摘要
计算金融是一门将机器学习的有关理论应用于金融研究的新兴学科.而股票收益预测则是金融研究的一个重要课题,在规避风险与投资决策中起着举足轻重的作用.作者基于统计学习的基本理论与计算金融的研究方法,将支持向量回归机这一新型神经网络应用于收益序列预测的回归分析,力求在克服数据过拟合现象的基础上寻找问题的全局最优解.通过交叉验证选择学习参数,实验表明基于二次规划与核函数理论的支持向量回归机能准确捕捉动态股票收益序列的波形特征,其预测性能与多层感知器以及广义回归神经网络进行比较,具有较为明显的优势.
Computational Finance is a new field which research applications of Machine Learning in finance. Stock returns prediction is important branch of finance, which play vital important role in finance to reduce risk and take better decisions. Based on theory of Statistic Learning and Computational Finance, this paper deals with the application of support vector regression （SVR） in stock returns prediction to solve the over-fitting problem and gain a globally optimal solution. Through selecting parameters by cross validation, the validity of SVR for stock returns prediction were analyzed through experiments on real-world stock data. It appears that SVR can describe the curve characteristic of nonstationary stock time series well and perform better than both the multi-layer perceptron （MLP） and the general regression neural networks （GRNN）. 4figs., 3tabs., 1 lrefs.
出处
《湖南科技大学学报（自然科学版）》
CAS
北大核心
2009年第2期70-73,共4页
Journal of Hunan University of Science And Technology：Natural Science Edition
基金
湖南省自然科学基金(09JJ3129)
关键词
股票收益预测
支持向量回归机
多层感知器
广义回归神经网络
Stock returns prediction
support vector regression （SVR）
multi-layer perceptron （MLP）
general regression neural networks （GRNN）.
