Skip to content

I will talk a bit on how we use machine learning in general in the area of labour market policy in DK, and how we relate this to our core business of producing results on employment and education.
As a specific example of our work, I will illustrate our statistical profiling of newly unemployed, both the technical/methodological side as well as the practical implementation and general experiences in this area, and some thoughts on further development.
Finally I will talk a bit on other more recent areas of developing datadriven solutions in the field of labour market policy, drawing perspectives to new possibilities deriving from machine-learning and modern Technology.

The presentation is about the nature and how to clean errors in occupational coding in order to measure patterns of occupational mobility (US, UK and Canada). Furthermore it is shed light on how occupational mobility matters for cyclical earnings inequality (based on Carrillo-Tudela, Visschers and Wiczer, 2019), unemployment and its duration distribution (based on Carrillo-Tudela and Visschers, 2019) and cleansing and sullying effects of the business cycle (based on Carrillo-Tudela, Sumerfield and Visschers, 2019).

Using a quasi maximum likelihood approach for a semi-structural model, we find highly precise and distinct estimates of consumption responses to idiosyncratic income shocks for different groups of households. Homeowners stratified by liquid wealth exhibit the most dispersion in their marginal propensities to consume. Time-varying estimates support strong patterns of heterogeneity by homeownership status and balance sheet liquidity, with economically and statistically significant increases in the sensitivity of transitory consumption for homeowners, especially those with lower liquid wealth, following the collapse in house prices with the Great Recession. These findings support consumption theories that include housing as an illiquid asset.