Publications
Training time, robots and technological unemployment with Alessio Moro and Michelle Rendall.
Journal of Economic Behavior and Organization , August 2026
Special issue Socio-Economic Dynamics and Health-Environmental Challenges
Working Paper | CEPR DP19343
We document that labor training requirements for high-skilled occupations increased in the U.S. from 2006 to 2019. These greater training requirements reduce the extent to which workers displaced from shrinking occupations can relocate to expanding (high-skilled) occupations, thus potentially affecting both the equilibrium occupational structure and the unemployment level. We build a quantitative model in which labor is displaced by task-replacing technological change embodied in robots (“tasks shock”) and the extent of occupational switching depends on the destination occupations' training requirements. We find that: (i) task-displacing technological change increases steady-state unemployment, but it reduces unemployment along the transition; (ii) in contrast, a comparable shock to capital embodied technological change produces larger unemployment rates with respect to the tasks shock, both in the transition and the steady state; and (iii) greater training requirements in high-skilled occupations increase steady-state unemployment and affect the occupational structure along the transition, but the shape of their effect depends on the size of the technological shock.
Working papers
Does Risk Shape Economies? Income Volatility and Structural Change with Alessio Moro and Andrea Mottola
Countries at similar levels of income can differ markedly in their sectoral composition. We show that income risk can account for part of these differences. In a multi-sector model with non-homothetic preferences, higher sectoral TFP volatility reduces the services share at a given level of GDP through a precautionary saving mechanism. Using U.S. time-series and household-level data, we find a negative and statistically significant relationship between several proxies for income risk and the services share. This relationship also holds for Latin American and Asian premature deindustrializers. We then build and calibrate a dynamic stochastic growth model of structural transformation and run a counterfactual in which we set early U.S. aggregate volatility to Great Moderation low levels. The counterfactual economy can explain about 17% of the services-share gap between the U.S. in 1879 and premature deindustrializers at comparable income levels.
The Skill Inside the Task: How AI and Robotics Reshape the Structure of Work
We examine how exposure to artificial intelligence (AI) and robotics reshapes the skill composition of occupations. Using O*NET data from 2006 to 2019, we construct indicators tracking over time the importance of seven broad skill categories within each occupation. We link these indicators to task-based measures of technological exposure at the occupational level. We then focus on the effect of AI and robotics in affecting the skill composition of high-, middle- and low-skilled groups of occupations. We find that AI primarily affects high-skill occupations by increasing the importance of Technical and Resource skill and decreasing that of System and Social skills. Robotics instead boosts Technical skills in middle and low-skill jobs and reduces Process skills in low-skilled occupations. Instead, neither AI nor robots affect the importance of Complex Problem skills.
A Theory of Structural Change, Home Production and Leisure (with Alessio Moro, Francisco Javier Rodríguez Román, Silvio Tunis).
Why do agents consume more services relative to goods as income grows? We present a theory of structural change assuming that a representative household satisfies final needs by means of two home-production functions in time and either goods or services from the market. When calibrating the model to U.S. data, roughly half of structural change is accounted for by technological change allowing services to display a larger time saving than goods in satisfying final needs. Also, even if preferences are homothetic, the calibrated model generates endogenous income effects, which account for the remaining structural change generated by the model.