The Bank of Korea’s headquarters in central Seoul (Image courtesy of Yonhap)
SEOUL, Feb. 6 (Korea Bizwire) –The Bank of Korea announced on February 6 that it has developed an inflation forecast model with improved predictive accuracy, utilizing big data, artificial intelligence (AI), and machine learning (ML) technologies.
Lee Changhoon, head of the digital technology R&D team at the Bank of Korea, shared these insights in a report titled “Real-Time Inflation Forecasting Using Big Data and Machine Learning Algorithms”.
During the development of the model, the Bank of Korea employed various forecasting methods, including tree-based ML algorithms, linear regression models, ensemble models (averaging forecasts from ML and linear regression models), and benchmark models such as random walk and ARIMA.
Comparative evaluation of the forecast accuracy of these models from January 2016 to September 2023 revealed that the ensemble model consistently outperformed the benchmark models across all forecasting horizons and evaluation metrics.
The Bank of Korea conducted real-time forecast simulations targeting periods of significant changes in South Korea’s inflation trends using the ensemble model.
The results accurately predicted a slight increase in July 2022 compared to the previous month and forecasted a slight decrease and a significant drop in the 3-month and 12-month forecasts, respectively.
For January of the current year, the monthly forecast initially predicted an inflation rate of 3.1% for the first two weeks, similar to December of last year. However, the forecast adjusted to 2.9% towards the end of the month, closely approximating the actual inflation rate of 2.8%, due to the expected decrease in inflation.
However, the study noted that the 3-month and 12-month forecasts had larger prediction errors compared to the monthly forecasts, and the monthly forecasts showed limited improvement from newly added information within the month.
Lee explained that traditional economic models struggle to reflect non-linear relationships and interdependencies among variables during significant shocks like the COVID-19 pandemic.
In contrast, ML algorithms can efficiently compute such shocks through their internal mechanisms without extensive effort.
A Bank of Korea official clarified that the current stage is exploratory, assessing the feasibility of using AI and ML for inflation forecasting, and not yet at a level suitable for official use.
The official expressed optimism that with further research to enhance the model’s reliability, accuracy, and stability, it could potentially be utilized for official forecasts in the future.
Ashley Song (ashley@koreabizwire.com)
