基于时间相关性的股票价格混合预测模型

A Hybrid model based on Time Correlation for Stock Price Forecasting

对于金融市场决策而言,金融时间序列的分析预测扮演着越来越重要的角色。但通常的分析预测模型没有考虑金融时间序列数据内部的时间相关性问题,这在很大程度上影响了预测模型精度的进一步提高。因此提出一种新的股票价格混合预测模型,分别用ARIMA和基于时间测地线距离的SVM处理金融时序的线性和非线性成分。实验表明,该混合模型可以有效克服传统SVM核函数利用欧式距离表征时序数据相关性的不足,从而显著提高组合模型的预测精度。

Financial time series analysis and forecasting has drawn considerable attention as an active research for financial decision -making. However, many hybrid methods combining linear and nonlinear models do not in- corporate the time correlations knowledge in the financial time series data, which influences the generalization per- formance of the hybrid models. This paper presents a novel two stage hybrid forecasting model ARIMATGDSVM to make full use of the time correlation knowledge in the complex financial series data. In the hybrid model, the ARI- MA model and the proposed SVM based on time geodesic distance are used to deal with the linear and nonlinear components respectively. Finally, the sum of the two prediction results got in the different stages is regarded as the final forecasting result of the hybrid model. The experimental results show that the proposed hybrid model overcomes the limitation of the SVM in expressing time ＇correlations by the Euclidean distance and improves greatly the forecasting accuracy.