A generative artificial intelligence approach to tracking Chinese Mainland's housing market sentiment using social media data

This paper develops a daily housing market sentiment index that leverages Chinese social media data and generative Artificial Intelligence (GenAI). We adopt a human-in-the-loop methodology and find that the GenAI assessments align closely with human evaluations. Compared to conventional methods, GenAI also offers methodological advantages and arguably provides a more reliable measurement of sentiment. Our empirical analysis demonstrates that the GenAI-driven sentiment index effectively captures public sentiment trends and robustly drives property sales. Furthermore, we utilise GenAI's strong comprehension abilities to identify cities mentioned in microblogs, thereby creating more granular sentiment indices at the city level. These city-level sentiment indices prove useful in predicting local property sales and revealing significant sentiment spillovers from major to smaller cities. Our research offers policymakers enhanced tools for timely market monitoring and underscores the transformative potential of GenAI in research and macroeconomic surveillance, enabling the analysis of previously unmanageable big and unstructured data.