A Gaussian additive model based on the stacking framework and its application in stock price forecasting

With the rapid development of big data technologies, employing stock price prediction models has become crucial in quantitative finance. Among the various prediction models, selecting appropriate features, mining factors, and quantifying the uncertainty of stock prices are essential challenges. To address these issues, in this paper proposes an improved Gaussian Process Additive Model based on the stacking method. The proposed approach introduces a "re-encoding and reduction" design in the first layer of the stacking framework, effectively filtering redundant features. Furthermore, the method uses the Gaussian Process Additive Model in the second layer of the stacking framework to capture interactions between features. By incorporating a Gaussian process prior, the model can provide uncertainty estimates for the predictions. Additionally, the flexibility in selecting sub-models and kernel functions within the Gaussian additive framework enhances the method's capacity to fit complex functions, offering a high degree of flexibility. Both simulation experiments and real-world data analysis demonstrate that the proposed method is highly competitive compared to other classical prediction models.