Generative AI Impact on Germany's Labor Market: An Occupational-Level Econometric Study

This study uses German occupational-level panel data from 2015 to 2025, applying difference-in-differences and event study methods to empirically analyze the impact of generative AI exposure on labor market outcomes across different occupations. Key findings: Occupations with high AI exposure showed higher unemployment-vacancy ratios (high demand-side pressure) and longer vacancy durations (skill mismatch) in the post-ChatGPT period; the study provides an empirical basis for policy formulation (such as skill retraining and social security adjustments), and methodologically adopts fine-grained occupational-level analysis and an open-source, replicable design.

Since the release of ChatGPT at the end of 2022, the penetration of generative AI into cognitive and knowledge-based work has attracted attention from academia and policymakers. However, there is a lack of large-scale rigorous empirical evidence on the actual impact of AI on employment. This study uses German occupational-level panel data to quantify the causal relationship between AI exposure and labor market indicators, filling this gap.

Data Sources: German KldB (German Classification of Occupations) 3-digit coded occupational panel data from 2015 to 2025, with the key time node being the release of ChatGPT at the end of 2022; core datasets include panel_dataset_controls.csv (control variables such as employment and wages) and AIOE_DataAppendix.xlsx (AI exposure scores). Core Variables: log_relation (log of the ratio of unemployed persons to job vacancies), log_vakanz (log of average duration of job vacancies in days), log_bestand (log of employment size), log_median_wage (log of median wage), ai_exposure (occupational AI exposure score), post (dummy variable for years 2023 and later). Econometric Methods: Baseline cross-sectional regression, continuous treatment difference-in-differences (DiD) model (controlling for occupational fixed effects and year fixed effects), event study method (dynamically estimating the trajectory of AI impact and parallel trend test).

Key Results: Occupations with high AI exposure had higher unemployment-vacancy ratios (high demand-side pressure) and longer vacancy durations (skill mismatch) in the post-ChatGPT period. Robustness Tests: Binary treatment DiD (high/low AI exposure groups), exclusion of the 2020-2021 pandemic interference period, and defining 2023 and later as the treatment period all support the core conclusions.

Policy implications of the study: 1. Increase support for skill retraining for affected occupational groups to facilitate transition to AI-complementary skills; 2. Adjust unemployment insurance and re-employment services to address structural unemployment needs; 3. Pay attention to regional labor market heterogeneity and formulate differentiated policies.

Methodological Contributions: Fine-grained occupational-level analysis (more accurate identification of AI impact transmission mechanisms), DiD design using ChatGPT's release as an exogenous shock, and replicable open-source code (Python/pyfixest library). Limitations: Covers only Germany (conclusions should be extrapolated with caution), AI exposure scores are based on U.S. occupational data (cross-cultural applicability needs verification), and short-term effects cannot predict long-term equilibrium outcomes. Future Directions: Expand to more countries and longer time spans, and explore the relationship between AI exposure and dimensions such as wages and job quality.