Abstract: The research on statistical characteristics of financial time series and neural network prediction is of great significance for grasping the law of financial market development and guiding long-term or short-term investment behavior. The NASDAQ semiconductor index was investigated by Empirical Mode Decomposition (EMD), Time-Dependent Intrinsic Correlation (TDIC) and Hilbert spectrum analysis, the NASDAQ semiconductor industry index was predicted by the priori-tested BP neural network method. Statistical analysis showed that IMFs presented certain periodicity. The result of the spectrum analysis showed that statistics behavior could be found in the semiconductor industry index. Eventually, it was found that the semiconductor industry index would keep fluctuating in the near future. Different BP neural network methods could be taken to direct the long-term investigation and short-term investigation. The BP neural network method could provide effective reference to investigators.