Abstract: In this paper, it is argued that investor attentioncan provide more information in forecasting realized volatility of the SSE composite index and SZSE component index. Time-delay detrended cross-correlation analysis (DCCA)is employed to investigate the lead-lag relationships between investor attention and realized volatility. We further use various ARMA and HAR models as benchmarks to verify the possible effect of investor attention on stock market volatility forecasting. The empirical results show that, in a low volatile market environment with few Internet search activities, investor attention represented by Baidu search volume cannot improve the volatility forecasting accuracy in the Chinese stock market. While in a high turbulent market condition with plentiful Internet search activities, the extended ARMA and HAR models incorporating investor attention can offer more accurate volatility forecasts than the benchmarks. Our findings have important implications in practice. For retail investors and institutional investors, they can previously identify stock market tendency and get profit opportunities. As for regulators, they can strengthen the market regulation and foster an efficient stock market.

Key words: volatility forecasting, Baidu Index, investor attention, stock market volatility