Beyond Econometrics: Using Google Trends and Social Media Data to Forecast Unemployment - OECD analysis of accuracy gains and robustness of predictions

Abstract

Google Trends has been used for less than two decades in academia to forecast outcomes, using various techniques. While most research has focused on developed countries, there are clear information gaps that have not been fully addressed. Previous studies in this field indicate that non-linear algorithms with feature set selection while using a large set of queries can yield better results across more countries. However, it is unlikely that these methods will be widely and rapidly adopted given the skills required. Therefore, the objective of this research is to explore whether the abundance of digital data sources, specifically Google searches, can aid agents as institutions and policy makers in their modeling efforts. The aim is to fill the gap in analysis for less influential countries and explore whether the use of Google searches data can be extended to multiple countries using a simple and agile methodology based on a widely used statistics-based modeling approach (ARIMAX). For this use we selected unemployment rate as the variable of interest. However, our findings show that only 30% of countries had promising results using Google-augmented ARIMAs. Thus, more computationally intensive empirical strategies would be needed to extract more predictive power out of Google queries information pool for unemployment rate modelling.