Leveraging alternative data sources for socio-economic nowcasting

Abstract

This article uses the French case to argue that alternative datasets can be used to improve the ability of policymakers to monitor socio-economic phenomena in real-time and at specific locations. In the first part of the article, we present an inflation nowcast based on internet data, both at the French level and for the city of Provins, the birthplace of the Yellow Vest movement. We then present a job-openings nowcast based on job postings online. Finally, we present a machine learning model used to predict soft wheat harvests in France in real-time. This indicator can anticipate wheat production levels and, therefore, food prices. In the sequence, we present an analysis of the French presidential election based on social media analytics. First, we employ NLP techniques to extract the hottest political topics from a pool of social media posts. Second, we use a transformers-based deep learning model to create a sentiment analysis of Emmanuel Macron and Marine Le Pen in the run-up to the ballot in order to estimate their vote intentions in real-time.