Stock Price Forecasting Model Research Based on Multi-Kernel Gaussian Process Regression and Stacking Algorithm

Abstract

With the rapid development of data collection and storage technologies, the variety of non-Euclidean predictors in the field of quantitative finance is increasing, and how to integrate them with the information of the traditionally used Euclidean factors to improve the performance of predictive models is an important research issue. Based on this, this paper proposes an integrated learning model based on multi kernel learning and Gaussian process regression. Specifically, on the one hand, the proposed method integrates the prediction information from different sources by using different kernel functions as a measure of the difference between elements in the Euclidean and non-Euclidean spaces and based on the Gaussian process regression model. On the other hand, the method improves the prediction accuracy and overall robustness of the prediction model by using the stacking algorithm to quadratically fit the output of the Gaussian process regression model as a new input variable. In addition, the learner used in the secondary fitting is model-free. Simulation analyses illustrate the effectiveness of the proposed method under different experimental settings. Finally, by predicting real stock price data, the results show that the proposed method has a smaller prediction error compared to some classical prediction models.