Abstract: This paper conducts a theoretical analysis of information friction and belief shock as well as a test with a SVAR model. The findings show that the fundamental shock can increase the expected macroeconomic indicator bias and the belief shock can decrease the expected macroeconomic indicator bias, and that the fundamental shock mainly acts on the actual macroeconomic indicators, while the belief shock mainly acts on the expected macroeconomic indicators. Further, the text recognition technology for big data is used to identify the contents of central bank communication both in the written form and verbal form, thus to identify the effective path of action of central bank communication. The results of the analysis show that central bank communication can indirectly act on expectation bias by moderating the beliefs of micro-subjects. Among them, the written-form communication and the narratives about economic fundamentals are suitable for regulating the output beliefs; and the verbal-form communication and the narratives about monetary policy tendencies are suitable for regulating inflation beliefs; the written-form communication is more effective than the verbal form in regulating expectation bias, and the narratives about economic fundamentals are more effective than the narratives about monetary policy tendencies. Therefore, central bank communication needs to play a full role in regulating the beliefs of micro agents and correcting expectation bias, especially the written-form communication and the narratives about economic fundamentals.

Key words: central bank communication, heterogeneous beliefs, narrative contents, narrative forms, expectation deviation