4.1 Methodology
To assess the impact of a severe crisis, such as the COVID-19 crisis, detailed information on household income is required. Due to the lack of up-to-date survey data, several different methods are used to forecast the impact of profound effects on the labor market at a micro-level. In the literature, two approaches are typically discussed [see, e.g., Gasior and Rastrigina (2017)]: re-weighting and modeling labor market transitions.
Re-weighting of the underlying micro-data can be used to adjust the micro-data to up-to-date macro-data. This approach has the advantage of accounting not only for changes in the labor market, but also for changes in the labor market structure. So far, several papers, such as Almeida et al. (2021a) and Li et al. (2020), have taken advantage of this modeling approach to estimate the impact of the COVID-19 crisis on household income, as well as its related indicators, such as the Gini index (income inequality) and poverty.
However, as argued by, e.g., Gasior and Rastrigina (2017) or Cantó et al. (2021), this approach has certain shortcomings. Firstly, the new pool of unemployed is assumed to have similar characteristics to that observed in the data, an assumption that can be disproved during the COVID-19 crisis, since its effects was driven by several lockdown measures and certain sectors were more severely impacted than others. Secondly, as regards the re-weighting approach, a detailed simulation of compensation schemes (such as STW schemes) cannot be directly taken into account. Thus, the potential heterogeneity across the income distribution of such schemes also cannot be accounted for.
Therefore, other papers, such as Christl et al. (2021a), Christl et al. (2021b), Cantó et al. (2021), Brewer and Tasseva (2020) and Figari and Fiorio (2020) have simulated adjustments to the underlying micro-data, using microsimulation techniques to model labor market transitions. The basic idea is to model transitions from employment to both unemployment and other compensation schemes (such as STW schemes). Given specific individual information, both the hypothetical unemployment benefit and wage compensation can be simulated and individual benefits can be estimated. This approach enables all micro-data to be updated, using all available information.
In this paper, we follow exactly this approach. We use detailed data of the EU Statistics on Income and Living Conditions (EU-SILC) in combination with EUROMODFootnote 3 to simulate the whole tax–benefit system of Austria. The version used is based on the policy year of 2020, combined with input data from EU-SILC 2018. Market income variables and non-simulated benefits are uprated to 2020, using specific uprating factors.Footnote 4 Labor market changes related to COVID-19 are simulated, using up to date detailed administrative data on the number of persons becoming unemployed and moving to STW schemes. This information allows us to replicate labor market changes by moving individuals from one state to another.
We then adjust the labor market characteristics and income of each individual, which changes the latter’s labor market status on micro-level. Additionally, we simulate the variables needed for the simulation of unemployment benefits (such as previous work history, previous wages, duration) and STW schemes (such as hour reduction, previous wages, duration). These adjustments are performed using the Labor Market Adjustment (LMA) add-on, which is a EUROMOD tool that can be used to simulate labor market transitions to employment, unemployment and monetary compensation schemes. The detailed description of the add-on can be found in the technical annex of Christl et al. (2021a).Footnote 5 Using EUROMOD, we can then recalculate the whole tax–benefit system, taking into account the new labor market status of individuals that have been observed as a result of the impact of COVID-19.
To identify those that transit to wage compensation schemes and unemployment, we use detailed information from the Public Employment Service Austria (AMS). These administrative data not only facilitate a detailed view of specific sectors, often argued to be a main driver of the unequal impact of the COVID-19 pandemic, but also an analysis of gender. Detailed information on the data will be discussed in Sect. 4.2.1.
4.1.1 Definition of simulation scenarios
Following Christl et al. (2021b), we base our analysis on the comparison of three different scenarios: firstly, a baseline scenario that is the 2020 policy scenario, not including the effect of COVID-19 and not including the impact of the pandemic on the labor market. Secondly, we consider a COVID-19 scenario that not only includes the simulation of related discretionary policy measures, but also the COVID-19-related adjustment of the labor market (transitions to both unemployment and STW schemes according to external, administrative information). Thirdly, we create a counterfactual scenario, in which we assume the COVID-19-related labor market shock; however, we assume the absence of the COVID-19-related discretionary policy measures. This allows us to estimate the impact of discretionary policy measures in mitigating the effect on household income during the COVID-19 crisis.
Let f be the tax–benefit function that depends on the tax–benefit structure (the specific policy rules in place), P as well as on the status of the labor market LM. We assume that the policy rules P can either constitute the standard rules that were in place before COVID-19, the so-called automatic stabilizers, \(P^{\mathrm{AS}}\), or can include the discretionary policy measures \(P^{\mathrm{COVID}}\). The labor market condition, LM, can either be a scenario without COVID-19-related changes affecting the labor market (\({\mathrm{LM}}^{{\mathrm{NoTrans}}}\)) or with COVID-19-related labor market transitions (\({\mathrm{LM}}^{{\mathrm{Trans}}}\)).
Therefore, we can define our three scenarios as follows:
-
Baseline scenario: \(f(P^{{\mathrm{AS}}}_{2020}, {\mathrm{LM}}^{{\mathrm{NoTrans}}}_{2020})\).
-
COVID-19 scenario: \(f(P^{{\mathrm{Covid}}}_{2020}, {\mathrm{LM}}^{{\mathrm{Trans}}}_{2020})\)
-
Counterfactual scenario: \(f(P^{{\mathrm{AS}}}_{2020}, {\mathrm{LM}}^{{\mathrm{Trans}}}_{2020})\)
Please note that in the counterfactual scenario, we assume that instead of entering into compensation schemes, people would only have access to the traditional automatic stabilization mechanisms, such as unemployment benefits. In this scenario, we assume the same loss in terms of hours worked, as in the COVID-19 scenario. However, the impact affects less people, since individuals that become unemployed reduce their working hours to zero, while under STW schemes, individuals can reduce their working hours to a certain level (retrieved from external data).
To estimate the direct COVID-19 effects \({\mathrm{PE}}^{{\mathrm{Covid}}}\) in 2020, we consider the changes between the first two scenarios, focusing on both changes in the labor market and policy changes (responses):
$$\begin{aligned} {\mathrm{PE}}_X^{{\mathrm{Covid}}}=X\Bigl (f\bigl (P^{{\mathrm{AS}}}_{2020},{\mathrm{LM}}^{{\mathrm{NoTrans}}}_{2020}\bigr )\Bigr )-X\Bigl (f\bigl (P^{{\mathrm{Covid}}}_{2020},{\mathrm{LM}}^{{\mathrm{Trans}}}_{2020}\bigr )\Bigr ) \end{aligned}$$
(1)
The function X can either constitute a certain income concept (disposable income or market income), but also indicators such as the AROP or the Gini coefficient.
We then define the policy effects of the traditional automatic stabilizers (in the absence of discretionary policy measures) related to a function, X, as the difference between the first and the third scenario.
$$\begin{aligned} {\mathrm{PE}}_X^{{\mathrm{AS}}}=X\Bigl (f\bigl (P^{{\mathrm{AS}}}_{2020},{\mathrm{LM}}^{{\mathrm{NoTrans}}}_{2020}\bigr )\Bigr )-X\Bigl (f\bigl ((P^{{\mathrm{AS}}}_{2020}, {\mathrm{LM}}^{{\mathrm{Trans}}}_{2020}\bigr )\Bigr ) \end{aligned}$$
(2)
Comparing the two policy effects (\({\mathrm{PE}}_X^{{\mathrm{AS}}}\) and \({\mathrm{PE}}_X^{{\mathrm{Covid}}}\)) allows us to gain an insight into the impact of STW and other discretionary policy measures.
4.1.2 Automatic stabilization coefficient
In crisis times, automatic stabilizers as well as discretionary policy measures play a central role in cushioning household income. To assess the income stabilizing effect of the Austrian tax–benefit system, as well as any of its individual components, we follow the approach of Dolls et al. (2012) that was also employed by Christl et al. (2021a) in a cross-country set up and by Christl et al. (2021b) for Germany and defines the Income Stabilizing Coefficient (ISC) as:
$$\begin{aligned} {\mathrm{ISC}} = 1 - \frac{\sum _i \Delta Y^{D}_{i}}{\sum _i \Delta Y^{M}_{i}} = \frac{\sum _i \Delta Y^{M}_{i} - \sum _i \Delta Y^{D}_{i}}{\sum _i \Delta Y^{M}_{i}} \end{aligned}$$
(3)
where \(\Delta Y^{D}_{i}\) is the disposable income change of an individual i and \(\Delta Y^{M}_{i}\) is the change in the market income of the individual i. An \({\mathrm{ISC}}=0.8\) would imply that 80% of the effect on market income is absorbed by the tax–benefit system.
Following this approach, we can further decompose the effect of several tax–benefit instruments, such as taxes, social security contributions and benefits, which are typically called automatic stabilizers. Additionally, and of special interest, is an analysis of the impact of discretionary policy measures (such as short-time work and other measures, e.g., the aforementioned one-off payments for the unemployed) on the automatic stabilization mechanism of the tax–benefit system.
We, therefore, define discretionary policy measures \({\mathrm{DPM}}_i\) as the sum of the benefit of STW \({\mathrm{STW}}_i\), the two one-off payments for the unemployed \({\mathrm{BUN}}_i^{{\mathrm{OOP}}}\), as well as the one-off payment for children \({\mathrm{BCH}}_i^{{\mathrm{OOP}}}\):
$$\begin{aligned} {\mathrm{DPM}}_i={\mathrm{STW}}_i+{\mathrm{BUN}}_i^{{\mathrm{OOP}}}+{\mathrm{BCH}}_i^{{\mathrm{OOP}}} \end{aligned}$$
(4)
We then further decompose the ISC:
$$\begin{aligned} {\mathrm{ISC}} = \frac{\sum _i \Delta Y^{M}_{i} - \sum _i \Delta Y^{D}_{i}}{\sum _i \Delta Y^{M}_{i}} = \frac{\sum _i \Delta T_{i} + \Delta {\mathrm{SIC}}_{i} - \Delta {\mathrm{BEN}}_{i} - \Delta {\mathrm{DPM}}_{i}}{\sum _i \Delta Y^{M}_{i}} \end{aligned}$$
(5)
where, \(T_i\) are taxes, \({\mathrm{SIC}}_i\) social insurance contributions, \({\mathrm{BEN}}_i\) benefits, \({\mathrm{DPM}}_i\) are all the discretionary policy measures paid or received by an individual i. Following this notation, we are able to decompose the income stabilization to the specific tax–benefit instruments.
4.2 Data
4.2.1 Administrative data for unemployment and STW
In order to evaluate the effect of a transition to STW or to unemployment, the labor market status of all individuals are adjusted, using monthly data from the Public Employment Service Austria (AMS). The simulation of the STW is based on the data available from March until December 2020, and takes into account information relating to the number of people in STW, the normal working hours and the reduction of working hours as a result of STW across sectors and gender. As an immediate impact of the COVID-19 crisis, unemployment increased. Figure 1 highlights that the first few months of the crisis, in particular, were critical: in April almost 600,000 people were registered as unemployed and almost twice as many were on STW. Therefore, more than 1.5 million people or almost 40 percent of the labor force, were either unemployed or in STW. Hence, STW has succeeded in limiting the impact of the COVID-19 crisis on the labor market and on unemployment.
Due to the lockdown and official closures, as well as the unequally distributed home-office possibilities, certain sectors have been more significantly affected that others. To capture this effect in the simulation of labor market transitions, we include detailed information in our estimates relating to the use of STW by sector.
Figure 2a highlights that from March until the end of the year 2020, more than half of the labor force within the sector, “accommodation and food service activities”, were either unemployed or in STW schemes. Focusing on the date, the effect of the first lockdown was greatest in this sector. At the end of April, more than 90% of people in this sector were either unemployed or in STW.
The utilization of short-time work differed substantially across sectors. More than every fourth employee worked to a limited extent in the areas of “accommodation and food service activities” and “arts, entertainment and recreation”, but STW was also used to a great extent in the sectors with the highest numbers of employees, namely the “wholesale and retail trade”, “repair of motor vehicles” and “manufacturing”. Our model will take these detailed sectorial differences into account.
Detailed administrative data allow us to calculate the share of the reduction in working hours, as highlighted in Fig. 2b and in Table 3 in the Appendix. This shows that the reduction in working hours peaked during the months of lockdown. During the period from March to December, the average reduction in working hours was equal to 53%. Nevertheless, the pattern in the reduction of working hours indicates huge differences across sectors. The reduction in working hours was highest in the sectors “arts, entertainment and recreation” and “accommodation and food service activities”, recorded in excess of 66%, while the sectors “mining and quarrying” and “water supply, sewerage, waste management” reported the lowest figures at less than 40%.
Due to general gender differences in employment by sector and the fact that certain sectors have been more significantly affected than others, our paper will also shed light on the gender differences in unemployment and STW, and the consequences on income. The gender difference in unemployment rate is highlighted in Fig. 3. We can see that before the COVID-19 crisis hit the Austrian labor market, the unemployment rate was slightly higher for males than for females. This, however, changed when first lockdown measures were introduced and general unemployment increased. While in February the unemployment rate of females was about 8% and the one for males about 10%, in April the rate increased to about 13% for males and 14% for females.
On the other hand, when looking at STW, administrative data reveal that the relative share of employees in STW (as a fraction of the number of employees) was higher for in the case of male employees at the beginning of the COVID-19 crisis. Nevertheless, since October female employees are on average slightly more likely to be in STW, compared to their male colleagues, as highlighted in Fig. 4a. During the last year, these two effects therefore, almost cancel one another out. On the other hand, Fig. 4b shows that there are gender differences in relation to the average reduction in working hours in STW. The reduction was greater in the case of female employees each month since the COVID-19 STW was introduced.
4.2.2 Estimating the duration of STW
There is no information in the administrative data for how long individuals stayed in STW schemes. We therefore set up a model based on survival probabilities to obtain estimates for the duration of STW scheme. As shown in Fig. 4a, we have detailed information on the number of individuals in STW in each month. This information is available by gender and each sector of activity. The STW started in March, and we assume that the total number of people entering in STW over the year is reached in the month with the highest share of people (April 2020). During the following months, we assume that some people managed to go back to work and no new persons entered into the scheme. Due to the second wave of COVID-19, there is a slight increase in the number of persons in STW in November and December. We assume that people entering in the schemes in November and December are employees who were already in STW in previous months.
To estimate the duration in STW, we sort the months in a descending order, based on the number of persons in STW in each month. This allow us to estimate the probability to go back to work in each month. Using these probabilities, we estimate the share of people staying from 1 up to 10 months, by sector of activity and gender.Footnote 6
Figure 5 shows the duration in STW of people that moved to short-time work by gender and sector. We see that related to the duration, there are no big differences across gender. However, we find substantial differences across sectors, highlighting the importance of taking sector specific information into account. While the duration for people working in public administration was on average very short (more than 90% of workers stayed 5 months or less in STW), workers in the hotels and restaurants were on average very long in STW (almost 25% of them stayed the full 10 months of 2020 in STW schemes).
