Stock Price Prediction Model Based on Generalized Gaussian Process Regression

Abstract

Stock price prediction models hold significant research importance in the realm of quantitative trading. Given this, the present paper introduces a regression model based on the generalized Gaussian process. Specifically, two distinct models are proposed: one is a generalized Gaussian process regression model incorporating L1 regularization, and the other is a generalized Gaussian process regression model utilizing reproducing kernel representation. The former is capable of handling high-dimensional linear features and conducting crucial variable selection, while the latter can effectively model the nonlinear relationship between covariates and response variables. Moreover, both models are adept at capturing the temporal dependence inherent in stock price series. Under diverse simulation settings, the simulation results consistently demonstrate the strong applicability and competitiveness of the proposed methods. Ultimately, empirical analyses are carried out on stock price series data exhibiting varying volatility trends. The results substantiate the superior predictive performance and substantial application value of the proposed methods.