To identify true sources of employment and income risk, ask workers

How much risk do people face in the labor market? What causes this variation in earnings and employment? Few questions are more fundamental to economic policy, including the Fed’s mandate to pursue maximum employment.

Yet efforts to answer these questions through economic models of earnings and employment face a problem: Most labor market data come with important limitations. “When you look at the outcomes for people, you only ever observe [their earnings] if they are working, and only in a certain job,” said Institute consultant and University of Minnesota economist Mariacristina De Nardi. “You only see the options they were offered and that they decided to take.”

This observed outcome could be only a partial representation of a worker’s actual skills and potential, which would be more valuable data to economists. “Just looking at the realizations of income, it is very hard to disentangle all of the possible forces and mechanisms” behind those outcomes, said Yale economist Orazio Attanasio, a 2024 Institute visiting scholar. “You have to make a very strong set of statistical assumptions to be able to extract from those data the objects you are interested in,” such as unemployment risk and the extent to which earnings growth is sustained over time.

In a recent Institute working paper, Attanasio, De Nardi, and co-authors illustrate one solution: Try a different type of data. Sufficiently detailed household surveys can elicit not just current labor market status but each respondent’s expectations and preferences about job offers and earnings. This allows the researchers to observe beliefs about opportunities that do not depend on whether those opportunities are ultimately accepted.

In “Subjective Earnings and Employment Dynamics,” the economists leverage the New York Fed’s panel-style Survey of Consumer Expectations (SCE) to reveal a new and arguably more accurate portrait of the risk and heterogeneity that shape outcomes for American workers. Compared with prior findings based on observational data, their survey-based findings point to a much smaller role for “productivity shocks” in shaping workers’ outcomes. Productivity shocks are bolts-from-the-blue that can knock a worker off course or provide an unexpected boost. Examples might be new technologies, health events, or the birth of a child, said De Nardi, “everything that makes you less or more productive at work with the same employer, and is not a permanent change.” Given the challenge of identifying other factors from observational data, prior research lumped much labor market risk under this broad heading. Even in studies that allow for changing employers, the results indicate that such generic productivity shocks play a big role.

In the SCE data, however, survey respondents express a relatively low expectation that their status quo will be upended by such events. “These productivity shocks, this catch-all, are much less persistent than in a big chunk of the literature,” Attanasio said. De-emphasizing the role of these exogenous events, he said, “makes a big difference for the way you represent reality” in economic models.

Worker, know thyself

Instead of productivity shocks, the SCE data identify more significant roles for two other factors in generating variation in worker outcomes.

The first is simply the extent to which people differ in skill and ability. In the SCE survey, people report their own earning potential (see sidebar). This might differ a lot from their pay at the job they happen to hold at any moment. Through this lens, said Attanasio, “diversity across people is substantially larger than what’s been estimated before.”

Should we trust people’s own perceptions of their earnings and job offer prospects? Yes, said De Nardi—in fact, the lack of this “latent” expression of labor market outcomes is precisely the problem with observational data. “What people think is how they act,” said De Nardi. “Your consumption choices, your savings choices, your job search choices are based on beliefs.” In simulations, the economists find that these beliefs lead to choices that align with subsequent labor market outcomes.

The second factor, which is a major source of risk for workers, will resonate with anyone who has struggled with a poor fit during their career: the quality of the job match. “I might be very good at working with a certain firm,” said Attanasio, “but not that good at working with another one or producing something a little bit different.” The economists find that other than entering unemployment, job changes are the biggest driver of individual earnings variability. Carefully worded survey questions in the SCE illuminate workers’ expectations for the impact of changing employers. This can mean a boost in productivity and earnings if the job match is a great one. But a switch can also take workers sideways or backward.

Why would job changes arise as a more important component of individual risk in this study compared with previous research? Unlike realized labor market data, said Attanasio, surveys elicit responses to hypothetical scenarios. For example, all respondents, employed or not, are asked whether they would accept a hypothetical offer and change their labor market status or current employer. “When working with realized data, one only observes outcomes of job changers, with no information on offers that have been rejected,” he said. “This poses difficult selection issues in the econometric study of these models.”

Different risks suggest different directions for policy

If better data shift our understanding of labor market risks and worker diversity, this has implications for policies intended to support employment and life cycle stability. Whereas prior research reinforced the value of buffering workers against surprise productivity shocks, these new findings highlight instead the heterogeneity across individuals and the importance of a well-matched job.

How might public policy address the heavy role the economists find for individual differences? The wide variation in earnings and income volatility “does not mean we can do nothing about it,” De Nardi said. On the contrary, “it depends on education, it depends on how we develop social-emotional skills. It depends on the family background one grows up with.” The findings complement the enormous body of research showing the long-term value of childhood interventions that build skills and economic resilience.

The substantial risk that the economists identify from better or worse job matches suggests a policy focus on “the fluidity of the labor market,” said De Nardi. “The more labor market frictions you have, the harder it is for people to find the right match.” For example, if workers are tied down by the fear of losing benefits, intimidated by high job-search costs, or otherwise wary of the downside of seeking a better role, that can raise the level of risk in the labor market overall.

“If you can improve that dimension—making people potentially more mobile, making it easier to find better matches for their skills—that could be useful,” said Attanasio.

Running some modern “heterogeneous agent” economic models can require heavy computing resources. In contrast, the economists highlight the relatively simple, linear nature of their model, where the rich survey data obviates the need for complex mathematical methods. “Questions that consider counterfactual, hypothetical scenarios simplify things considerably,” said Attanasio. As the research evolves, the economists contemplate future steps to make their subjective data even more useful, including gathering respondents’ expectations at multiple time horizons and pairing survey responses with detailed, life-long work history.

Their research into employment risk will also evolve alongside a potentially generational shock to the labor market. Whatever the ultimate scale of change, artificial intelligence promises to disrupt productivity, tasks, job search, compensation, unemployment, and just about any aspect of the labor market one might think to model. “There is no question that AI is a structural change,” said De Nardi.

No matter how AI ultimately figures into economists’ labor market models, the findings of Attanasio, De Nardi, and their co-authors illustrate how the direct perspectives of workers can be essential data to measure the risks those workers face.