A multi-factor quantitative model based on weighted SIR-SAVE and model averaging

Abstract

In financial quantitative research, there are many kinds of predictors that are closely related to stock prices, so how to design effective machine learning algorithms for such high-dimensional prediction problems is a hot issue. Based on this, this paper proposes a stock factor quantization and prediction model based on the idea of sufficient dimensionality reduction and model averaging. The proposed method is based on the principle of conditional independence, which can greatly retain the validity of the predictors on the stock price while solving the dimensionality catastrophe problem. In addition, the method adaptively learns the link function between the quantized factors and stock prices by using the model averaging method to effectively weigh the variance and bias of the prediction model. In the actual data analysis, this paper chooses the weighted SIR-SAVE as the adequate dimensionality reduction method, the mutual information criterion as the assignment rule of model averaging, and some classical high-dimensional regression techniques as the comparison method, and the empirical results show that the proposed method has a higher accuracy under the evaluation indexes of mean squared error and absolute error, and possesses a certain degree of robustness. Finally, since the sub-models in the model averaging are free, the proposed method is also applicable to the problem of classifying the prediction of stock ups and downs.