Transcript

Tim Sablik: My guests today are Katarina Borovičková and Claudia Macaluso. Katarina and Claudia are both senior economists in the Research department at the Richmond Fed. Katarina and Claudia, welcome back to the show.

Katarina Borovičková: Thank you very much for having us back.

Claudia Macaluso: Yeah, it's great to be here.

Sablik: Our topic for today is a three-part Economic Brief series that you recently published on RichmondFed.org. The articles take a look at long-term unemployment. Long-term unemployment is the share of workers who have been unemployed for 27 weeks or longer. So, what has been happening to that share recently?

Macaluso: Before getting to what's been happening, one thing that's useful to think about is how unemployment duration, as we economists call it, is actually measured. The main source of labor market data in the United States is the Current Population Survey, or CPS. That's a survey that goes out to households every month.

Among the many things they ask, there are questions about employment status. One of them asks whether people are at work and, if they are not, if they have a job. If the answer is no to both of those questions, then that's when you are called non-employed. You have to answer another question, which is, "Are you looking for a job, and would you be able to take one?" That's when you become unemployed. These are not workers that are taking a break or rethinking some of their goals. These are workers who are not at work and looking for a job and will be able to take one, and they have been doing that for at least 27 weeks, which is a long time.

Borovičková: Since July 2022, the share of workers who have been unemployed for a long time has been increasing. Even though in the last few months we can say that this share has stabilized somewhat, it is still at the elevated level compared to, say, the beginning of 2023.

This is unusual. The U.S. has not been in a recession any time since 2022. So, we have seen something that we typically see in a recession outside of a recession.

Sablik: Katarina, maybe you can talk a little bit more about which mechanisms seem to be behind the recent rise in long-term unemployment.

Borovičková: It's very helpful to think about a model of inflows and outflows into unemployment, and the easiest way to think about it is to think about a bathtub. There are two sources of how the volume of water can change in the bathtub. There is a faucet through which water flows into the bathtub and there is a drain through which the water flows out of the bathtub. What can increase the volume of water in the bathtub is when the inflow of water is faster, or the inflow of water into the bathtub is the same but the water is not flowing out of the bathtub quickly enough.

Let's now link that to unemployment. The volume of water in the bathtub are the unemployed people and the inflow is coming from employment. There are people who are employed and lose their jobs, and if they start losing jobs at the faster rate, that is going to contribute to the increase in the number of people who are remaining unemployed. On the other hand, the drain has become smaller — people are just not leaving unemployment at a faster rate. What this represents is the job finding probability.

Both of these are measurable in the CPS data that Claudia has just described, and we did look at that. What we find is a very clear pattern that even though we have seen a little bit of a change in the separation rate — meaning how quickly people become unemployed — the whole increase in unemployment is driven by the fact that people are not leaving unemployment at a fast enough rate. Something has happened to the job finding rate.

Sablik: Right, the drain has slowed in the bathtub.

Claudia, given the differences between these two forces, depending on which one dominates — in this case, it seems to be the job finding rate — does that suggest different things about the state of the economy?

Macaluso: It does. A typical business cycle starts with a spike in layoffs. People lose their jobs in larger numbers than before — this tends to be relatively sudden and then subside relatively quickly. Then, the job finding rate slows down — a lot of people have lost their jobs and their ability to hop on to another job is slowing down. The job finding rate recovers after a recession quite a bit slower than the layoff rate.

It is a really unusual environment. We did not have a recession. We did not have a big spike in layoffs. But we do have a low job finding rate, and declining at that. This is what sometimes people describe as the "low hire, low fire" environment.

I just want to stress that it's quite unusual because the outflow is becoming slower and slower. The job finding rate is slower. You have both these experiences of sending a million resumes and nothing really responds to that, but also the fact that the unemployment rate keeps inching up.

Sablik: Yeah, and we can dig a little deeper into that. I'm sticking with the metaphor because I really like the bathtub metaphor.

So, [given] the unemployed workers — the water in the tub — and thinking about long-term unemployment — that's the water sitting in the tub for longer and longer periods — you mentioned in general terms the slower draining process, slower job finding process. What are some of the reasons why long-term unemployment might rise?

Macaluso: The first thing that we are going to look at is the reason for searching. Why are people in unemployment and now looking for new jobs? Well, they might have lost their previous job, so they might have quit. We would think of those as people who separated from the previous job.

You could also be an entrant or a new entrant. You could be just out of college or just out of high school, or you could have taken a longer pause — in the data, it's going to be five years — and are now looking for a new job.

Entrants and new entrants are particularly vulnerable because they don't have a lot of specific human capital. So, in general, they tend to have slightly longer job finding rates, even more so when the labor market is somewhat depressed.

The entrants and new entrants, sort of young workers, are also linked to the elephant in the room: AI. There's been quite a bit of talk about AI displacing entry-level workers and, in general, displacing all sorts of job tasks and job titles. Another natural step that we take is thinking about AI exposure. How can we measure whether jobs are disappearing because of AI?

One last thing that we look at is a little more subtle. It's what we call the latent worker types. Think of them as a summary measure of a person's labor market experience. There are some people who have much more stable work lives and some who don't. Some who don't participate at all — they tend to make choices that keep them outside of the labor market.

Sablik: Are there specific characteristics that affect which type of worker you are?

Borovičková: In the CPS data, we have several observable characteristics of workers, including age, location, industry that they work, or education. The thing is that these observable characteristics do not seem to be picking up a lot of the variation that we have seen in the labor market experience. For example, you can take people who, in the dataset, look exactly the same to us: same age, same gender, same location, and same education. But one person tends to be moving between unemployment and employment all the time, while the other person is employed all the time.

So, we try to infer what a worker's type is based on their realized experience in the labor market. For this article, we have been using a methodology that was proposed by Ahn, Hobijn and Sahin, where we use the labor market experience in the CPS data and, based on those, classify workers into three groups.

They would call the first group as the primary types. These are the well-attached people — they don't have trouble finding jobs in case they lose one. Most of the time, we see them as employed and, if they happen to be unemployed, they transition out of unemployment very quickly back to employment.

The second group they label secondary types. These are people who experience a lot of unemployment spells and a lot of movements between employment and unemployment. This is the group of workers that accounts for most of the unemployed people. Throughout the business cycle, the movements in the job finding probability of this group are very important.

The third group that they call the tertiary group is a group of workers that is most of the time out of labor force. Sometimes they come back to employment, sometimes to unemployment. But most of the time, they are out of the labor force. We do not focus on this group of workers because in the recent episodes they do not seem to be contributing much to the changes in the unemployment rate that we have been seeing.

Sablik: Claudia, you mentioned the elephant in the room: AI, another big thing that could be affecting workers. What can you tell us about AI exposure for workers?

Macaluso: While the data sources that Katka was mentioning for labor market histories are very well established, there isn't anything as comparable for AI. So, let me tell you what we do, with the caveat that this is really where the work in progress is happening in the profession.

We want to think about AI exposure the way we thought about other technology like industrial robots or computers. There are essentially some tasks that AI can do and that workers may not need to do or may not be needed to do any longer. One way to measure that is to compare the description of jobs for every single job title in the economy to what is written into patents for AI products. Those patents have exactly the description of what AI can do. The larger the share of tasks that has overlap with AI, the more exposed we're going to say that that job is.

One thing that I want to flag is that this is all somewhat theoretical. We haven't really seen workers at work doing or not doing things.

So, the one I just described is the Webb index. There's another one which is developed in cooperation with Anthropic by Alondo and others. The tasks in jobs are classified both by an LLM and by human eyes and they come up with a set of tasks that are highly substitutable and some that are not. It's a very explicit measure of the percentage of your tasks that could be taken by a computer.

Sablik: Thinking about how this AI exposure might intersect with long-term unemployment, there have been some reports in the press that workers in industries that are more affected by AI could be having a harder time finding jobs. Did you find any evidence of that in your research?

Borovičková: Yes, we did. The most striking finding that we have is the difference in the job finding rate between the primary and secondary workers. What we have found is that the job finding probability has declined for everybody, but it has declined especially for the primary workers. This is surprising because the primary workers are those who usually do not have trouble in the labor market, but it is especially this group that experiences the elevated risk of having trouble finding a new job once they become unemployed.

Naturally, we try to think whether this is driven by AI exposure. What we did is that we classified workers within each group based on their occupations that they have and looked at people who, before becoming unemployed, were in occupations that would be very exposed to AI based on one of the measures that Claudia mentioned. We do see that within the primary workers group, the workers who are more exposed to AI are indeed experiencing a bigger drop in the job finding rate. The same is happening within the group of secondary workers. Nevertheless, it is the primary workers that are driving a lot of the decline in the job finding rate that we have seen in the data.

Sablik: Within that primary group, is the reduction in the job finding rate for industries exposed to AI enough to drive the overall increase in long-term unemployment that you're seeing in the data?

Macaluso: Well, it actually is not. The workers who are exposed to AI, especially the primary workers, have the largest draw. But it's worth recalling that there's very few of them in unemployment at any given time. It is the secondary-type workers that make up the bulk of the unemployment fluctuations. The AI-exposed workers make up, at most, about one third of the change in the job finding rate, while the other ones make up the rest. Overall, the aggregate is still a story of secondary workers, those that cycle through employment and unemployment very, very often.

Sablik: I guess a search for a culprit continues.

Katarina and Claudia, thank you so much for joining me to talk about your research.