Colombian Agricultural Sector’s Early Estimator of Gross Domestic Production Using Nowcasting and Big Data Methods

Abstract

Facing challenges like the COVID-19 pandemic, statistical production increasingly relies on non-traditional data sources for timely and accurate information. In this regard, The National Statistical Office of Colombia (DANE, by its acronym in Spanish) initiated a project, supported by the Statistics Advisory Council, to develop an early estimator for the Colombian agricultural sector. This paper presents the results for the implementation of a Ridge model and Zero Shot Classification to estimate the Gross Domestic Product (GDP) of the agricultural sector, leveraging Google News and Google Trends. Results reveal that these alternative sources offer valuable insights into economic trends. Combining machine learning techniques with Nowcasting methods yielded precise projections. The Ridge method demonstrated the lowest estimation error, providing an early GDP indicator for the agricultural sector of 8,188 billion Colombian pesos for 2022 Q2, 30 days ahead of official publication.

Keywords

• Nowcasting
• Forecasting
• google trends
• Machine Learnig
• Big Data