"Enhancing Inflation Forecast Accuracy with Artificial Intelligence Technology"

It is expected that artificial intelligence (AI) technology will be introduced into macroeconomic forecasts such as inflation predictions in the future.

According to the 'BOK Issue Note, Real-time Inflation Forecasting Using Big Data and Machine Learning Algorithms' released by the Bank of Korea on the 6th, major central banks including those of the United States and Europe are currently researching forecasting models using machine learning.

Machine learning refers to one of several AI technologies and is an algorithm designed for computers to learn patterns inherent in data on their own.

The report also attempted to develop a forecasting model that predicts the flow and real-time outlook of inflation by utilizing data analysis technologies such as big data, AI, and machine learning, and the accuracy was found to be at a considerably high level.

According to the forecasting model, when calculating the Mean Directional Accuracy (MDA), which evaluates how accurately the rise and fall of inflation from January 2016 to September 2023 are predicted, the MDA for all forecast horizons was above 0.6, demonstrating a significantly high predictive power, the report emphasized.

Additionally, when real-time forecasts were conducted for October 2020 and July 2022, periods when inflation trends changed significantly, the July 2022 forecast predicted a slight increase compared to the previous month in the current month forecast, and slight decreases and sharp declines in the 3-month and 12-month forecasts, respectively.

Since October 2023, when the development of the forecasting model was completed, the forecast error has been very small, within 0.2 percentage points (based on the current month forecast).

This study began as the limitations of existing forecasting models were pointed out. After the COVID-19 pandemic, inflation forecasts by major central banks significantly underestimated actual figures, which was criticized as a limitation because existing forecasting models based on traditional economic theories did not adequately reflect the magnitude and persistence of actual inflation shocks.

In response, major central banks such as the U.S. Federal Reserve (Fed) and the European Central Bank (ECB) recently announced research results on inflation forecasting using tree-based machine learning, suggesting that these models can serve as a complement to existing forecasting models.

Lee Chang-hoon, head of the Digital New Technology Team at the Bank of Korea’s Digital Innovation Office, who authored the report, stated, "This study is significant in that it attempted a new method rather than traditional approaches for price forecasts," and added, "We plan to continue efforts to improve through follow-up research so that the forecasting model can enhance the predictive power of the Bank of Korea’s official forecasts."

© The Asia Business Daily(www.asiae.co.kr). All rights reserved.