Analysis of financial market trend based on autoregressive conditional heteroscedastic model and BP neural network prediction

Abstract

High-frequency data such as stock prices are aggregated into low-frequency monthly data for modeling. However, the summation method only applies to high frequency data in the form of flow, and the summation method reduces the sample size. Based on this, this paper uses the mixing model to construct the financial status index, which can model the data of different frequencies and compensate for the defects of the same frequency data modeling to some extent. Moreover, based on principal component analysis and text mining technology, this paper constructs two kinds of sentiment indexes, and studies the influence and prediction of two sentiment indexes on the closing price of stock market. In addition, in the empirical analysis, this paper establishes the GARCH model and BP neural network prediction model and predicts the closing price. Finally, this paper compares the pros and cons of predictive models and sentiment indices. The research shows that the BP neural network model established by using the lag variable of the Web text sentiment index as the input layer variable is more reliable and can be widely used in the stock market.