Job Transitions during the Covid-19 Pandemic

class: center, middle, inverse, title-slide

.title[
# Job Transitions during the Covid-19 Pandemic
]
.subtitle[
## For Better of for Worse?
]
.author[
### Melanie Arntz, Sarra Ben Yahmed, Eduard Brüll and Michael Stops
]
.date[
### NGE Spring Workshop in Hohenheim and Mannheim - 20.04.2023
]

---

class:topbar
#Motivation

---
class:topbar
#Introduction
.smallest[
###Differential impact of COVID-19 pandemic across regions and occupations
- Demand for various occupations changed drastically
 + Some sectors recovered quickly while others faced high rates of short-time work
 + Similar differences between regions (i.e. tourist regions vs. others)
- Long-term inequalities could arise due to the severity of impact on certain occupations
  + Workers in severely affected occupations might need to adapt to the crisis by finding work in other occupations
  + Workers in hard hit occupations were less likely to find new jobs in the same occupation due to the fall in demand
  + Lack of social protection in part-time and temporary help jobs in hospitality/retail increased the incentive for affected workers to change occupations
  
> Opportunities to move to jobs that require a similar skill set vary greatly between workers in different types of affected occupations and between local labour markets.

]
---
class:topbar
# This paper
.smallest[
### Objectives

- Analyse how the COVID-19 crisis affected occupational mobility in Germany
- Consider how job transitions depended on the occupational structure of the local labour market prior to the crisis
- Examine how workers' long-term labour market outcomes responded to the occupation-specific shocks in different local labour markets.

### Approach
- Develop a new measure of occupational switching opportunities
    + Combine measures of occupational distances with vacancy information
- Compute forecast-based counterfactuals and a generalised difference-in-difference setting to get at causal effects of changes in occupational switching opportunities during the crisis 
- Examine changes in switching rates and job quality using individual social security data
]

---
class:topbar
# Contribution
.smallest[
### Our paper contributes to three strands of economic literature :

1. Workers' responses to local labor market shocks
  - *Redondo (2022):* Job transitions of workers in the construction sector during the Great Recession in Spain
  - *Arntz et al (2022):* Effect of job loss depends on the evolution of demand for tasks in a region
  - **Our contribution:** Long-term effects due to local occupation structure and occupation-specific shocks

2. Job transitions after the COVID-19 shock
    - *Hensvik et al.(2021):* Job seekers in Sweden redirected their search effort towards less badly hit sectors
    - *Costa Dias et al. (2021):* Pandemic was detrimental to workers' careers in the UK 
    - **Our contribution:** Occupational mobility due to the COVID-19 crisis in Germany

3. Long-term effects of occupational mobility on labor market trajectories
  - **Our contribution:** Examine crisis-induced shifts in occupations and how they differ from normal mobility.

]

---
class: inverse, middle, center
# Measuring the COVID-19 schock on occupations 
<html><div style='float:left'></div><hr size=1px width=1100px></html>
---
class:topbar
# Measuring the COVID-19 schock on occupations
.smallest[

**Novel measure of occupational switching opportunities due to labor demand changes**
- Individuals in the same occupation in different areas may have different incentives to switch due to:
    - Differences in shock to labor demand across space
    - Ability to switch depends on availability and similarity of tasks in other occupations
- Use changes in vacancy postings to identify shifts in labor demand  
`$\rightarrow$`  Use vacancy data from Federal Employment Agency  
- Use BIBB survey to measure overlapping task-content between occupations    
`$\rightarrow$` Compute Gathmann and Schönberg (2010) occupational distance from BIBB task information
- Compare available vacancies `$v_{tro}$` inoccupation `$o$`, region `$r$` and month `$t$` to distance-weighted average of vacancies in all other occupations in the same region `$v_{tro}^\prime$`  
`$\rightarrow$`  Compute vacancy ratio: `$\text{vr}_{tro} = \frac{v_{tro}^\prime}{v_{tro}}$`   
`$\rightarrow$` Measure of local opportunity to switch for each occupation

]
---
class:topbar
# Forecast-based counterfactuals for the vacancy ratio
<img src="data:image/png;base64,#04-presentation_files/figure-html/occ63_prediction_cf-1.svg" style="display: block; margin: auto;" />
---
class:topbar
# National deviations from prediction
<img src="data:image/png;base64,#04-presentation_files/figure-html/nat_dev-1.svg" style="display: block; margin: auto;" />
---
class: topbar
# Exposure measure
.smallest[
Using both forecast-based counterfactual and actual data, we can compute measures for exposure to the Corona-induced changes in the vacancy ratio:

`$$\text{Average Exposure}_{ro} = \frac{\sum\limits_{t >  Feb.\hphantom{x}2020} \left(\text{vr}_{tro} - \widehat{\text{vr}_{tro}}\right)}{\sum\limits_{t >  Feb.\hphantom{x}2020}\widehat{\text{vr}_{tro}}}$$`

This measure captures the shift in the availability of vacancies outside an individual's occupation relative to vacancies within their occupation during the Covid-19 pandemic. It depends on an individual's pre-crisis occupation `$o$` and region of residence `$r$`.
]
---
class: topbar
# Average exposures by region for different occupations

.pull-left[
<img src="data:image/png;base64,#04-presentation_files/figure-html/map_h-1.svg" style="display: block; margin: auto;" />
]
.pull-right[
<img src="data:image/png;base64,#04-presentation_files/figure-html/map_m-1.svg" style="display: block; margin: auto;" />
]
---
class: topbar
# Exposures and switching rates
<img src="data:image/png;base64,#04-presentation_files/figure-html/switch_exp-1.svg" style="display: block; margin: auto;" />
---
class: inverse, middle, center
# Data
<html><div style='float:left'></div><hr size=1px width=1100px></html>
---
class:topbar
# Individual social-security data
.smallest[
###Monthly panel of  individual social security records
- Detailed information on employment status and occupation
- 80% of the German labor force (excluding civil servants)
- Data from 2016 to 2021
- Balance panel (carryforward leavers source occupation info)
- Detailed information on job switches
- Detailed wages (censored, but censoring irrelevant for most affected occupations)
- 2% sample at the moment but could expand to universe of data

###Main outcomes

- Switching rates by occupation
- Pre-computed AKM-firm-effects to identify switches from low-paying to high-paying firms
- Add further occupational info from the BIBB data to get at changes in occupational quality
]
---
class:topbar
# Switching rates during the pandemic
<img src="data:image/png;base64,#04-presentation_files/figure-html/descriptive_switches-1.svg" style="display: block; margin: auto;" />
---
class: inverse, middle, center
# Empirical Strategy
<html><div style='float:left'></div><hr size=1px width=1100px></html>
---
class:topbar
# Empirical strategy
.smallest[
###Directly using the prediction-based counterfactual

Simplest possible estimation using the prediction-based counterfactual:
  `$$Y_{itro} =  \beta  \text{Exposure}_{tro} + \alpha_r + \gamma_o + \lambda_t +  \varepsilon_{itro},$$`
for outcomes `$Y$` like regional occupational switching or occupational quality measures of switches

**Disadvantages:** 
- Unlikely that reactions to the shock are immediate
- Timing unknown

**Alternative:**  
Keep only average exposure by occupation and region and get at timing using an event-study approach
]
---
class:topbar
# Empirical strategy
.smallest[
###Difference-in-differences  with continuous treatment

Get timing from an event-study specification:
`$$Y_{itro} =  \sum_{t\neq0} \beta_t \left( \text{Average Exposure}_{ro} \times D_t \right) + \alpha_r + \gamma_o + \lambda_t +  \varepsilon_{itro}$$`

The reference period `$(t=0)$` is February 2020.

**Advantages:**  
- Check differential pre-trends
- Can estimate timing and use in direct specification

**Disadvantages:**   
- Stronger parallel trends assumption than classical DiD
]
---
class: inverse, middle, center
# Results
<html><div style='float:left'></div><hr size=1px width=1100px></html>
---
class: topbar
# Occupational switching by avg. exposure terzile
<img src="data:image/png;base64,#04-presentation_files/figure-html/occ_switch_result1-1.svg" style="display: block; margin: auto;" />
---
class: topbar
# Occupational switching by avg. exposure event study
<img src="data:image/png;base64,#04-presentation_files/figure-html/occ_switch_result_ev1-1.svg" style="display: block; margin: auto;" />

---
class: topbar
# Occupational switching by avg. exposure event study
<img src="data:image/png;base64,#04-presentation_files/figure-html/occ_switch_result_ev2-1.svg" style="display: block; margin: auto;" />
---
class:topbar
# Occupational switching by avg. exposure event study

---
class:topbar
# Occupational switching by avg. exposure event study

<img src="data:image/png;base64,#04-presentation_files/figure-html/occ_switch_result_ev4-1.svg" style="display: block; margin: auto;" />
---
class:topbar
# Occupational switching by avg. exposure event study

---
class: inverse, middle, center
# Conslusion
<html><div style='float:left'></div><hr size=1px width=1100px></html>
---
class: topbar
# What's next?
1. Change estimation strategy to use time-varying exposure given that we see immediate reactions during high-demand slump periods
2. Use an additional measure that use past occupational switching instead of a task distance 
3. Analyse destination and source jobs together:
  - More permanent or the same?
  - Use BIBB survey data on job quality measures 
  - Use AKM-Effects to compare previous current firm of workers

---
class: topbar
#Preliminary conclusion

- New insights into the effect of the Covid-crisis on occupational switching
  + Novel measure combining labour demand and task-based switching possibilities was used
  + Hospitality sector suffered the most `$\rightarrow$` strongest increase in switching,
  + Almost imediate reaction during biggest demand changes 
  + During lockdowns, less-educated, low-wage workers in marginal employment relations were affected the most
- First step to study where excess switchers end up
  + For better or for worse?