The leverage of artificial intelligence (AI) and statistical methods can quickly process the languages of a myriad of people and generate high-quality analyses. For this study, I build big data from over 128,000 corpora te analysis reports that are written by business analysis experts and extract meaningful industrial monitoring information from it. Natural language processing (NLP) and statistical computing techniques have been employed to process text data and the procedure is implemented through automated algorithms without human assistance.
All quantitative figures are ignored from the analyst reports to compile analysts’ views completely through qualitative (language) contents. Particularly, business confidence indicators by industry and region as well as the factors behind them have been composed in algorithms based on analysts’ narratives. Furthermore, through keyword analysis, analysts’ views on macroeconomic variables like the exchange rate and interest rate are compiled in algorithms to generate novel numeric indicators.
The proposed text-based indicators has been shown useful in complementing traditional macroeconomic data such as the Gross Domestic Product (GDP) and Business Survey Index (BSI) as well as identifying the factors behind the change of business conditions in industry. Empirical analysis using Granger causality tests has verified the usefulness of the text-based indicator in predicting future economic prospects by showing a one-way causal correlation between the text-based indicator and the cyclical indicator of leading composite index, which is undiscovered when the KOSPI market consensus indicator is used instead of the text-based indicator. The result suggests text data contains useful information that cannot be expressed in numerical data.
The text-mining procedure used in this study is automated by computer algorithms including web-scraping to retrieve data and investigate business conditions and factors to sum and visualize the findings.
Advances in technology have accelerated automation and bolstered efficiency. Application of NLP for macroeconomic analysis is still at the fledgling stage, but given the progress in the recent technology like the generative pre-trained transformer (GPT), AI may soon achieve the level of collecting data and deriving conclusion on the economy on its own. My findings justify the need for continued research in text analysis and development of AI-enabled analytics to enhance efficiency in economic research.