Gaussian process regression model optimized based on stacking framework and its application in financial quantitative trading

In the field of quantitative finance, stock price prediction serves as an important reference for investment decisions. However, traditional forecasting models often encounter the problem of dimensionality disaster when dealing with high-dimensional factor data, and they struggle to capture the complex nonlinear relationships between stock prices and influencing factors. Based on this, this paper proposes a Gaussian process regression model optimized within a stacking framework. On the one hand, the proposed method re-encodes high-dimensional data through a feature random sampling strategy in the first layer of the stacking framework, thereby alleviating the dimensionality disaster problem. On the other hand, the standardized output of each base learner is used as a new input feature, and the Gaussian process regression model is employed as the second layer of the stacking framework. This approach allows the model to fit the unknown nonlinear connection structure by selecting an appropriate kernel function and provides uncertainty quantification of stock price predictions from a probabilistic perspective. Extensive simulation experiments and actual data analysis demonstrate that the proposed model exhibits certain advantages over some existing classical forecasting models.