Automation, Spatial Sorting, and Job Polarization

We present evidence showing that more expensive cities - measured by rental costs - have not only invested proportionately more in automation (measured by investment in Enterprise Resource Planning software) but also have seen a higher decrease in the share of routine abstract jobs (clerical workers and low-level white collar workers). We propose an equilibrium model of location choice by heterogeneously skilled workers where each location is a small open economy in the market for computers and software. We show that if computers are substitutes to middle skill workers - commonly known as the automation hypothesis - in equilibrium large and expensive cities invest more in computers and software, substituting middle skill workers with computers. Intuitively, in expensive cities, the relative benefit of substituting computers for routine abstract workers is higher, since workers must be compensated for the high local housing prices. Moreover, if the curvature of the production function is the same across skills, the model also delivers the thick tails in large cities' skill distributions presented by Eeckhout, Pinheiro, and Schmidheiny (2014).