Job Market Paper
Occupational Reinvention: Evidence from a century of task & technology
data
Abstract: Technological change reshapes occupations by both substituting for
and
assisting workers, yet standard measures hold task content fixed and collapse these forces into
a
single index. We assemble an 80-year, task-level panel for U.S. occupations by digitising the
historical Dictionary of Occupational Titles with a computer vision model and harmonising it
with
modern O*NET data. Using text embeddings, we map detailed task descriptions to patent texts to
construct time-varying, occupation-level exposure to substitution- and assistance-intensive
innovation. To avoid mechanical endogeneity from task reweighting, exposure is anchored in a
predetermined “frozen” baseline of tasks and task weights. We combine these baseline
exposures with
differential growth across technology fields in a shift-share IV design to identify the causal
effects of substitution- and assistance-intensive innovation on occupation-level employment and
wages. The results imply that measuring technology without accounting for evolving task content
systematically distorts our understanding of both the history and the future of work.
Work in Progress
Old Skills, New Skills
Abstract: Combining over one million Adzuna online job advertisements per year from September 2016 and September 2024, matched to the Lightcast Open Skills taxonomy, we examine how UK employers' skill requirements have evolved. Following Deming and Noray (2020), we classify skills within each occupation as emerging, disappearing, or core. Around 14% have emerged or sharply declined, with the largest churn in IT, health, and Science, Engineering and Technology roles: AI, cybersecurity, and machine learning have risen, while older digital and routine administrative skills have fallen. Yet the biggest shifts come from changes in core skills - communication, management, and client services remain the most widely sought, and increasingly so. In 2024, 36% of vacancies mentioned an IT skill and 21% an analysis skill, showing the diffusion of technology-related competencies well beyond traditionally technical roles.
AI Adoption and Workforce Composition
Abstract: This paper examines how AI adoption reshapes workforce composition at the firm level, focusing on two underexplored margins: the substitutability or complementarity of AI and migrant labour, and distributional effects across ethnic groups. Using UK firm-level panel data linking LinkedIn-derived worker histories, AI adoption indicators, and firm financials, we exploit Brexit-induced variation in EU labour supply as an instrument to identify causal effects. Firms with greater pre-referendum reliance on EU workers faced differential post-Brexit labour supply shocks, providing exogenous variation in both migrant labour availability and AI adoption incentives. We complement the main analysis with an exploratory examination of GitHub activity among AI workers to better characterise how AI roles function in practice.
Indirect Experience
Abstract: We study the effect of an unprecedented shock on the behavior and beliefs of individuals socially connected to those directly exposed. Using an original survey that we designed and administered to migrants living in the United States, we show that individuals from countries hit earlier by Covid-19 adopted preventive measures before U.S. mandates and cited events in their home countries as the main driver of this response. Over time, differences in legally mandated behaviors converged, but differences in non-mandated precautionary behaviors persisted. Respondents from earlier-hit countries also overestimated Covid-19 deaths in the U.S., consistent with heightened perceived risk. These findings show that exposure to novel shocks through social ties - which we define as indirect experience - can shape beliefs and generate persistent behavioral responses.
Working Papers
Occupational Skill Content and Technological Change
Abstract: Technological change events fundamentally change the type of tasks
performed by human labour within occupations. We develop a predictive model, utilising machine
learning techniques, and find that occupational skill intensity data can predict, to a high
degree of accuracy, technological change event exposure, as measured by indices developed by
Webb
(2020). We link these predictions to skills data from a library of newspaper job vacancy adverts
to
understand how skill intensities have changed over time, and use this to predict historical
occupational technological exposure. Change in occupational technological exposure, as predicted
by changing skill intensities, is highly associated with important labour market outcomes.
Uncertain Health and
Wealth Inequality
Best Dissertation Prize, MSc Economics, UCL (2017)
Abstract: Precautionary saving is a key driver of wealth inequality within
models
of the Bewley-Huggett-Aiyagari canon. However, models with savings rates calibrated solely to
idiosyncratic income risk find it difficult to replicate the vast wealth inequality empirically
observed in the United States. This paper looks at a potential source of increased precautionary
savings — idiosyncratic medical expenses shocks. This paper: i. establishes an
identification
procedure for medical expenditure shocks across the entire life cycle, ii. finds that
idiosyncratic
shocks are very highly persistent, iii. establishes the extent to which these shocks contribute
to
wealth inequality through the effect on savings behaviour.
Public Opinion and Immigration
Abstract: Immigration in many high-income countries is often a fraught
political
issue. The share of migrants has been rising yet typically more people want lower than higher
immigration, though views on the issue are often very polarised with strongly-held views on both
sides. Given this, understanding public opinion on immigration is obviously important and there
is a large and growing academic literature on the subject. These studies often investigate how
attitudes vary with demographics among people in the same country at the same point in time. But
it is
also likely that attitudes respond to country-level macro variables like the level and mix of
immigration and the general state of the economy. To investigate the influence of macro-level
variables
requires data on multiple countries and years so that there is enough variation in the variables
of
interest. This data is relatively rare. To address this, we introduce a novel high-dimensional
dataset,
harmonising data on the 28 EU countries from Eurobarometer surveys over the period 2002-2019,
and
investigate the influences of macro-level variables on attitudes towards immigration.
