本研究在於探討如何應用支援向量機(Support Vector Machine)在台灣股市中,分別採用分類(Classification)與迴歸(Regression)技術建立個股投資預測模型,並且輔以粒子群最佳化(Particle Swarm Optimization)進行參數最佳化與變數篩選。一直以來台灣股市被認為極難分析且預測,所以本研究希望能夠有效地整合財務基本面與技術分析面的資訊,以建立穩健的投資預測模式。 本研究分析模式為根據公司各季所公布的財務報表與自行計算出的技術指標進行預測。在採用台灣股市歷史資料進行投資模擬後,實證本研究的方法可以產生不錯的獲利曲線。同時,本研究所得的模擬結果經過分析,也證明優於類神經網路預測模式或是買進及持有的投資策略。因此,本研究認為在同時整合財務基本面與技術分析面的資訊下,使用支援向量機和粒子群最佳化建立個股投資預測模型,不失為一種可以有效獲利而且相當穩健的投資策略。
This study is to explore how to apply the Support Vector Machine (SVM) method to build the prediction model from classification and regression techniques for individual stock in Taiwan Stock Market and leverage Particle Swarm Optimization (PSO) for parameter optimization and feature selection. Usually researchers think that it’s difficult to have an efficient prediction about the trends of Taiwan Stock Market. It’s the goal of this study to create a robust financial forecast model based on both fundamental and technical analysis information. This research uses quarterly financial reports and simple technical indexes as independent variables for financial prediction. The empirical results show that the performance of PSO+SVM is much better than that of artificial neural network or Buy & Hold investing strategy. Therefore, it is proven to be an efficient and robust investing strategy by combining both SVM and PSO data mining techniques.