Nowcasting the Economic Impact of COVID-19

This information was sourced from "Artificial Intelligence in Action: Selected ADB Initiatives in Asia and the Pacific," Asian Development Bank, 2024, https://www.adb.org/sites/default/files/publication/963831/artificial-intelligence-action-asia-pacific.pdf. Licensed under Creative Commons Attribution 3.0 IGO (CC BY 3.0 IGO).

The COVID-19 pandemic has become one of the biggest global health crises, but its impact goes beyond health. Experts project that the poorest countries will feel the impact of the pandemic for years, while ADB warned in 2020 that the global economy could suffer up to $8.8 trillion in losses because of the pandemic. Asia and the Pacific account for about 30% of the overall decline in global output, which is equivalent to economic losses worth $2.5 trillion.

Policymakers and businesses need timely data to base decisions that would support economic recovery. Official statistics are released on a monthly, quarterly, or yearly basis. However, some developing countries experience a significant lag in releasing this data; it can take anywhere from a few weeks to several months for it to be released.

This initiative saw the development of an MVP that could demonstrate how multiple and varied datasets could be used for nowcasting to determine COVID-19’s impact on GDP growth rate in select countries in Asia and the Pacific. The MVP was developed by Ddata, which was selected in ADB’s “Nowcasting the Economic Impact of COVID-19,” besting 65 other teams. Ddata used a combination of big data and machine learning to develop its nowcasting model that links and aggregates thousands of datasets to predict the movement of a target indicator, in this case, GDP growth rates. Raw data came from multiple sources, had different types, and had varying release dates to enable the model to reflect holistic economic measurements, eliminate noise, and increase the model’s accuracy in making predictions.

The solution offered by Ddata had two main features:

• Data ingestion, transformation, and modeling. Ddata reformatted various datasets into a common format and linked them together. All data transformations were recorded for transparency. A quality check was also performed to ensure maximum accuracy. Ddata combined linear and nonlinear methods (i.e., neural networks) to develop the model.
• Backfilling. Ddata trained and tested the model using data from the 2008–2009 financial crisis because this period, like the COVID-19 pandemic, was marked by economic volatility. This was done to check if it could accurately predict the economic shocks and recovery using data from that period. In cases where data was missing, “backfilling” was done using machine learning to fill in the missing data points using the available variables and their relationships.

ADB and Ddata agreed to narrow the focus of this initiative to the quarterly GDP growth rates of five countries in the Asia and the Pacific region (Armenia, Fiji, Myanmar, Pakistan, and the Philippines). These countries had different geographical locations and varied in terms of the types of data they release and the frequency with which these official statistics are released, making them good representative cases for the Asia and Pacific region.

Ddata presented its final output to ADB on 29 July 2021. The model accurately nowcasted GDP growth over time for 6–12 months across all countries using multiple data sources. The accuracy was particularly good for high-data countries such as the Philippines. Meanwhile, the model was able to identify data-point variations for low-data countries such as Pakistan and Myanmar. The nowcast results were released weekly, thus ensuring timeliness.

The model was in line with or even better than similar models in terms of analysis errors. Ddata also developed cloud-based automated pipelines that allowed the creation of release-based updates.

The nowcasting model has progressed to the beta stage of testing by the end of the MVP development. Ddata intended to gather more data to further improve its model, which is scalable, by incentivizing people to add data to the peer-to-peer distributed ledger technology-based data-sharing platform they created to crowdsource alternative datasets.