Stock price series prediction optimization framework based on function information stacking and model averaging

Abstract

Stock price series prediction has always been a hot issue in the field of quantitative finance. The commonly used models include ARIMA, GARCH, LSTM neural network and BP neural network. Aiming at these models, this paper proposes an optimization framework based on function information stacking and model averaging. The proposed method uses intra-day price information as auxiliary information and extracts functional features based on functional principal component analysis (PCA). Considering that the underlying model structure between the characteristic variables and the residual series obtained from the original time series prediction model is unknown, this paper uses Stacking method to enhance the data of the characteristic variables to reduce the impact of noise information on the prediction model. In addition, to solve the parameter optimization problem of the original model, this paper proposes a model averaging method using distance covariance weighting to deal with it. In the actual data analysis, this paper takes the LSTM neural network as an example to explore the effectiveness and robustness of the proposed method, and the results show that the proposed method has certain competitiveness. Finally, the proposed optimization method can be used to improve other time series prediction models.