Resilience to Automation: the Role of Task Overlap for Job Finding
By Diego Dabed, Sabrina Genz and Emilie Rademakers
Technology changes the demand for different skills. Automation, for example, eliminates some routine tasks, reducing the demand for related skills. On the other hand, new technologies can involve entirely new tasks that require new skills, or they might require greater proficiency in existing tasks. The changing nature of skills is important because it affects how quickly workers who are displaced by technology can find new jobs. Even if automation does not change the net number of jobs, displaced workers might bear significant costs if it takes them a long time to find a new job.
This paper investigates to what extent newly unemployed people are able to find new jobs that match their skill sets. To do this, we focus on a sample of newly unemployed job seekers in Flanders, Belgium. Importantly, we measure the similarity of workers’ job experience to the job opportunities they target for a new job. We measure this similarity by comparing the words used in formal job descriptions. We can then use this similarity to predict the probability of finding a new job and how long the worker is likely to be unemployed. And we can do this for different groups of workers based on the kind of technology they might have been exposed to in their former jobs.
This analysis provides us with some important takeaways. First, our empirical analysis shows that only the most similar jobs outside of jobseekers’ own labor markets are relevant for job finding.
Second, the duration of unemployment varies with the type of technology exposure. Using the job descriptions of the worker’s previous employment, we identify their “exposure” to different technologies by the nature of the words used in the task description. Automation exposure corresponds to tasks that are routine in nature, that is, they have a high “Routine Task Intensity.” Presumably, these jobs are more likely to be automated. Similar word clusters are used to identify software and AI exposure. Using our similarity measures we predict unemployment duration with two models, one allowing labor markets to overlap and one with no overlap. We find that greater Routine Task Intensity in the former job is associated with longer unemployment duration, suggesting that workers displaced by automation have a harder time finding a new job, perhaps because automation might be eliminating jobs with routine skills. Comparing our two models, we find that the positive gradient between Routine Task Intensity and unemployment is unaffected. This implies that overlapping job markets requiring very similar skills, provide little relief from having higher Routine Task Intensity in the former job. On the other hand, for recent software innovations, we find that overlapping markets provide discernible improvement in the expected job finding rates of exposed job seekers. We show that current AI exposure is linked with faster job finding.
These findings provide important policy implications that we explore using simulations. We find increasing the matching efficiency with the most similar markets increases job seekers’ exposure to automation in close job markets and reduces their job finding probabilities. Our counterfactual simulation points to potential harm in recommendation algorithms that direct job seekers to explore job markets that are most similar to their previous job’s task content. Rather, it may be more relevant to actively direct job seekers towards markets that provide better job opportunities, as can be found in examples of sectoral work programs (Katz et al., 2022) and active labor market policy for the unemployed (Behaghel et al., 2022).