This European study presents an innovative approach to short-term forecasts of unemployment using data on Google search words, which allows them to predict unemployment trends as early as one month before the official figures are released by the Federal Employment Agency.

In times of crisis, there is a particularly strong demand for early and reliable forecasts of economic and labor market trends. For lack of up-to-date primary data, and due to rapid structural changes, traditional forecasting techniques have reached their limits in terms of providing a complete description of reality. Substantial legislative interventions, such as Germany's expansion of short-time work, create additional problems for the established forecasting models.

This study presents an innovative approach to short-term forecasts of unemployment using data on Google search words, which allows them to predict unemployment trends as early as one month before the official figures are released by the Federal Employment Agency. As the first practical applications of the model show, the predictions are remarkably accurate: The trend reversal of the past months was predicted fairly well. According to the model, unemployment figures in Germany will continue to drop in June.

"Although the internet is a vast source of instantly accessible data, which responds quickly to changes in economic and political conditions, it has been widely neglected by science. Using these data for unemployment forecasts is a promising approach, given that more than 86 percent of all job-seekers leave some kind of trace on the internet as they search for jobs online. Of course, our forecasting technique based on Google data cannot replace traditional models that analyze causal relationships. But it provides fast and reliable information that can serve as an early warning system for policymakers," said Zimmermann.

The study was published in Applied Economics Quarterly but is available to download via IZA