Bank of Korea: "Deep Learning Technology Effective in Predicting GDP Growth Rate"

The big data-specialized deep learning (machine self-learning) technology developed by the Bank of Korea has been found to be effective in predicting the growth rate of the Gross Domestic Product (GDP).

On the 3rd, the Bank of Korea announced this through a report titled "Development of a Real-Time Current Quarter Economic Forecasting (GDP nowcasting) System Using Digital New Technologies."

According to the report, an evaluation of a real-time economic forecasting system that combines the Dynamic Factor Model (DFM), used by major central banks, with the deep learning model LSTM (Long Short-Term Memory) showed that LSTM had relatively higher predictive power for GDP growth rates.

In fact, for the first quarter of 2020, when COVID-19 was prevalent, LSTM calculated the preliminary GDP growth rate as -1.1%, close to the actual figure of -1.4%, whereas DFM predicted -0.4%, showing a difference.

The report explained that both models captured in advance that GDP would sharply decline in the first and second quarters of 2020 and rebound in the third quarter, but LSTM demonstrated higher predictive power in situations with significant economic uncertainty, such as the spread of COVID-19.

According to the report, DFM efficiently extracts information contained in each variable by utilizing the structural relationships among economic variables, while the LSTM algorithm is effective in capturing nonlinear and interdependent relationships among multiple variables using deep learning networks.

The Bank of Korea stated, "We plan to apply the accumulated know-how of utilizing digital new technologies from this system development to various research and survey tasks such as economic forecasting and market monitoring, and actively share research outcomes through domestic and international exchanges."

This content was produced with the assistance of AI translation services.