Failing Young and Temporary Workers? The Impact of a Disruptive Crisis on a Dual Labour Market

2.1 Covid-19 in Portugal and Policy Responses

The government closed schools and imposed circulation restrictions on the border with Spain shortly after the first confirmed Covid-19 case on March 2nd 2020. On March 18th, the President declared the State of Emergency, which lasted until May 3rd, when it was substituted by a less severe but still fairly constrained State of Calamity. As a consequence, all retail was shutdown, with the exception of supermarkets, pharmacies, and gas stations. Restaurants were only allowed to serve take-away. Further restrictions on circulation and mandatory homeworking for compatible jobs were also decreed.^[6]

The strong impact on the Portuguese economy is documented in Carvalho et al. (2020), using data from electronic purchases. They provide causal estimates of decreases in year-on-year growth rates of 16, 37, and 28 percentage points on overall purchases in March, April and May, respectively. The authors also show that the impact was very uneven across sectors. The Hospitality sector (including restaurants, coffee shops and accommodation), Fashion and Beauty, and Transportation were among the most affected. The differential impact across sectors induced asymmetric effects of the crisis. Average wages in the Hospitality, and Fashion and Beauty sectors are smaller than the national average. The share of low educated and foreign workers in Restaurants is higher than the national average. At the same time, 2022, show that the sectorial composition of the municipal economies leads to asymmetries in the geographic effect of the crisis.^[7] Importantly, as shown in Peralta et al. (2021), the incidence of non-permanent contracts in these sectors was larger than for the overall economy.

These sizeable consequences were met with policies targeting the labour market, of which we highlight three. Firstly, the Portuguese Covid-19 special furlough scheme was implemented on March 26th The scheme covered labour costs for the whole duration of compulsory shut-down, and was in place until July 2022.^[8] Administrative data from Social Security shows that 1.2 million workers were supported, mostly concentrated in the Retail and Hospitality sectors.^[9] Firms that benefited from the furlough policy could not dismiss workers on permanent contracts. They could, however, choose not to renew temporary contracts, stop hiring workers with no formal contract, and not replace workers who depart voluntarily.^[10]

The bulk of the effects on unemployment should arise from these margins by recipient firms, or from firms that did not qualify or apply for the scheme. Importantly, furlough allows firms to suspend employment contracts or reduce employees’ working hours, with social security covering two-thirds of the wage. Therefore, keeping the workers is costly.

Secondly, remote work was made compulsory for compatible occupations. Survey data from Statistics Portugal shows that there were around 1 million remote workers in the Spring of 2020, mostly highly educated and high-income ones. Their average wage was 50% higher than that of non-remote workers, and 70% of them had a higher education degree.

Finally, access to unemployment benefits was eased as of July 25, encompassing permanent or temporary employees, and halving the minimum qualifying contract duration in the past 24 months. Since our period of analysis ends in August, this legal change is unlikely to affect the flows of unemployed individuals. Moreover, mandatory job search and training were suspended between March and May.

2.2 Labour Market in Portugal

Permanent workers in Portugal benefit from one of the highest levels of employment protection across the OECD, which, according to the European Commission, increases both the reluctance of firms to hire permanent employees (European Commission 2018) and the importance of non-permanent employment (OECD 2017).

Figure 1 shows that temporary employment, i.e. wage and salary workers whose job has a pre-determined termination date, accounts for 22% of all employees in Portugal, above the OECD average of 12%, and only exceeded by Spain (27%) and Poland (24%). This makes Portugal an interesting laboratory to study the impact of a major disruptive shock (such as the pandemic), with policies that create asymmetric incentives regarding job separations for different types of workers.

Figure 1:

Share of temporary employment (as % of employees), 2018.

Source: OECD.

Moreover, legal restrictions on nominal wage cuts, combined with the long recent period of low inflation, leave employers with little margin to adjust real wages. As a consequence, in periods of crisis, employment (and especially temporary employment) becomes the main margin of adjustment (Carneiro et al. 2014; Martins and Portugal 2019).

This is confirmed by Figure 2, constructed with data from the Labour Force Survey. It shows that the YoY quarterly change in employment in 2020 was positive in the first quarter and only marginally negative in the second, for permanent workers, in sharp contrast with the strongly negative for temporary ones.

Figure 2:

YoY change in temporary and permanent contracts. Source: Labour Force Survey (Statistics Portugal).

Importantly, temporary contracts are more prevalent in some groups of the population, as shown in Figure 3, which displays the share of temporary employment as a percentage of total employment between the first quarter of 2018 and the second quarter of 2020, split by gender, age, and education level. Given the concentration of low and medium skilled workers in this group, they face significant difficulties in finding a new job (Blanchard and Portugal 2017), and are prone to be long-term unemployed. In sum, the characteristics of the labour market place some workers in a more vulnerable position in the face of a crisis.

Figure 3:

Temporary employment in Portugal. Source: Labour Force Survey (Statistics Portugal).

According to panel (a), the share of temporary contracts decreased in 2020, without any consistent difference between female and male workers. The age differences, shown in panel (b), are the most striking. The share of workers aged less than 25 years old with temporary contracts is around four times that of those older than 35. In the 25–34 age interval, the prevalence of temporary work is twice as much as that of the older individuals. Finally, from panel (c) it is clear that temporary employment is more prevalent among individuals with upper secondary education. The fact that individuals with basic, primary education, or less, represent the group with the lowest share of temporary contracts is mostly driven by age. These education levels are more common in the oldest cohorts, who also have a more stable relationship with the labour market.

Figure 2 suggests that the decrease in temporary contracts is explained by an increase in job separations that hits them disproportionally.

3.1 Data

We use administrative data from Instituto do Emprego e Formação Profissional (IEFP), the Portuguese Public Employment Service, provided at the municipality level.^[11] As of age 16, unemployed individuals must register in Public Employment Services to receive unemployment benefits and have access to active labour market policies. In return, they must comply with a personal employment plan and actively seek employment. The Public Employment Service collects job offers and advertises them both online and through its network. When offered a suitable or socially necessary job, unemployed individuals must accept the offer, or they risk losing the benefit.^[12]

Jobless individuals have two incentives to register: qualify for the active labour market policies, and receive unemployment benefits, when eligible. Between 1999 and 2019, the number of individuals registered at the Public Employment Services represented an average of 94% of the unemployed population (according to Statistics Portugal).^[13]

IEFP provides monthly data on the number of unemployed individuals registered and the number of new job placements that take place for those registered at one of the job centers distributed across the country. Our sample comprises data on the 278 municipalities of the Portuguese mainland between October 2016 and August 2020.^[14] We drop September in the four years, given our identification strategy, explained in Subsection 3.2.

Data on the unemployed is split into gender, age group, and education level. Data for job placements is disaggregated by gender. We also use data on the main reasons for registration at the job centers, i.e. dismissals, voluntary quits, mutual agreement dismissals, end of temporary jobs, self-employment or former inactivity.^[15] It is worth noticing that while data on unemployment refers to the situation at the end of each month (stock), data on job placements and the motives for registering at IEFP refers to the movement throughout the month (flow). Summary statistics of all variables for the average municipality are provided in Table 1.

The average number of adults registered in Public Employment Services per municipality is 1274, with a minimum of 23 and a maximum of around 26 thousand. The number is higher for females and adults between 35 and 54 years old, when compared to males and other age categories, respectively. Registrations also vary according to the level of education, with those with upper secondary education having the highest number of registrations. The average number of job placements per month and municipality is 24.

In terms of motives to register at the IEFP, end of temporary job is the most frequent (on average 73 people per municipality), followed by dismissal from previous job (19) and the registration of formerly inactive worker (15). Finally, we note that job placements are very low when compared to the flow of registrations at the Public Employment Service.

We complement the unemployment data with four further data sources.^[16] Firstly, we use the last available wave of Quadros de Pessoal (2018), a linked employer–employee dataset covering the universe private-sector firms based in Portugal with at least one wage earner, aggregated to the municipal level. We retrieve the share of temporary workers per municipality and per sector and we also use it to relate temporary jobs to the working-from-home measure of Dingel and Neiman (2020). At the municipal level, the average share of temporary employment contracts is 33%, with a standard deviation of 8%.

Secondly, we use municipal-level data from SIBS (acronym for Sociedade Interbancária de Serviços, in Portuguese), which manages the integrated banking network in Portugal, comprising Automated Teller Machines (ATM), Point-of-sales (POS) terminals, and other electronic payment technologies such as mobie e-money. SIBS is the largest player in the electronic payments market and it runsthe interbank compensation system through a contract with the Portuguese Central Bank.^[17] The data comprises electronic payments, i.e. payments with bank cards, including those with contactless technology, and several digital money solutions (both mobile phone and net banking based), made in Portugal, by domestic and foreign costumers.^[18] This data allows us to compute a real-time measure of the crisis per sector of activity.

Thirdly, we use sectoral-level administrative data from the Ministry of Social Security and Employment on the number of workers benefiting from the furlough scheme implemented by the Portuguese government per sector of activity. Finally, we use individual data from the Labour Force Survey between the first quarter of 2018 till the second quarter of 2020 to highlight some stylized facts about the impact of the Covid-19 crisis.

3.2 Methodology

We implement a difference-in-differences (DiD) event study. In Section 5, we use triple difference-in-differences to assess the role of duality and exploit heterogeneous effects. Our identification strategy is easily explained by analysing Figure 4. The treatment and comparison groups are sets of months. The treatment group is represented by the blue lines in both panels, i.e. it comprises the months between October 2019 and August 2020. The comparison group comprises the remaining lines, i.e. the same sequence of months lagged one, two, or three years, respectively. The treatment period includes the months between March and August for all the years between 2017 and 2020.

Figure 4:

Identification strategy. Time-series of Log of average unemployment and new job placements. The blue line represents the period between October 2019 and August 2020 (the treatment group), of which October 2019 to January 2020 correspond to the pre-treatment period, and March 2020 to August 2020 correspond to the treatment period. The remaining lines represent the control group, i.e. the same series lagged one (Oct. 2018–Aug. 2019), two (Oct. 2017–Aug. 2018) and three (Oct. 2016–Aug. 2017) years.

The identifying assumption is that, absent the pandemic, and the implemented policies (detailed in Subsection 2.1) the year-on-year growth rate between the month of March (resp., April, May, June, July, and August) 2020 and the corresponding month in 2019 would be equal to a weighted geometric mean of the year-on-year growth rates for the same month, lagged between 1 and 3 years. Thus, the treatment group (Oct. 2019–Aug. 2020) can be decomposed into two parts, the pre-treatment period (Oct. 2019–Jan. 2020), used to assess parallel trends, and the treatment period (Mar. 2020–Aug. 2020), used to assess the impact of the pandemic.

The evidence in Figure 4 suggests that the outcome variables are on parallel trends prior to the shock. We provide a more formal test in Section 4 below, in which we run event studies for these outcome variables, using the same set of months for the control and treatment groups. Figures C.3–C.5 in the Appendix C show analogous evidence with the unemployed split by gender, age, and education levels.

The treatment period begins in March 2020. Recall that the first case was diagnosed on March 2nd. Carvalho et al. (2022) show that there is no evidence of changed behavior in electronic purchases before March. This is confirmed by Google mobility data shown in Figure C.2 (borrowed from Carvalho et al. 2022) that displays no evidence of changes in mobility in February or the first two weeks of March. Moreover, note that around the Carnival holiday of February 24, we observe a downward peak in workplaces and retail, accompanied by a mirror increase in parks and other open areas. This is convincing evidence that the country was living a normal life at the end of February. Hence, we rule out potential anticipatory behaviours that would threaten our identification strategy.

4.1 Transitions in a Dual Labour Market

We now analyse the evidence about the reasons that led the individuals to register with the Public Employment Service during the Covid crisis. Since these are flow variables, the results measure the impact on new unemployment each month, net of composition effects.

Figure 6:

Motives to register with IEFP.

The data splits the motives into four categories that are explicit job separations (end of temporary contract, dismissal, voluntary quit, mutual agreement dismissal) and two additional categories for transition from inactivity and registration of self-employed. Figure 6 shows the event study for each motive. The evidence confirms the importance of temporary contracts for the transitions during this period. The most important ones, for the purposes of our analysis, are the first two. Note that dismissals also include temporary workers who are dismissed before the end of the contract. The pandemic had no sizeable impact on the remaining transitions, except for the drop in inactive workers that register as job-seekers, which is a natural consequence of the lockdown period.

Panels (a) and (b) exhibit very clear spikes in terminations of temporary contracts and dismissals, respectively, lasting until the end of the analysed period. The effects were particularly strong in April, with YoY growth rate increases of 87 p.p. and 216 p.p. for end of temporary contracts and dismissals, respectively. Although the effect is lower for termination of temporary contracts, the base effect is much stronger (Figure C.8 in Appendix C shows that almost two thirds of job separations in 2020 are due to the termination of temporary contracts.)^[22]

4.2 Duality in Municipal Labour Markets and the Pandemic Shock

We now explore whether unemployment is driven by municipal heterogeneity in the dual nature of the labour market, i.e. we analyse the possibility that municipalities with a higher share of temporary contracts are more impacted by the crisis. We use the municipal share of temporary contracts in 2018, the last available year. As we show in Figure C.9, in Appendix C, this share is strongly correlated with that of previous years, which suggests that we capture a structural feature of the local labour markets.

Table 3 shows the estimate of α ₃ in Eq. (3) for the stock of unemployed (column 1). The estimate indicates that there are more registrations at the Public Employment Services in municipalities with a higher share of temporary employment. The number of newly registered unemployed increases by 1.5% with an increase of 1 p.p. in the share of temporary workers in a municipality. Alternatively, a one standard deviation (8%) increase in the share of temporary contracts amounts to an 11.6% change in the number of registries. This effect is sizeable, particularly given that the average share of temporary workers is 33%.

Columns 2 to 8 report similar regressions where the outcome variable is the flow of monthly registrations by motive, for all the motives shown in Figure 6. Results show that new registrations in Public Employment Offices due to the termination of temporary contracts are higher in municipalities with a higher share of temporary workers. The interaction with the share of temporary contracts is not significant for the remaining motives. This provides further evidence that temporary workers were more severely affected by job loss during the early months of the pandemic.

5.1 Heterogeneous Effects of Covid-19 on the Labour Market

In this section, we use a triple difference-in-differences strategy to explore the heterogeneous effects of the Covid-19 shock on the outcomes of interest for (i) gender, (ii) age groups, and (iii) education levels. Results are reported in Table 4. As mentioned in Subsection 3.1, data on new job placements is only disaggregated by gender, thus the heterogeneity in terms of age and education focuses exclusively on unemployment.

Columns 1 and 2 of Table 4 report the estimates of β ₃ and β ₄ from (4) for the gender heterogeneity specification, where β ₄ represents the additional impact for females. In all specifications, the estimates of β ₃ represent the impact for the reference group. When we estimate heterogeneous impacts other than gender, we expand the interaction of β ₄ to include a full set of interactions for each category, i.e. three coefficients instead of one for the age specification (column 3 of Table 4), and four for the education one (column 4 of Table 4).

Results for gender heterogeneity indicate that Covid-19 increased male unemployment by 33.8% (column 1) and decreased new job placements by 24.1% (column 2) between March and August 2020. Taken together, our results show that women were most severely hit by the pandemic: while there is no statistically significant difference in unemployment, women suffer an additional drop of 17.5% in placements after March, when compared to men. The absence of gender differences in unemployment is in line with the findings of Casarico and Lattanzio (2020) for Italy, and Hupkau and Petrongolo (2020) for the UK. The negative effect that we identify on placements suggests that women are less likely to find a job following an unemployment episode during the pandemic; this adds to the long list of differential gender impacts due to a higher proportion of female workers in the most affected industries (ILO 2017), disproportionate take up of household chores and childcare after school closures and work from home restrictions (Del Boca et al. 2020; Farré et al. 2020).

To study the impact on different age groups, we use the unemployed aged more than 55 as the reference group. Our findings in column 3 of Table 4 show this is the least affected group. There is a very strong impact on youth unemployment after March 2020, amounting to an additional increase of 20.8% and 25.8%, for individuals younger than 25, and between 25 and 34 years old, respectively. These results are consistent with the prevalence of non-permanent contracts among younger workers in Portugal documented above. Younger workers are also less unionized than older ones (Portugal and Vilares 2013) and more vulnerable to precarious working conditions.

With regards to education levels, column 4 of Table 4 shows that individuals with higher education, the reference group, experienced an increase of 23.9% in unemployment after March 2020. The differential impact of the crisis was 17.5% higher for individuals with upper secondary education, 15% for individuals with lower secondary education, and 9.6% for individuals with basic education, than for those with higher education. Individuals with less than four years of formal education do not behave differently than the highly educated ones. This may stem from the fact that these workers are concentrated in essential jobs that kept working during the lockdown; however, the lack of statistical significance may also be explained by the relatively small number of workers with this level of education.

The inverted u-shaped relationship between education levels and the labour market impact of the pandemic can be explained by the fact that workers with secondary education are usually not employed in the subset of service sectors compatible with home working, and are thus in a more vulnerable position in this crisis. Indeed, data from Statistics Portugal shows that during the second quarter of 2020, 4.7% of the employed population with lower secondary education or less was working from home, compared with 53.8% of the population with higher education degrees (INE 2020).

5.2 Does Labour Market Duality Drive the Heterogeneous Effects?

So far, we have established that (i) the increase in unemployment was higher in more dual municipal labour markets, (ii) temporary contract terminations were more pervasive in these municipal labour markets, and (iii) the impact of the crisis falls disproportionally on females (through the hiring margin), younger workers, and those with secondary education. Since these worker groups are the ones that are over-represented in temporary contracts, this also constitutes evidence that the termination of temporary contracts plays a major role in our results. We now combine the two pieces of evidence and test whether the heterogeneous impact (along gender, age, and education) is greater in municipal labour markets with a stronger duality character.

The estimates of α ₃ from (3) for each group of workers are presented in Table 5. We report the tests for the null hypothesis that the estimated coefficients are equal in Tables B.4 and B.5 in the Appendix B. As regards gender, the value of the χ ² statistic is 4.03, i.e. the coefficients are statistically different at 5%. More specifically, a one standard deviation increase in the share of temporary contracts in a dual labour market increases the number of newly registered female workers in response to the pandemic shock by 12.6%, and that of males by 10.7%.

We now analyse the results with respect to age. Interestingly, the estimated coefficients are monotonic in the age of the individuals. The tests of the difference of the coefficients in Table B.4 indicate that there is no statistical difference between the two groups who are younger than 34. Conversely, the coefficients for both these groups are different from the ones of older workers. The negative impact of the pandemic shock in dual labour markets is therefore more concentrated in younger workers. More specifically, a one standard deviation increase in the share of temporary workers leads to an increase of between 14.2% and 16.5% in the number of newly registered young workers in the Employment Office. Conversely, the impact on older workers amounts to between 7.4% and 11.4%.

Regarding education, the effect follows an inverted U-shaped pattern. In municipalities with a higher share of temporary contracts, Covid-19 impacts less severely individuals with primary education or less, and individuals with higher education. The strongest impact falls on the those with lower and upper secondary education, as the results of the tests of the equality of the coefficients reported in Table B.5 confirm. A one-standard deviation increase in the share of temporary workers in the municipality increases the number of registered unemployed people with upper and lower secondary education by 12.6% and 13.7%, respectively. The impact on highly educated workers is non significant.

These results strongly suggest that the effect of the crisis in female, young and middle-educated workers is driven by the duality of the labour market. The crisis has asymmetric effects depending on the workers’ ties to the labour market, and measures like the furlough scheme do not seem to be enough to protect certain groups of the labour force.

In order to correct for possible seasonality in temporary contracts, we re-estimate the regressions excluding the municipalities in the top quartile of the distribution of tourist overnight stays, which are bound to have a peak of jobs in the Summer. The results, presented in Tables B.6 and B.7, in Appendix B, are similar to the baseline, with the exception of the gender differences, which are less precisely estimated.