A recent paper from the European Central Bank (ECB) reveals that analyzing qualitative comments in Purchasing Managers' Index (PMI) releases using ChatGPT can significantly enhance GDP forecasting capabilities. The ECB has been leveraging artificial intelligence in recent years to improve predictions by scraping price data from the web and classifying data with large language models.
This new study highlights that even limited text from PMI releases can offer meaningful insights into economic development, thereby improving real-time assessments of economic activity. The research uniquely focuses on the narrative, tone, and anecdotes within PMI press releases.
Economists used ChatGPT to generate activity sentiment scores from these narratives and anecdotes, integrating them into the current quarter's growth forecast, known as nowcasting. The authors found that the enhanced PMI text scores greatly improved the accuracy of GDP forecasts when compared to two benchmark indicators.
They emphasized that just two pages of text, rather than millions of newspaper articles, were sufficient to improve the existing indicators that are hard to surpass.
