Real-time macroeconomic projection using narrative central bank communication

Unlike the central banks of most developed economies, the People's Bank of China (PBC) does not release its macroeconomic forecasts to the public but instead carries out narrative communication. We apply a hurdle distributed multinomial regression to PBC communication texts in real time, addressing the ultrahigh dimensionality, sparsity, and look-ahead biases. In addition, we embed text-based indices into mixed-data sampling (MIDAS)-type models and conduct forecast combinations for prediction. Our results argue that the predictive information from communication texts improves the real-time out-of-sample prediction performance. We connect textual analysis and real-time macroeconomic projection, providing new insights into the value of central bank communication.