-
You're (not) Hired: Artificial Intelligence and Early Career Hiring in the Quarterly Workforce Indicators
April 2026
Working Paper Number:
CES-26-27
Using detailed tabulations from matched employer-employee administrative data, I document evidence of an immediate, sizable, and persistent decrease in the level of early career (22-24 year old) hires following introduction of ChatGPT within the industry-state cells that are most exposed to AI. The decline in hires is the primary cause of large observed declines in employment over the subsequent period. Regressionadjusted employment of early career workers in the most AI-exposed quintile of industry-state cells declined by 12% over the 10 quarters following the introduction of ChatGPT, even as employment in lessexposed industries has remained stable. The rate of hiring largely recovered by early 2025, attributable to a smaller employment base. Earnings growth of early career workers in the most exposed industries slowed slightly relative to those in less exposed industries. Although the most AI-exposed quintile of detailed industries is dominated by a handful of industry sectors, I find that the association of higher AI exposure with reduced early career employment and fewer hires is observed across most sectors of the economy. Timing of effects in event studies is consistent with an immediate effect on hiring following introduction of ChatGPT. However, triple difference estimates provide some evidence of earlier trend shifts on employment, hiring, and separations around the onset of the COVID pandemic. I discuss potential explanations, including the increase in remote work and increased educational attainment among workers in AI-exposed occupations. Nonetheless, job gains to early career workers and backfill hires show evidence of discontinuous decline at the time of ChatGPT's release in comparison to older workers in the same industries. A local projections analysis at the NAICS industry group level shows that industries with high AI exposure are not particularly sensitive to unexpected fluctuations in monetary policy on average relative to other industries in employment, hiring, or separations. A historical decomposition suggests that up to one quarter of relative early career employment declines through 2025q2 may be attributable to monetary policy shocks through 2023, but the analysis does not find evidence that these shocks can explain the rapid decline in hires at the most AI-exposed firms in comparison to others.
View Full
Paper PDF
-
Unemployment Insurance Extensions, Labor Market Concentration, and Match Quality
April 2026
Working Paper Number:
CES-26-24
I investigate whether the effects of UI extensions are different for workers exposed to higher levels of local labor market concentration, a potential source of employer market power. I exploit measurement error in state unemployment rates that led to quasi-random assignment of UI durations in the U.S. during the Great Recession. Using matched employer-employee data from the Longitudinal Employer-Household Dynamics program, I find that UI extensions lengthen nonemployment durations by one week and cause economically meaningful but not statistically significant increases in earnings. The UI-earnings effect is significantly lower at higher levels of concentration, while there is no difference in the UI-duration effect. The lower UI-earnings effect is driven by the extremes of the distribution of concentration. My results suggest that match improvements from UI are attenuated at higher levels of concentration.
View Full
Paper PDF
-
Establishment-Level Life Cycle and Analysts' Forecasts
February 2026
Working Paper Number:
CES-26-12
This paper examines how multi-unit firms' life-cycle stages affect analyst forecast accuracy. While prior studies focus on the firm-level life cycle, we utilize the Census data and focus on the establishment level. We find that analyst forecast accuracy is lower for multi-unit firms whose establishments are in different life-cycle stages than those in the same life-cycle stage. This finding suggests that the forecasting difficulty of more diversified firms can be attributed to the different life-cycle stages of each establishment. We also find that for firms whose units are in different stages, analyst forecast accuracy is lower if the establishments in earlier stages are larger (i.e., generate more revenue) than those in later stages. As a comparison, we estimate the life-cycle stages using firms' segment classifications in their 10-K filings. We find that analysts' forecast accuracy is lower when firms report fewer segments than the number of establishments, suggesting that aggregating more establishments for segment reporting could complicate analysts' forecasting. To our knowledge, this is the first study that focuses on the establishment-level life cycle. This study highlights that firm-level life cycles should not be taken without caution, as aggregating multiple units' life cycles may be misleading. In order to provide better forecasts to investors, analysts should have a deeper understanding of firms' subunits, especially when the establishments are in different life-cycle stages.
View Full
Paper PDF
-
Expectations versus Reality in Business Formation
February 2026
Working Paper Number:
CES-26-11
Using administrative data on 17 million U.S. business applications linked to outcomes, we compare potential entrants' expectations about employer entry and first-year employment with realizations. On average, applicants overestimate employment, mainly because many expect to enter but do not. Among those who expect and achieve entry, employment is typically underestimated. Expected employment predicts entry and realized employment, but conditional on entry realized employment rises less than one-for-one with expectations. Expectation errors are highly heterogeneous and systematically related to application characteristics and local economic conditions, and they predict near-term employment outcomes. A parsimonious model with heterogeneous priors, learning, and pre-entry selection rationalizes these patterns.
View Full
Paper PDF
-
Non-Random Assignment of Individual Identifiers and Selection into Linked Data: Implications for Research
January 2026
Working Paper Number:
CES-26-06
The U.S. Census Bureau's Person Identification Validation System facilitates anonymous linkages between survey and administrative records by assigning Protected Identification Keys (PIKs) to person records. While PIK assignment is generally accurate, some person records are not successfully assigned a PIK, which can lead to sample selection bias in analyses of linked data. Using the American Community Survey (ACS) and the Current Population Survey Annual Social and Economic Supplement (CPS ASEC) between 2005 and 2022, we corroborate and extend existing findings on the drivers of PIK assignment, showing that the rate of PIK assignment varies widely across socio-demographic subgroups. Using earnings as a test case, we then show that limiting a survey sample of wage earners to person records with PIKs or successful linkages to W-2 wage records tends to overestimate self-reported wage earnings, on average, indicative of linkage-induced selection bias. In a validation exercise, we demonstrate that reweighting methods, such as inverse probability weighting or entropy balancing, can mitigate this bias.
View Full
Paper PDF
-
Manufacturing Dispersion: How Data Cleaning Choices Affect Measured Misallocation and Productivity Growth in the Annual Survey of Manufactures
September 2025
Working Paper Number:
CES-25-67
Measurement of dispersion of productivity levels and productivity growth rates across businesses is a key input for answering a variety of important economic questions, such as understanding the allocation of economic inputs across businesses and over time. While item nonresponse is a readily quantifiable issue, we show there is also misreporting by respondents in the Annual Survey of Manufactures (ASM). Aware of these measurement issues, the Census Bureau edits and imputes survey responses before tabulation and dissemination. However, edit and imputation methods that are suitable for publishing aggregate totals may not be suitable for estimating other measures from the microdata. We show that the methods used dramatically affect estimates of productivity dispersion, allocative efficiency, and aggregate productivity growth. Using a Bayesian approach for editing and imputation, we model the joint distributions of all variables needed to estimate these measures, and we quantify the degree of uncertainty in the estimates due to imputations for faulty or missing data.
View Full
Paper PDF
-
Job Tasks, Worker Skills, and Productivity
September 2025
Authors:
John Haltiwanger,
Lucia Foster,
Cheryl Grim,
Zoltan Wolf,
Cindy Cunningham,
Sabrina Wulff Pabilonia,
Jay Stewart,
Cody Tuttle,
G. Jacob Blackwood,
Matthew Dey,
Rachel Nesbit
Working Paper Number:
CES-25-63
We present new empirical evidence suggesting that we can better understand productivity dispersion across businesses by accounting for differences in how tasks, skills, and occupations are organized. This aligns with growing attention to the task content of production. We link establishment-level data from the Bureau of Labor Statistics Occupational Employment and Wage Statistics survey with productivity data from the Census Bureau's manufacturing surveys. Our analysis reveals strong relationships between establishment productivity and task, skill, and occupation inputs. These relationships are highly nonlinear and vary by industry. When we account for these patterns, we can explain a substantial share of productivity dispersion across establishments.
View Full
Paper PDF
-
Earnings Measurement Error, Nonresponse and Administrative Mismatch in the CPS
July 2025
Working Paper Number:
CES-25-48
Using the Current Population Survey Annual Social and Economic Supplement matched to Social Security Administration Detailed Earnings Records, we link observations across consecutive years to investigate a relationship between item nonresponse and measurement error in the earnings questions. Linking individuals across consecutive years allows us to observe switching from response to nonresponse and vice versa. We estimate OLS, IV, and finite mixture models that allow for various assumptions separately for men and women. We find that those who respond in both years of the survey exhibit less measurement error than those who respond in one year. Our findings suggest a trade-off between survey response and data quality that should be considered by survey designers, data collectors, and data users.
View Full
Paper PDF
-
The Rise of Industrial AI in America: Microfoundations of the Productivity J-curve(s)
April 2025
Working Paper Number:
CES-25-27
We examine the prevalence and productivity dynamics of artificial intelligence (AI) in American manufacturing. Working with the Census Bureau to collect detailed large-scale data for 2017 and 2021, we focus on AI-related technologies with industrial applications. We find causal evidence of J-curve-shaped returns, where short-term performance losses precede longer-term gains. Consistent with costly adjustment taking place within core production processes, industrial AI use increases work-in-progress inventory, investment in industrial robots, and labor shedding, while harming productivity and profitability in the short run. These losses are unevenly distributed, concentrating among older businesses while being mitigated by growth-oriented business strategies and within-firm spillovers. Dynamics, however, matter: earlier (pre-2017) adopters exhibit stronger growth over time, conditional on survival. Notably, among older establishments, abandonment of structured production-management practices accounts for roughly one-third of these losses, revealing a specific channel through which intangible factors shape AI's impact. Taken together, these results provide novel evidence on the microfoundations of technology J-curves, identifying mechanisms and illuminating how and why they differ across firm types. These findings extend our understanding of modern General Purpose Technologies, explaining why their economic impact'exemplified here by AI'may initially disappoint, particularly in contexts dominated by older, established firms.
View Full
Paper PDF
-
The Geography of Inventors and Local Knowledge Spillovers in R&D
October 2024
Working Paper Number:
CES-24-59
I causally estimate local knowledge spillovers in R&D and quantify their importance when implementing R&D policies. Using a new administrative panel on German inventors, I estimate these spillovers by isolating quasi-exogenous variation from the arrival of East German inventors across West Germany after the Reunification of Germany in 1990. Increasing the number of inventors by 1% increases inventor productivity by 0.4%. I build a spatial model of innovation, and show that these spillovers are crucial when reducing migration costs for inventors or implementing R&D subsidies to promote economic activity.
View Full
Paper PDF
