FTE) furlough spells (see section 3.2 for more detail). The financial support reduces an employer's costs for personnel by around 50% (70% after 1st of May), while workers will retain more than 80% of their original pay (KPMG, 2020). The salary cap for financial support is 44,000 SEK (around 4,150 EUR) per month. 6

As we can see above, the degree of generosity of the programs differs somewhat between the countries, especially regarding the maximum possible reduction in working hours. In Section 3.2, we will account for these differing degrees of working time reduction. Overall, however, the furlough programs were designed in similar fashion, especially regarding replacement rates and salary caps. By and large, the incentives of affected businesses to participate in the respective furlough program were substantial across all four countries (OECD, 2020).

3. DATA

3.1. Data Sources

During the pandemic, the administrations of the Nordic countries started to produce weekly reports on the new number of individuals being laid‐off or put on furlough. Most of the reports they issued during these weeks focused on inflow into unemployment and furlough. Thus, we have access to high‐frequency weekly inflow data on the new number of unemployment as well as furlough spells for all regions of Denmark, Norway, Finland, and Sweden for the years 2019 and 2020. 7 In addition, for Sweden and Denmark we also have data on the stock number of people currently on furlough, which allows us to also examine outflows from the respective furlough program.

For Denmark, we received data on the weekly number of new unemployed through Statistics Denmark. We received furlough data from Erhvervsstyrelsen, the Danish Business Authority which manages the program. For Sweden, we received data on the weekly number of new unemployed through the national employment agency. Furlough data was collected through Tillvaxtverket, the government agency managing the furloughs. For both Denmark and Sweden, the furlough programs were newly introduced due to the Corona crisis, which means that no prior data exists (in Sweden, the first data on furloughs is from week 12, for Denmark from week 11). In our data we replace the missing observations for Sweden and Denmark prior to week 12 and in 2019 with zeroes, consistent with the fact that the program did not exist. For Finland, we downloaded the data from the Helsinki Graduate School of Economics webpage. Helsinki GSE created a special webpage collecting and analyzing data around the COVID‐19 pandemic. 8 The Norwegian data we received from NAV, the Norwegian Labour and Welfare Administration. The furlough programs of both Finland and Norway existed prior to the pandemic, which gives us data on the weekly number of new furlough spells also for 2019. 9

3.2. Calculating Full‐time Equivalents for Furloughs

As it has been described above, the institutional arrangements regarding part‐time/partial furloughs differ between the four countries. For instance, in Denmark every person being furloughed is on full‐time furlough, meaning that working time is reduced by 100%. In contrast, a furloughed person in Sweden continues to work partially, since working hours can only be reduced by a maximum of 60% (up to 80% after 1st of May). In Finland and Norway, both part‐time (i.e., a partial reduction in working hours) and full‐time furlough (100% reduction) is possible. Since the working time reduction of a furlough spell indicates how severely a labor market has been hit by the crisis, we want to take this into consideration. Specifically, to account for the different intensities of the furlough spells and to make them more comparable, we will express the number of furloughs as full‐time equivalents (FTE). To do so, we first need information on the number of part‐time as well as of full‐time furlough spells. Second, we have to find a way to account for the average degree of the hours reduction the part‐time furloughed are taking (which we do not have in the data).

Receiving the number of partial furlough spells is relatively straightforward. For Denmark, the share of part‐time furloughs is zero, since everyone on the furlough program needs to reduce working time by a 100%. In Sweden, only part‐time furloughs are possible, which means that everyone in our furlough data is part‐time furloughed. For Norway, we have weekly information on the number of part‐time as well as of full‐time furlough spells, but only on the national level. We use this share of part‐time furlough spells on the national level as a proxy to calculate the number of part‐time furloughs on the regional level. For Finland, we only received data on the number of full‐time furlough spells. However, a government report on the Finnish furlough program from May 2020 finds that only around 15% of all furloughs are actually part‐time (Elinkeinoministeriö, 2020). Thus, for Finland we will use the 15% stated in the report to infer the part‐time share for all Finnish regions.

In a second step, we need to take into account the degree of the hours reduction the part‐time furloughed are taking in order to calculate the corresponding FTE. This data does not exist for any of the countries, neither on the individual nor aggregate level. Therefore, we decided to use the maximum possible reduction of working time possible in Sweden (60% before 1st of May, 80% thereafter), and use this degree of hours reduction also for the part‐time furloughed in the other countries to calculate the FTE. The vast majority of furlough spells of the other three countries are actually full‐time, namely 72% for Norway, 85% in Finland, and 100% in Denmark. Thus, the assumption about the working time reduction of the part‐time furloughed do not matter greatly for these three countries, since most furlough spells are full‐time. In the Appendix A.3 we present robustness checks where we change the assumed working time reduction for the partially furloughed. Overall, we receive qualitatively similar results. 10

3.3. Descriptive Statistics

The main variables of interest in our study are the weekly new unemployment and furlough spells, both measured on the regional level. All our dependent variables are measured in FTEs as explained above, and we normalize them by the population of the respective region and year. Table 2 shows the average number of weekly new unemployment, furlough as well as the cumulative sum of weekly new unemployment and furlough spells for the weeks 11 to 21 and the years 2019 and 2020, respectively.

TABLE 2.

Descriptive statistics

Variable	Denmark	Finland	Norway	Sweden
Number of observations (regions)	210(5)	798(19)	462(11)	882(21)
Population (1,000)	1,162.88	291.00	486.17	493.88
Mean of regions
Min.	589.76	29.88	241.24	59.64
Max.	1,846.02	1,708.43	1,241.12	2,409.46
New weekly unemployment spells (mean of regions) a
2019	116.41	162.36	39.80	59.50
2020	186.92	167.31	95.94	115.58
New weekly furlough spells (mean of regions) a
2019 − 19.56			7.73	−
2020	341.62	359.32	530.52	232.08
Cumulative unemployment and furlough spells a
2020	6,272.38	6,314.31	7,136.35	4,604.45

Notes: ^a Only weeks 11 to 21, all numbers per 100,000 population.

As we can see in the table, from 2019 to 2020 the average weekly number of new unemployment spells increased by about 3% in Finland, by more than 50% in Denmark and more than doubled in Norway and Sweden. More dramatic is the growth in furlough spells, shown in the bottom lines of Table 2. Two things are worth noting: First, we see how important it is to also obtain data on furlough spells when studying labor markets during the COVID‐19 crisis: The average number of new weekly furlough spells are around 2 to 6 times higher than the average number of new weekly unemployment spells. Second, it becomes already evident from this table that the labor markets of all Nordic countries were severely hit by the COVID‐19 pandemic.

4. EMPIRICAL ANALYSIS

4.1. Specification and Identification of Labor Market Effects

Our data is structured as a panel with a country‐region (cr) cross section and a year‐week (jw) time dimension. Hence, the observational unit is at the cr,jw‐level. Our main outcome variables (y) are (i) weekly new unemployment spells, (ii) the weekly new unemployment plus furlough spells, and (iii) the cumulative sum of these spells over time. Our regression model is given by

$$y_{cr,\mathit{jw}}={\eta }_{r,j}+{\alpha }_{c,w}+{\beta }_{c\mathit{,}w}{D}_{j\mathit{=}\mathit{2020}}+{\epsilon }_{cr,\mathit{jw}},$$ (1)

where y _cr,jw denotes the respective outcome for region r of country c in week w of year j. η _r,j are region‐year‐fixed effects capturing the influence of variables that do not vary over the weeks and countries (e.g., a country's industry composition). α _c,w denote country‐week fixed effects controlling for seasonal fluctuations in the respective outcome, and D _j = 2020 is a dummy which equals 1 if the year is 2020, and zero else. The main coefficient of interest is β _c,w which measures deviations in the respective outcome in week w in 2020 compared to the same week w in year 2019. 11 Week 10 serves as the baseline, i.e., β _c,10 is normalized to 0. Standard errors are clustered on the country‐region level.

4.2. Results

Figure 2 present the results from estimating Equation (1). Panel a uses the weekly new unemployment spells as outcome variable, whereas Panel b is based on the weekly new unemployment plus furlough spells. Note that the figures use different scales, since the number of furlough spells is much larger than the number of unemployment spells in all four countries. A couple of things are notable when looking at the two figures. First, the coefficients for the periods prior to the lockdown in week 11 are quantitatively small, move basically in parallel, and do not exhibit a trend. This confirms that during the first weeks of 2020 the labor markets of the four countries were on similar trajectories once accounting for region‐year and country‐week fixed effects. This parallel trend changes abruptly in the week of the lockdown (week 11), when the number of new unemployment spells increases tremendously in Denmark, Finland and Norway. Sweden lags behind this development of its neighbors by a few weeks, with the peak number of new unemployment spells being in week 14. Overall, panel a of Figure 2 shows that the pandemic dwarfs other regional and seasonal specific labor market fluctuations.

FIGURE 2.

Seasonally and regionally adjusted unemployment/furloughs per 100,000. Notes: The figure shows the event‐study coefficients estimated from equation (1), including 95%‐confidence intervals (standard errors clustered on the country‐region level). The blue shaded vertical line indicates the week of the lockdowns in Denmark, Finland and Norway (week 11). Panel a employs new weekly unemployment spells, panel b new weekly unemployment plus furlough spells, panel c cumulative unemployment spells, and panel d cumulative unemployment plus furlough spells as the respective outcome (all per 100,000 population) [Colour figure can be viewed at wileyonlinelibrary.com]

When studying weekly new unemployment plus furlough spells together (panel b of Figure 2), a similar but more dramatic picture emerges. Again, and in line with Bennedsen et al. (2020) as well as Alstadsæter et al. (2020b), we find the increase to be sudden and sharp for Denmark and especially for Norway. In Sweden and Finland, the labor market worsens more gradually, with the peak number of weekly new unemployment plus furlough spells being in week 14. In sum, we find that the two strict lockdowns of Denmark and Norway had an immediate and strong effect on their national labor markets. The somewhat less strict and later lockdown of Finland (see Table 1) delayed the worsening of the labor market by around 2 weeks. Interestingly, also the Swedish labor market seems to have been hit hard by the escalating pandemic, but with a slightly better performance compared to its neighbors.

The differential timing in the surge of the weekly new numbers may mask some differences in the total sum of unemployment and furlough spells across the four countries. Therefore, we also employ the cumulative sum of new unemployment and furlough spells as dependent variables. Panel c displays the regression coefficients when using cumulative new unemployment, and panel d when employing cumulative new unemployment plus furloughs as the respective outcome. When looking at the combined measure (panel d), we again find that the labor markets of Denmark and Norway seem to have suffered the most. This mirrors what we have already observed in the mobility data shown in Figure 1: The lockdowns of Norway and Denmark seem to have had the largest impact, followed by Finland and Sweden.

In order to estimate the differences between Sweden and its neighbors more directly, we employ an event‐study difference‐in‐differences (DID) analysis in which Sweden serves as the control group and where treatment takes place in week 11 12:

$$y_{cr,\mathit{jw}}={\eta }_{r,j}+{\alpha }_{c,w}+{\gamma }_{jw}+{\beta }_{c,w}{D}_{w\ge 11}+{\epsilon }_{cr,\mathit{jw}},$$ (2)

where γ _jw denote week‐year fixed effects. In this model, β _c,w denotes the DID between country c and Sweden (the omitted category) between week w and week 10.

Results are reported in Figure 3 where we focus on the cumulative sum of the weekly unemployment and furlough spells as outcome variable. We see that after week 10, Denmark as well as Norway see a strong spike in the cumulative sum of unemployment and furlough spells relative to Sweden. After week 13, the coefficients for both Denmark and Norway decline gradually, but remain significantly larger compared to Sweden up to week 21. Finland's trajectory is more similar to the Swedish development, but its cumulative sum of unemployment and furlough spells also remain significant higher by week 21, 2020.

FIGURE 3.

Seasonally and regionally adjusted cumulative unemployment & furloughs per 100,000. Notes: The figure shows the leads and lags estimated from equation (2), including 95%‐confidence intervals (standard errors clustered on the country‐region level). The outcome variable is the cumulative sum of unemployment plus furlough spells per 100,000 population. The blue shaded vertical line indicates the week of the lockdowns in Denmark, Finland and Norway (week 11) [Colour figure can be viewed at wileyonlinelibrary.com]

In order to quantify the differences, we use the estimated coefficient of week 21 from our DID estimation (Equation (2)), and compare it to the overall level of the same outcome variable in the weeks 11 to 21 of 2020. Specifically, we use the DID coefficient of the cumulative sum of the weekly new unemployment plus furlough spells for Denmark in week 21, which is about 1,360 per 100,000 population (depicted in Figure 3, as well as in Table A.1 in the Appendix). We relate this DID estimate to the cumulative sum of the weekly new unemployment plus furlough spells for Denmark by week 21, which is around 6,300 (see Table 2). Thus, following the Swedish model of no strict lockdown, Denmark would have accumulated 22% less unemployment plus furlough spells up to calendar week 21. The estimate for Denmark appears to be in the same ballpark but somewhat larger than what Sheridan et al. (2020) find using bank transaction data from Swedish and Danish bank clients. Specifically, they find a 25% drop in spending for Sweden versus a 29% drop for Denmark, and interpret the difference as the causal effect of the lockdown. This difference points to an impact of the lockdown of about 14%, based on the drop of activity in Denmark (≈4/29). 13

5. DISCUSSION

It became (and still is) a vigorously debated question during the pandemic how much of the economic collapse resulted from people voluntarily choosing to stay home to avoid infection versus government‐imposed restrictions on activity. Exploiting the large policy variation between the Nordic countries we find that around 20 to 25% of the slump on the labor market are due to lockdown measures. This suggests that strict measures do add additional economic burden on top of the threat of the virus. Thus, our study provide guidance on the question how to trade‐off between public health and the health of the economy, which makes it highly policy‐relevant. A better understanding of the existence and size of this trade‐off allows for better fine‐tuning of the measures as well to use it in overall social welfare calculations.

In this respect, it seems important to note that the health vs. economy trade‐off seems not have been the key driver behind the differential lockdown decision of Sweden. Rather, it appears that concerns regarding the negative side effects on health and societal well‐being played a critical role why Sweden decided for lighter restrictions. 14 One of the most prominent examples of this differential assessment was the Swedish decision regarding school closures. During the first wave of the pandemic, all neighboring countries of Sweden closed kindergardens and elementary schools, whereas Sweden kept them open. 15 Interestingly, the public health agencies in Denmark and Norway also recommended keeping kindergardens and elementary schools open but they were overruled by their governments.

This directs us to the question why Sweden responded so differently compared to its neighbors. A crucial reason many observers pointed to is the Swedish constitution (see Jonung, 2020). In Sweden, some public agencies such as the public health agency (Folkhälsomyndigheten) are independent from the government. The Swedish constitution forbids the ministerstyre a direct interference with the decisions made by the public agencies. There is no similar constraint to the power of ministerstyre in Norway or Denmark. This independence gave the Swedish public health agency (Folkhälsomyndigheten) and its state epidemiologist Anders Tegnell a key role in handling the pandemic. A key policy goal of his agency was to implement measures that can be sustained for a long time. In order to ensure compliance with social distancing measures over time, the agency put more weight on recommendations instead of on mandated strict lockdowns. 16

After the first wave of the pandemic in March/April 2020, one might observe some policy learning in the sense that the stringency of measures among the Nordic countries converged to some degree. For instance, during the second wave of COVID‐19 in early winter 2020/21, Sweden's stringency index sometimes even exceeded the one of Norway or Denmark. It is interesting to note that the Swedish stringency index stayed rather constant over the course of the pandemic, whereas the other countries showed much more fluctuations, experiencing a sequence of multiple lockdowns and re‐openings. 17 Overall, however, it appears that Sweden and its neighbors continued to follow their differential approach: Norway, Denmark and Finland responded with very harsh measures even to smaller outbreaks of COVID‐19 cases, following a strict suppression strategy. In contrast, Sweden did refrain from mandatory lockdowns also during the second and third wave, tolerating a much higher case load compared to its neighbors. So while there happened some convergence in stringency, different approaches across the Nordics can still be observed.

6. CONCLUSION

This paper studies the labor market effects of non‐pharmaceutical interventions (NPIs) to combat the COVID‐19 pandemic. We focus on the Nordic countries which showed one of the highest variations in NPIs despite having similar exposure to the spread of COVID‐19 at the onset of the pandemic. Empirically, we use novel data on weekly new unemployment and furlough spells from all 56 regions of the Nordic countries to compare the labor market outcomes of Sweden with the ones of its neighbors.

We find that the labor markets of all four countries were severely hit by the pandemic, with Sweden performing slightly better than its neighbors. Specifically, we find the worsening of the Swedish labor market to occur with a time lag of 2 to 3 weeks compared with its neighbors, and that its cumulative sum of new unemployment and furlough spells remains significantly lower up to week 21 of 2020.

Juranek and Zoutman (2020) show that the lockdown in Denmark and Norway was successful in terms of reducing the pressure on the health care system and mortality. However, our study indicates that the lockdown comes at a cost in terms of labor market performance. Whether the benefits outweigh the costs depend in part on ethical judgement which is beyond the scope of this paper. Nevertheless, further economic evaluations to assess the different trade‐offs a society is facing when combating a pandemic appears a fruitful avenue for future research.

Supporting information

Figure A.1: Seasonally and regionally adjusted stock number of furloughs per 100,000

Notes: The figure shows the event‐study coefficients estimated from Equation (1), using the cumulative stock of furloughs rather than inflows (all per 100,000 population). The whiskers indicate the 95%‐confidence intervals (standard errors clustered on the country‐region level). The blue shaded vertical line indicates the week of the lockdowns in Denmark (week 11).

Table A. 1: Difference‐in‐Difference results

Table A. 2: DID results when part‐time furloughs reduce working time by 50%

Figure A.2: Workplace visits in the Nordic countries and all U.S. states

Notes: The figure shows how workplace visits changed compared to the median weekly value, using the 5 week period from January 3 to February 6, 2020 as comparison. The U.S. states are shown in shades of light‐grey colors. The blue shaded vertical line indicates the date of the lockdowns in Denmark, Finland and Norway from Table 1, which is around March 13 (week 11). The dashed vertical line indicates Easter holidays (week 16). Source: Google LLC “Google COVID‐19 Community Mobility Reports.” https://www.google.com/COVID19/mobility/ [July 15, 2020].

Juranek, S. , Paetzold, J. , Winner, H. & Zoutman, F. (2021) Labor market effects of COVID‐19 in Sweden and its neighbors: Evidence from administrative data. Kyklos, 74(4), 512–526. Available from: 10.1111/kykl.12282