基于稀疏贝叶斯极限学习机算法的股票价格预测

Stock Price Prediction Based on Sparse Bayesian Extreme Learning Machine

股票价格模型是金融理论分析与实证分析的重要基础,学术界与金融业界对其建模预测一直保持着极大的兴趣,但由于股票价格表现出高噪声性、强非线性、随机分形结构以及长记忆效应等特点,需要融合优化算法、统计学习方法与金融学理论对其建模分析。基于传统方法的股票价格过程建模预测结果往往精度不够好,所建立的模型泛化能力较差。基于稀疏贝叶斯极限学习机(SBELM)方法对股票价格进行建模预测,SBELM既能保持传统极限学习机(ELM)算法训练过程简捷的优点,又具有稀疏贝叶斯学习机自动选择隐藏层节点数的优点。利用上证综合指数2014-2015年的市场数据,比较基于SBELM方法的建模预测与基于贝叶斯极限学习机(BLEM)、ELM以及BP神经网络学习算法的建模预测,结果表明,基于SBELM方法的市场指数模型预测精度最高、泛化能力最强,具有较好的应用价值。

Stock price model is an important basis of finance theoretical and empirical analysis. Academia and financial sector have maintained great interest in its modeling and predicting. Because of the high noise,strong nonlinearity,random fractal structure and long memory effect of stock price,we need integrate optimization algorithm,statistical learning and financial theory to set up its model and then analyze it. The modeling of the stock price process based on the traditional method do not achieve precise results often,so those models have a poor generalization ability. In this paper,we propose a modeling based on the sparse Bayesian extreme learning machine（ SBELM） to forecast the stock price. SBELM not only keeps the simplicity of the traditional extreme machine learning（ ELM） in training process but also has the advantage of automatically choosing the number of hidden layer nodes. By using the data of Shanghai Composite Index from 2014 to 2015,experiments are conducted to compare the model prediction of SBELM with it of Bayesian ELM（ BELM）, ELM and back-propagated neural network（ BPNN）. The results show that the prediction accuracy and the model generalization ability of SBELM are the best,so this method has a high application value.