The Bank of Korea: "News-Based Economic Indicators Show High Correlation with Official Statistics"

(Data provided by Bank of Korea)

An analysis has revealed that economic indicators created based on news have a high correlation with official statistics.

On the 16th, the Bank of Korea stated in its BOK Issue Note titled 'Economic Forecasting Using News Text' that "news text-based economic indicators lead official statistics by 0 to 9 months and show a high correlation with official statistics."

In the real sector, there are few economic indicators and official statistics usually have a time lag of more than a month. However, news text big data delivers diverse and vast information quickly and includes qualitative information such as expert opinions, which is expected to help improve the accuracy of economic forecasts.

For the investigation, the Bank of Korea selected 15 sectors of high economic interest, including production, employment, prices, stock prices, and housing prices, and established target indicators for each sector to create text indicators predicting the flow of these indicators. To create the text indicators, all economic news articles posted on internet portal sites from January 2005 to March of this year were analyzed.

Comparing the created indicators with official statistics, the text indicators showed a correlation of 0.35 to 0.73 with official statistics at lead times of 0 to 9 months and demonstrated significant causality in most cases at lead times of 1 to 7 months, indicating that they contain important information.

The Bank of Korea explained, "Looking at the end of June 2020, when the impact of COVID-19 was significant, GDP forecasts for the second quarter of 2020 were closer to observed values when text indicators were added compared to using only official statistics, and changes in other economic variables were better captured."

The Bank of Korea added, "News text includes qualitative information such as various experts' opinions and forecasts and is available in real time, so it is necessary to integrate and quantify this information for use in economic forecasting." It also stated, "It is useful for quickly and accurately grasping economic trends and is expected to contribute to reducing human errors compared to using news through qualitative methods."