Forecasting labor needs for digitalization: A bi-partite graph machine learning approach

Abstract

We use a unique database of digital, and cybersecurity hires from Swiss organizations and develop a method based on a temporal bi-partite network, which combines local and global indices through a Support Vector Machine. We predict the appearance and disappearance of job openings from one to six months horizons. We show that global indices yield the highest predictive power, although the local network does contribute to long-term forecasts. At the one-month horizon, the “area under the curve” and the “average precision” are 0.984 and 0.905, respectively. At the six-month horizon, they reach 0.864 and 0.543, respectively. Our study highlights the link between the skilled workforce and the digital revolution and the policy implications regarding intellectual property and technology forecasting.