Research on Stock Price Multi-Factor Prediction Model Based on Bayesian Model Averaging

Abstract

Factor mining is a crucial component in constructing stock price prediction and quantitative models, where factor mining methods based on feature dimensionality reduction, such as Principal Component Analysis (PCA) and sufficient dimensionality reduction, are widely utilized. However, the connection structure between the quantitative factors extracted using these methods and stock prices is unknown, leading to potential issues like overfitting or underfitting in prediction models. In light of this, this paper proposes a multi-factor prediction model based on Bayesian model averaging. On one hand, the proposed method employs the concept of model averaging instead of model selection, effectively balancing the variance and bias of prediction models. On the other hand, it can adaptively choose sub-models that play a crucial role in predicting stock prices, thereby enhancing overall prediction accuracy. Empirical data analysis indicates that the proposed method, compared to PCA-based Lasso and Ridge regression, exhibits smaller mean squared error and possesses a certain level of robustness. Lastly, by incorporating other model averaging techniques such as the Bagging algorithm, the generalization ability of the proposed method can be further improved.