粒子群支持向量回归在金融时间序列预测中的应用

Application of Particle Swarm Support Vector Regression in Financial Time Series Prediction

金融时间序列一直以来以其非线性、非平稳、信噪比低等特性成为时间序列预测中的难题。支持向量回归(SVR)在对时间序列进行预测时会有模型不稳定、预测精度不高等问题。上述问题的部分原因是模型中的参数选取可能会对预测结果造成影响,因此对于支持向量回归的参数选取问题给出粒子群支持向量回归模型(PSO-SVR)。该模型用粒子群优化算法代替传统的k折交叉验证法求出支持向量回归中的参数,再构建出PSO-SVR模型对金融时间序列进行预测。通过与传统k折交叉验证支持向量回归、BP神经网络以及随机森林模型(RF)预测后得到的均方误差、决定系数等比对,比对各项指标发现PSO-SVR模型要优于其余两者。鉴于PSO-SVR在稳定度和预测精度两方面的优势,表明该模型在金融时间序列的预测上有较好的体现。

Financial time series has been a difficult problem in time series prediction with its non-linear, non-stationary and low signal to noise ratio. Support vector regression (SVR) has the problems of model instability and low prediction accuracy when predicting the time series. Part of the above problem is that the parameter selection in the model may affect the prediction results, so the par-ticle swarm support vector regression model (PSO-SVR) is given for the parameter selection problem of support vector regression. This model uses the particle swarm optimization algorithm instead of the traditional k-fold cross-validation method to find the parameters in the support vector regression, and then constructs the PSO-SVR model to predict the financial time series. The comparison shows that the PSO-SVR model is better with the traditional k-fold cross-validation support vector regression, BP neural network and random forest model (RF) prediction than the other two. Given the advantages of PSO-SVR in both stability and prediction accuracy, it shows that the model is well represented in the prediction of financial time series.