Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Large Language Models and Sectoral Employment in Sweden: A Multi-Cutoff Difference-in-Differences Analysis

Wolff Perez, Leo LU and Torstensson, Teo LU (2025) NEKH02 20242
Department of Economics
Abstract
This thesis investigates whether recent Large Language Models (LLMs)—notably
ChatGPT—have influenced sectoral employment in Sweden. Using a multi-cutoff
Difference-in-Differences (DiD) design centered on the dates November 2022 (initial
ChatGPT release), September 2023, March 2024, we compare employment changes in high
vs. low-exposure sectors classified via exposure metrics from Eloundou et al. (2023).
Analysis of monthly panel data (2021–2024) finds no statistically significant divergence in
employment trends between high- and low-exposure sectors post November 2022. While
point estimates suggest mild positive effects for high-exposure industries, these lack robust
significance but may reflect employer expectations that... (More)
This thesis investigates whether recent Large Language Models (LLMs)—notably
ChatGPT—have influenced sectoral employment in Sweden. Using a multi-cutoff
Difference-in-Differences (DiD) design centered on the dates November 2022 (initial
ChatGPT release), September 2023, March 2024, we compare employment changes in high
vs. low-exposure sectors classified via exposure metrics from Eloundou et al. (2023).
Analysis of monthly panel data (2021–2024) finds no statistically significant divergence in
employment trends between high- and low-exposure sectors post November 2022. While
point estimates suggest mild positive effects for high-exposure industries, these lack robust
significance but may reflect employer expectations that LLMs will augment, rather than
entirely replace, human labor. Sweden’s strong labor institutions and gradual tech adoption
may temper large-scale disruption, underscoring the need for longer-term monitoring. This
study contributes a Swedish-specific assessment of LLMs’ labor impact, employs multiple
DiD cutoffs to capture staggered AI adoption, and provides policy insights on how
institutional features can mediate emerging AI technologies. (Less)
Please use this url to cite or link to this publication:
author
Wolff Perez, Leo LU and Torstensson, Teo LU
supervisor
organization
course
NEKH02 20242
year
type
M2 - Bachelor Degree
subject
keywords
Employment, LLM, AI, DiD
language
English
id
9182614
date added to LUP
2025-05-16 10:48:01
date last changed
2025-05-16 10:48:01
@misc{9182614,
  abstract     = {{This thesis investigates whether recent Large Language Models (LLMs)—notably
 ChatGPT—have influenced sectoral employment in Sweden. Using a multi-cutoff
 Difference-in-Differences (DiD) design centered on the dates November 2022 (initial
 ChatGPT release), September 2023, March 2024, we compare employment changes in high
vs. low-exposure sectors classified via exposure metrics from Eloundou et al. (2023).
 Analysis of monthly panel data (2021–2024) finds no statistically significant divergence in
 employment trends between high- and low-exposure sectors post November 2022. While
 point estimates suggest mild positive effects for high-exposure industries, these lack robust
 significance but may reflect employer expectations that LLMs will augment, rather than
 entirely replace, human labor. Sweden’s strong labor institutions and gradual tech adoption
 may temper large-scale disruption, underscoring the need for longer-term monitoring. This
 study contributes a Swedish-specific assessment of LLMs’ labor impact, employs multiple
 DiD cutoffs to capture staggered AI adoption, and provides policy insights on how
 institutional features can mediate emerging AI technologies.}},
  author       = {{Wolff Perez, Leo and Torstensson, Teo}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Large Language Models and Sectoral Employment in Sweden: A Multi-Cutoff Difference-in-Differences Analysis}},
  year         = {{2025}},
}