bigNOMICS · Big Data for Forecasting Economic Developments

bigNOMICS takes inspiration from the explosive growth in the amount of information collected leading to the big data era. Such a data avalanche has created new opportunities to design novel data sets that allow pressing economic policy issues to be answered. The goal of bigNOMICS is to improve our understanding of the economic and financial environment along two directions: better data and better models. First, big data makes it possible to build a set of fast-moving measures that provide early signals of the direction of the economy. However, big data in itself does not immediately result in better economics insights, if the statistical methods adopted do not account for the complexity of such data. Hence, a second goal of bigNOMICS is to adopt complex statistical models from the machine learning and artificial intelligence literature to extract relations and patterns to turn data into policy insights.

The project focuses on four important lines of research:

Exploiting economic news to improve nowcasting and forecasting economic and financial indicators

News often contains unanticipated information about the health of the economy and financial markets and their evolution over time: such as the progress of stocks and consumption, the future changes in fiscal regulations. This enables market participants to learn about recent economic events and trends, helping them to adjust their expectations about future developments of the economy and financial markets. The project collects economic news from US and European news outlets. A set of natural language processing techniques is applied to calculate sentiment indicators as well as extract emotions, such as anxiety, panic, confidence and enthusiasm, expressed by the text. These indicators are then used in traditional economic models as well as within more complex machine learning models to improve, nowcast and forecast of economic and financial indicators.

The use of loan-level data to track regional variations in household debt and default

The project analyses a dataset of millions of loans across Europe to ex­tract proxies for the level of financial indebtedness of households in Europe, and study its evolution during the 2008 financial crisis. Understanding the driv­ers of the recent downturn is of par­amount importance for policy makers in order to learn how to avoid socially and economically stressful events as well as decide how to optimally react should these events happen again. A further range of models from the machine learning literature is adopted to study the drivers of loan default. Many factors might influence simultaneously the decision of a borrower to default, but only few of these factors can be tuned by policy makers. A better under­standing of the drivers of loan default could help policy makers to identify in­terventions to reduce delinquency cas­es, thus containing costs caused by the inefficient allocation of resources.

The trade effects of European anti-dumping

Despite the growth in international trade due to the increased integration of national economies into a global economic system, as well as advance­ments in telecommunications and logistics, the project observes persis­tent and even the intensifying adoption of trade protection measures, such as anti-dumping. Anti-dumping measures are often adopted by governments as a tool to protect domestic firms and industries. A large data set is exploit­ed on import and exports from/to the EU at a very fine level of product de­tail to study the impact of anti-dump­ing measures on trade flows. Machine learning techniques are applied to iden­tify early signs of duty avoidance due to the implementation of anti-dumping measures.

Seismonomics: Measuring economy activity through seismic noise

Background seismic noise refers to the persistent vibration of the ground due to a multitude of causes, from wind and other atmospheric phenom­ena, ocean waves as well as human activities, such as traffic and industri­al activities. Accordingly, background seismic noise is extracted to monitor human activity and use it to improve existing nowcasting and forecasting models for a set of economic indicators.