基于近邻互信息的SVM-GARCH股票价格预测模型研究

为了克服传统线性模型分析处理收益率数据非线性因素的不足，本文提出一种新的基于近邻互信息特征选择的SVM-GARCH预测模型。该模型利用SVM处理高维非线性数据的优势，不仅包含了股指序列自身的历史数据信息，而且通过近邻互信息的方式融合了与目标股指数据关系密切的周边证券市场的相关变化信息。仿真实验结果表明，该模型在时序数据除噪、趋势判别以及预测的精确度等方面均优于传统的ARMA-GARCH模型。

A SVM-GARCH Model for Stock Price Forecasting Based on Neighborhood Mutual Information

In order to overcome the limitations of the traditional linear model in dealing with the nonlinearity in time series, a novel SVM-GARCH forecasting model is proposed based on the neighborhood mutual information. By constructing high dimensional input variables, the proposed nonlinear model not only absorbs the historical information in the time series data but also incorporates the stock market information in different regions through feature selection by the neighborhood mutual information. Empirical studies demonstrate that the proposed model is superior to the traditional linear ARMA-GARCH model in terms of data denosing, trend discrimination and prediction accuracy etc.