Bank of Korea Develops New Economic Forecasting Model 'PrismNow' Utilizing Big Data

Utilizing Latest Techniques Including Machine Learning
Short-Term GDP Forecast Valid Within 2 Quarters

The Bank of Korea has developed 'PrismNow,' a model that provides useful information for short-term economic outlook judgments within two quarters, including Gross Domestic Product (GDP). PrismNow is a new economic forecasting model that utilizes the latest techniques such as big data and machine learning in addition to traditional econometric models, and it has recently been recognized as an advanced forecasting model system by the Bank for International Settlements (BIS) and the European Central Bank (ECB).

On the 19th, the Bank of Korea introduced PrismNow, a new short-term economic forecasting model system that can be used for short-term GDP forecasts by leveraging the latest techniques such as big data, through a report titled "Model Synthesis System for Real-Time Economic Forecasting and Risk Assessment: PrismNow" published in the 'BOK Economic Research.'

PrismNow, released by the Bank of Korea this time, is an economic forecasting model that fuses and decomposes short-term forecasts derived from various models such as traditional econometric models, big data models, and machine learning models, similar to how a prism disperses light to separate visible rays. While macro models built on historical data had limitations in forecast accuracy when economic environments changed rapidly, this model incorporates the latest techniques such as big data and machine learning that can quickly and flexibly consider large-scale economic information, overcoming the structural rigidity of traditional econometric models and enhancing predictive power.

According to the Bank of Korea, an evaluation of PrismNow's forecasting ability showed that it outperforms existing individual models or other model synthesis methods in predictive performance. Notably, even in situations where domestic and international economic uncertainties have greatly expanded, such as before and after COVID-19, the recent Ukraine crisis, and the Israel-Hamas conflict, PrismNow demonstrated considerably stable forecasting capabilities.

Additionally, internal tests indicated that PrismNow significantly reduces forecasting burdens by automating the acquisition of large-scale information and the estimation processes of multiple models, representing a considerable improvement in the efficiency of forecasting operations.

Examining the forecasting power of individual models included in PrismNow by prediction period, traditional econometric models in simple forms showed superior forecasting ability for the current quarter predictions where most performance information is available, whereas big data and machine learning models generally exhibited better predictive power for the next quarter predictions where no performance information exists. For example, around the time of the rapidly changing economy due to COVID-19, models reflecting text and sentiment indicators were analyzed to have the best forecasting performance.

During the development of PrismNow, the Bank of Korea directly visited the BIS and ECB to present detailed results of this model, receiving favorable reviews, and agreed to further strengthen cooperation with these institutions in the field of model systems in the future.

Seobum Seok, head of the Macroeconomic Model Team at the Bank of Korea's Economic Modeling Office, stated, "PrismNow, developed this time, is a model that provides useful information for short-term economic outlook judgments within two quarters and is expected to significantly contribute to improving the accuracy of economic forecasts." He added, "This model is the first effort to advance the Bank's economic forecasting system since the establishment of the Economic Modeling Office in March this year, and we are currently developing a large-scale FRB/US-type model to be used as a core model in medium-term forecasts."

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