Testing big data in a big crisis: Nowcasting under COVID-19.

Area of Science:

• Economics
• Econometrics
• Data Science

Background:

• The COVID-19 pandemic disrupted standard economic forecasting models.
• Traditional macroeconomic variables proved insufficient for real-time economic predictions.

Purpose of the Study:

• To present novel methods for macroeconomic forecasting in unconventional times.
• To enhance the accuracy of economic predictions using big data and advanced statistical frameworks.

Main Methods:

• Construction of an extensive European dataset with over a thousand time series, including big data indicators.
• Development of a dynamic Bayesian framework to merge diverse data sources with multiple forecasting methods.
• Introduction of a 'selection prior' for model selection among competing forecasting approaches.

Main Results:

• Demonstrated the added value of big data indicators for nowcasting Gross Domestic Product (GDP).
• Identified key variables effective for predicting GDP during the COVID-19 crisis.
• Validated the efficacy of the dynamic Bayesian framework in a crisis scenario.

Conclusions:

• The proposed methodology offers a robust solution for economic forecasting in volatile periods.
• Forecasting institutions can adopt these methods to improve predictions during future crises.
• Integration of unconventional data sources is crucial for accurate real-time economic assessment.