Endogenous labour flow networks

Despite our success in demonstrating the importance of labour flow networks, I always thought that real policy relevance would come from accounting for worker-level behaviour and heterogeneity; something quite challenging with the plain formalism of random walks on graphs. In this work, we crack down this challenge by implementing an agent-computing model with household micro data and leisure-consumption behaviour. Furthermore, it is the first model that can generate empirical labour flow networks endogenously (instead of assuming them as exogenous). Real-world impact came shortly after completing this work, as departments in the UK government have adopted this toolkit.

In the last decade, the study of labour dynamics has led to the introduction of labour flow networks (LFNs) as a way to conceptualise job-to-job transitions, and to the development of mathematical models to explore the dynamics of these networked flows. To date, LFN models have relied upon an assumption of static network structure. However, as recent events (increasing automation in the workplace, the COVID-19 pandemic, a surge in the demand for programming skills, etc.) have shown, we are experiencing drastic shifts to the job landscape that are altering the ways individuals navigate the labour market. Here we develop a novel model that emerges LFNs from agent-level behaviour, removing the necessity of assuming that future job-to-job flows will be along the same paths where they have been historically observed. This model, informed by microdata for the United Kingdom, generates empirical LFNs with a high level of accuracy. We use the model to explore how shocks impacting the underlying distributions of jobs and wages alter the topology of the LFN. This framework represents a crucial step towards the development of models that can answer questions about the future of work in an ever-changing world.