基於支持向量回歸的選股模型實證研究 —以台股市場爲例

當今世界的發展下，隨著信息技術的快速發展，計算機進行資料處理越來越流行。近幾年來機器學習技術的火爆，更加催生了將機器學習用在金融、經濟等領域的熱潮。本文選擇了機器學習領域的成熟算法支持向量機的分支——支持向量回歸，當作基礎的算法，搭配以網格搜索、主成分分析法對模型進行參數尋優和對資料進行降維處理。選取2009年第一季度到2018年第四季度的財報資料共170個指標和收盤價資料，利用2009年到2015年的資料對上述的支持向量回歸模型進行訓練，再利用2016年到2018年的資料進行回測。回測結果表明，支援向量回歸模型對於台股市場有一定的預測能力，當使用主成分分析法提取特徵個數為十個的時候，整體模型的報酬率表現最好，當特徵個數增加或者減少時，一定程度上可以增加模型的擬合程度，但是會增加樣本內和樣本外的R^2的差，導致模型的一般化能力減弱。

關鍵字

機器學習；支持向量回歸；支持向量機；量化交易；選股模型

並列摘要

Since the great development of today's world, computer data processing has become more and more popular with the rapid development of information technology. In recent years, the popularity of Machine Learning technology has spawned a boom in applications of finance and economy. This paper chooses the Support Vector Regression (SVR) as the basic algorithm, a branch of the mature Machine Learning approach, Support Vector Mchine (SVM). With the application of grid search and principal component analysis method, the parameters of the model are optimized and the data dimension is reduced. 170 indicators and close price data from Q1 2009 to Q4 2018 are selected, when data from 2009 to 2015 is used to train the SVR model, and data from 2016 to 2018 is used for back testing. The back test result shows that SVR model has predictive ability for the Taiwan stock market. When ten features are extracted by principal component analysis, the return of overall model reaches the best. When the number of features increases or reduces, the fitting degree of the model can be increased to some extent, but the difference of R^2 between sample and out of sample increases, resulting in a weakness of the model generalization ability.