The central Bank of Korea (BOK), as part of its research, compared the results of deep learning-based calculations and traditional econometrics methods to make short-term forecasts on monthly exports and the daily won-dollar exchange rate.

According to bank's research paper, which analysed prediction tests of monthly exports from Korea and daily Korean won-US dollar exchange rates, deep learning (an advanced form of artificial intelligence) produced more accurate results in predicting outcomes, even with the sorts of non-granular data sets which are normally used for conventional econometric models.

Moreover, deep learning produced results with a noticeably narrow margin of error compared with the traditional method, especially when predicting exchange rate forecasts.

The working paper says that compared to conventional econometric approaches such as VECM, the deep learning approach shows not only better prediction powers, but also more informative error bands that may indicate periodical developments of uncertainty in the economy.

In this regard, the paper concludes that deep learning is useful, even with data that seems to not be coordinated with any machine learning approach, but there is room for improvement. The study is no more than a prototype that can be improved with “enhanced data.”

__________________________________________________

Sources:

https://www.bok.or.kr/viewer/skin/doc.html?fn=202008280531279490.pdf&rs=/webview/result/E0002902/202008
https://www.bok.or.kr/imerEng/bbs/E0002902/view.do?nttId=10060043&menuNo=600342
https://www.theinvestor.co.kr/view.php?ud=20200902000794
https://www.centralbanking.com/central-banks/economics/macroeconomics/7676766/deep-learning-can-beat-other-forecast-methods-bank-of-korea-research?utm_medium=email&utm_campaign=Do%20not%20use&utm_source=CB.DCM.Editors_Updates&im_amfcid=17761296&im_amfmdf=cbe28bac3b65e119f1a09eccccaa3cc9