BOK, Korean Statistical Society hold joint forum

[123RF]
[123RF]

As the economic environment shifts rapidly in the wake of the Fourth Industrial Revolution, new technologies such as machine learning should be adopted to overcome the limits of traditional statistical methods, a forum heard Friday.

Park Min-gyu, a professor at Korea University, made the remarks in a keynote address at the forum, titled "AI, Data and Economic Statistics: A Changing Environment and New Approaches." The forum was held at the Bank of Korea on Friday morning. The Bank of Korea and the Korean Statistical Society jointly hosted the event.

Park said, "Since the Fourth Industrial Revolution, the spread of new technologies such as AI, big data and the Internet of Things has been rapidly changing the environment for compiling economic statistics, and this calls for a diversification of methodology that embraces both existing statistical techniques and new analytical methods."

He then suggested, "To ensure the accuracy and timeliness of official statistics, traditional statistical techniques such as data integration and calibration should be combined with machine learning technology based on high-frequency data such as card sales and search trends."

Machine learning refers to technology that enables computers to learn patterns and rules from data on their own, allowing them to make predictions or judgments about new data.

He cited practical applications of machine learning-based transfer learning, such as macroeconomic nowcasting and cross-country growth rate forecasting. "It is important to take a balanced statistical approach that maintains traditional statistical techniques as the basic framework while applying new technologies in a complementary way, and rigorously evaluates their achievements and limitations based on statistical theory," he said. Nowcasting refers to a technique that uses high-frequency data, such as card sales and search trends, to quickly grasp and estimate current economic conditions or indicators.

In the second session, Park Se-ho, a professor at Hongik University, reinterpreted the gap between microdata and macro statistics from the perspective of distributional shift. Distributional shift refers to a phenomenon in which the distribution of data deviates from the expected distribution. He also presented a cyclical framework, from detecting distributional shifts to correcting them, to secure the stability of estimates. Song Kyung-woo, a professor at Yonsei University, proposed a direction for building a "self-evolving AI assistant" that flexibly adapts to environmental changes. He also introduced methods to statistically control hallucinations and uncertainty in AI responses.

Kim So-jung, a manager on the Bank of Korea's Economic Statistics Research Team, then emphasized the need to diagnose and control "agent bias." Park Jin, a manager on the bank's Distributed National Income Team, diagnosed the causes of the gap between micro- and macro-level data in household distributed income accounts. He also proposed ways to improve their compilation.

Bank of Korea Deputy Governor Kwon Min-su said in a welcoming address that day, "With the recent rapid advancement of AI technology and changes in the data environment, the way statistics are produced and analyzed is also undergoing major change. It is becoming increasingly important to utilize new forms of data and extract meaningful information from vast amounts of data in order to capture economic phenomena more quickly. In addition, how to maintain statistical reliability in an environment where the distribution and structure of data are changing has emerged as an important challenge."


kimstar@heraldcorp.com