Understanding Unemployment in the Era of Big Data : Policy Informed by Data-Driven Theory

On one hand, unemployment is a central issue in all countries. On the other the economic policies designed to mitigate it are usually built on theoretical grounds that are validated at an aggregate level, but have little or no validity from a micro point of view. This situation is a cause for concern because policies are designed and implemented at the level of individuals and organisations, so ignoring realistic micro-mechanisms may lead to costly outcomes in the real world. Ironically, the data to inform theoretical frameworks at the micro-level has existed in labour studies since the 1980's. However, it is only now that we count with analytical methods and computational tools to take full advantage of it. In this paper we argue that big data from administrative records, in conjunction with network science and agent-computing models offer new opportunities to inform unemployment theories and improve policies. We introduce a data-driven model of unemployment dynamics and compare its predictions against a conventional theory built on assumptions that are common among policy models. We show that these assumptions, while reasonable at a first glance, lead to erroneous predictions that have real-world consequences