DANE Early GDP Estimator — Colombia’s ML agricultural nowcasting model

Colombia's national statistics institute DANE developed a machine-learning nowcasting model that produces an early estimate of agricultural-sector GDP approximately 30 days before the official quarterly figure is published by the national accounts team. The initiative sits within DANE's broader 'Data for Now' programme, which also applies satellite imagery analysis, night-lights data and natural language processing to accelerate other parts of the statistical production process.

To generate a timely proxy indicator of agricultural GDP ahead of the official release, supporting early economic signalling relevant to fiscal and agricultural policy planning.

The system ingests Google News headlines, Google Trends search volumes and DANE's own administrative data series, applying Ridge regression and Zero Shot Classification (a transformer-based method) to generate the early estimate.

In its first published benchmark the model estimated agricultural GDP for Q2 2022 at 8,188 billion Colombian pesos. Of the methods tested, Ridge regression achieved the lowest mean prediction error, outperforming both the Zero Shot classifier and simpler baselines. The work was peer-reviewed and published in the SciELO academic journal Estudios de Economía and was featured by the Global Partnership for Sustainable Development Data as an example of ML-driven statistical innovation.

Formal integration of the indicator into official DANE statistical releases, and uptake by other national statistics offices, remain works in progress.