Abstract: In recent years, influenced by factors such as the shift in monetary policies of major economies and frequent favorable policies in the stock market, China's stock prices have experienced intensified volatility. Investor sentiment in the stock market is highly sensitive to policy changes, and policy interest rates are in a downward cycle. In this context, studying the impact of interest rate changes on stock prices from the perspective of investor sentiment decomposition has important practical significance. This paper uses the Regional Vector Autoregressive (RVAR) model to decompose investor sentiment, and adopts the Regional Dependent Local Projection (RLP) model to examine the transmission effect of interest rate changes on stock prices in China under different channels of investor sentiment. The findings show that when investor sentiment is low, interest rate changes cause multi-level stock price movements in the same direction and opposite direction through blind and directional investor sentiment channels, respectively. When investor sentiment is high, interest rate changes cause multi-level stock prices to experience first in the same direction and then in the opposite direction through blind investor sentiment channels, and first in the opposite direction and then in the same direction through directional investor sentiment channels. Meanwhile, in the zone of high investor sentiment, the self reinforcing characteristics of investor sentiment in China's stock market are mainly manifested in blind investor sentiment rather than directional investor sentiment. Therefore, in order to ensure the stability of stock prices after interest rate adjustments, China should fully leverage the investment leading role of smart investors over ordinary investors, and guide and manage the emotions and capital flows of ordinary investors in the stock market through a small number of smart investors after interest rate changes.

Key words: interest rate policy, investor sentiment, stock prices, multi-level stock market, RLP model