Research
Working papers and advanced projects
When to Broaden, When to Focus: Job Search and Market Tightness
This paper investigates whether directing job search toward tight labor markets improves re-employment. Such
targeting is theoretically attractive as it reduces aggregate congestion. I link Swiss unemployment records to
clickstream search data to measure which markets jobseekers target. To address selection, I exploit
quasi-random assignment to caseworkers differing in their tendency to redirect clients toward tight markets.
Moving from the 10th to the 90th percentile raises six-month job-finding by 2.7%. The channels are
heterogeneous: jobseekers who become unemployed in slack markets gain from mobility, consistent with prior
findings from policies that encourage broader search. The new insight is that jobseekers who become unemployed
in tight markets benefit from a narrower focus on this initial market. Effects are orthogonal to other
caseworker behaviors and do not come at the expense of job quality. My findings suggest that market tightness
offers a simple basis for targeted search advice.
Media: Die Volkswirtschaft / La Vie économique (forthcoming)
Adapting to Scarcity: The Role of Firms in Occupational Transitions
This paper examines the circumstances under which firms facilitate occupational transitions, complementing
prior work that focuses on workers' decisions. We link unemployment insurance records with application diaries
and clickstream data from a recruitment platform to causally assess how candidates' occupational histories
shape recruiters' hiring decisions. We find that the average candidate from a different occupation faces a
7% lower contact rate than equally qualified candidates who last worked in a recruiter's searched
occupation. Using a new measure of skill overlap, we show that 60% of this penalty reflects that movers
meet fewer skill requirements than incumbents. Occupational experience and qualifications further reduce the
mover penalty, such that certain candidates returning to a prior occupation face no penalty at all. Finally,
recruiters adapt to scarcity and contact more movers in tight occupations. Changes in firm behavior account
for one-third of the increase in movers' application success in tight versus slack labor markets.
Media: RFBerlin Research Insights
Job Search and Employer Market Power
This paper provides a framework for thinking about how the job search of workers affects the market power of
employers. We use data from the French public employment service on the search behaviour of the unemployed
to follow job-seekers through each step of this process: a search query returns a long list of vacancies
presented with very limited information; a `click' on a vacancy in this list reveals more information,
notably the wage; and a clicked vacancy may then receive an application. Job-seekers appear to search
sequentially, applying straight after a click rather than first assembling a portfolio of possible jobs; few
of the job offers they generate are ever refused; and applications rise less than proportionally with the
number of vacancies in a market, suggestive of increasing marginal application costs. Guided by these
findings, we model the decision to apply for a job as based on whether its expected utility exceeds the
marginal cost of an application, and estimate the model working backwards: the decision to apply given a
click, then the decision to click given the long list. We show how this model can be used to compute
measures of the extent of employer market power. Because the wage is only revealed by a click, a higher wage
raises the probability of filling a vacancy primarily by attracting more applications, and the implied
markdown formula combines the wage elasticity of applications with a modified version of popularly used
measures of concentration ratios based on the `consideration set' of clicked vacancies.
Generative AI and unemployment
We study the impact of generative artificial intelligence on unemployment using detailed register data from
the Swiss unemployment insurance system including information on occupation, industry, education, and past
wages. We combine established occupation-level AI exposure measures from the literature with new measures of
firm AI usage from a survey at the industry-time level. We argue conceptually that unemployment, rather than
shifts in employment, is the relevant margin for measuring the welfare costs of AI-driven labor market
disruptions. In a difference-in-differences design around the release of ChatGPT, unemployment in highly
exposed occupations increases by approximately 20 percent relative to less exposed occupations. The results
are robust to controlling for industry-specific shocks. We decompose the rise in unemployment into changes
in the separation rate and the job-finding rate.
Can technology free time? Experimental evidence from an AI chatbot helping caseworkers solve employers' hiring difficulties
In late 2024, the French public employment service introduced a new AI tool to support caseworkers dedicated to
assisting client recruiters in their hiring decisions. Caseworkers support employers through two channels:
screening applications from job seekers who apply on their own, and sourcing candidates who have not applied.
The tool was designed to help caseworkers when sourcing candidates, and save them time for other tasks. We
leverage the randomised implementation of this intervention to measure its impact on job-filling
probabilities. We find that the intervention successfully increased the probability of filling posted jobs by
between 10 and 15 per cent. While the intervention was intended solely to facilitate the sourcing of job
seekers who had not initially applied for the job, we find that the increase in hires comes from both job
seekers who were sourced and job seekers who had applied on their own and were screened by the caseworker. Our
interpretation is that the AI tool freed up caseworkers' time, allowing them to devote more time to screening
applications (and other tasks).
Work in progress
Incidence of corporate taxation
Which Job Openings Lead to Employment? The Role of the Consideration Scope in Job Search
Policy reports
Artificial intelligence and the Swiss labor market
We analyse early impacts of large language models (LLMs) on unemployment and job postings across occupations in Switzerland.
Media: covered in all major Swiss news outlets. The following include quoted expert commentaries / interviews: NZZ; KOF-Studie: KI verändert den Schweizer Arbeitsmarkt dramatisch and NZZ; Hochgebildet, Kadermann — und plötzlich arbeitslos; KOF Magazine interview; “I expect AI to make us all richer”
The effect of unemployment benefit sanctions on search effort and re-employment outcomes
Media: Die Volkswirtschaft (DE/FR); La Liberté: Les pénalités infligées par l'assurance chômage peuvent conduire à la pauvreté
Swiss job tracker
A real-time index of open vacancy postings in Switzerland. The R code aggregates data scraped from 70 job portals and 50K company webpages; I implemented an algorithm that deals with scraper issues and unstable job portals in real-time to obtain a clean index.
Media: Launch on Swissinfo (SRG SSR); NZZaS: Studie zur zweiten Welle: 200 000 Selbständige fürchten um ihre Existenz
Teaching
- 2019-2020: TA for the Public Finance bachelor's course by Pascal St-Amour
- 2020-2024: TA for the Econometrics master's course by Martin Huber
Other
- Contributed a patch to fix an error in the did_multiplegt_dyn R package (de Chaisemartin & D'Haultfoeuille 2024)
- Together with Pascal Josephy, I founded the web design and web development agency jkweb in 2011 (at the age of 17). The company by now employs over 30 programmers, designers and other specialists in Zurich and Basel. jkweb has recently acquired novu and is now called novu.