Sentiment and stock price volatility: a multilayer heterogeneous graph network analysis of the new energy vehicle market

This study examines how media and investor sentiment jointly shape price dynamics in the new energy vehicle (NEV) sector, an industry characterized by strong policy sensitivity, concentrated investor attention, and network-based spillovers. Understanding the interaction between these sentiment sources is economically relevant because it sheds light on how information flows and behavioral forces jointly influence asset prices in strategically important and rapidly evolving markets. To address this question, we develop a multilayer heterogeneous graph network model that integrates recurrent neural networks (RNNs) and graph neural networks (GNNs). This framework conceptualizes sentiment as a sequential and network-embedded process, in contrast to existing studies that typically analyze sentiment in isolation. Empirical evidence shows that media sentiment has a statistically and economically significant impact on prices by amplifying volatility through investor sentiment, particularly in high-uncertainty periods. Investor sentiment exerts short-horizon effects, whereas media tone is more persistent but of lower intensity. Transmission is concentrated in a small number of hub firms within the supply chain, and a month-length horizon effectively summarizes the interaction between short-run shocks and persistent information. The analysis reveals multi-level dependencies, with key firms acting as sentiment hubs that propagate shocks across the network, thereby contributing to systemic risk. These findings extend existing theories of sentiment transmission by identifying structural and dynamic channels that connect information flows to price formation. Policy implications follow directly from these results. Timely clarification of market-moving information within one to two trading days, intensified monitoring during the first three trading days following salient events, alignment of disclosure practices with monthly cycles, and targeted transparency at hub firms can help market participants process information more efficiently and mitigate destabilizing amplification effects.