In this seminar, I will discuss three recent Bank of England research projects which use large, ‘naturally occurring’ datasets of high frequency data to better understand UK macro-economic issues. Each showcases a different way of making text count in a macroeconomic application.
The first project uses 15 million job vacancy adverts posted online between 2008 and 2016. Using text analysis, we map vacancies into the standard ONS classifications to match them with other official labour force statistics. The combined dataset enables us to find whether unwinding occupational or regional 'mismatch' between workers and job vacancies would have boosted productivity and output growth in the UK’s unprecedented post-crisis 'productivity puzzle' period.
In a second application using the same data, we use a number of supervised and unsupervised machine learning techniques to derive a data-driven classification for jobs which cuts across wage, sector, region, and occupation. This classification surfaces careers not apparent in current taxonomies of jobs and could help to capture rapidly changing labour market conditions.
In the third research project, we ask whether the text of popular UK daily newspapers contains signals about the future direction of macroeconomic and financial variables. We run a horse race of different methods to turn text into time series representing both sentiment and uncertainty.