본문으로 바로가기

World Economy Brief

Publications

To list

Developing an International Macroeconomic Forecasting Model Based on Big Data

  • Author Sang-Ha Yoon
  • Series24-18
  • Date2024-06-13
In the era of big data, economists are exploring new data sources and methodologies to improve economic forecasting. This study examines the potential of big data and machine learning in enhancing the predictive power of international macroeconomic forecasting models.

The research utilizes both structured and unstructured data to forecast Korea's GDP growth rate. For structured data, around 200 macroeconomic and financial indicators from Korea and the U.S. were used with machine learning techniques (Random Forest, XGBoost, LSTM) and ensemble models. Results show that machine learning generally outperforms traditional econometric models, particularly for one-quarter-ahead forecasts, although performance varies by country and period.

For unstructured data, the study uses Naver search data as a proxy for public sentiment. Using Dynamic Model Averaging and Selection (DMA and DMS) techniques, it incorporates eight Naver search indices alongside traditional macroeconomic variables. The findings suggest that online search data improves predictive power, especially in capturing economic turning points.

The study also compares these big data-driven models with a Dynamic Stochastic General Equilibrium (DSGE) model. While DSGE offers policy analysis capabilities, its in-sample forecasts make direct comparison difficult. However, DMA and DMS models using search indices seem to better capture the GDP plunge in 2020.

Based on the research findings, the author offers several suggestions to maximize the potential of big data. He stresses the importance of discovering and constructing diverse data sources, while also developing new analytical techniques such as machine learning. Furthermore, he suggests that big data models can be used as auxiliary indicators to complement existing forecasting models, and proposes that combining structural models with big data methodologies could create synergistic effects. Lastly, by using text mining on various online sources to build comprehensive databases, we can secure richer and more real-time economic data. These suggestions demonstrate the significant potential of big data in improving the accuracy of international macroeconomic forecasting, particularly emphasizing its effectiveness in situations where the economy is undergoing rapid changes.

File

Prev Next List

공공누리 OPEN / 공공저작물 자유이용허락 - 출처표시, 상업용금지, 변경금지 공공저작물 자유이용허락 표시기준 (공공누리, KOGL) 제4유형

대외경제정책연구원의 본 공공저작물은 "공공누리 제4유형 : 출처표시 + 상업적 금지 + 변경금지” 조건에 따라 이용할 수 있습니다. 저작권정책 참조