In Search of a Job: Forecasting Employment Growth in the US Using Google Trends

Abstract

We show that Google search activity on relevant terms is a strong out-of-sample predictor of future employment growth in the US and that it greatly outperforms benchmark predictive models based on macroeconomic, financial, and sentiment variables. Using a subset of ten keywords, we construct a panel with 211 variables using Google's own algorithms to find related search queries. We use Elastic Net variable selection in combination with Partial Least Squares to extract the most important information from a large set of search terms. Our forecasting model, which can be constructed in real time and is free from revisions, delivers an out-of-sample R 2 statistic of 65% to 88% for horizons between one month and one year ahead over the period 2008-2017, which compares to between roughly 30% and 60% for the benchmark models.