论文部分内容阅读
发行流通股的上市公司财务数据是高维、复杂的,在利用财务指标对股票进行投资选择时往往难以全面考虑。为了从样本股的大量财务指标中提取出低维、有效的特征信息来构成支持向量机(SVM)的训练集,提出了一种启发式算法(HA)对原始财务数据进行预处理,在保存原始数据特征信息的同时提高了训练精度和训练效率。实证结果中,基于该启发式算法的支持向量机选股模型(HA-SVM)最终构造的股票组合的年收益显著高于同期基准组合的年收益。另外,进一步将被广泛使用于降维和数据特征提取的主成分分析法(PCA)与该启发式算法进行对比分析,结果表明,HA-SVM模型的训练准确率、预测准确率以及所选股票组合的年收益情况均显著高于PCA-SVM模型。
The financial data of listed companies that issue tradable shares are high-dimensional and complex, and it is often difficult to take full account of the investment choices of stocks using financial indicators. In order to extract the low dimensional and effective feature information from a large number of financial indicators of the sample shares to form a training set of Support Vector Machine (SVM), a heuristic algorithm (HA) is proposed to preprocess the original financial data, The original data feature information at the same time improve the training accuracy and training efficiency. In the empirical results, the annual returns of the stock portfolios finally constructed by the support vector machine model (HA-SVM) based on the heuristic algorithm are significantly higher than the annual returns of the benchmark portfolios. In addition, principal component analysis (PCA), which is widely used in dimensionality reduction and data feature extraction, is further compared with the heuristic algorithm. The results show that the accuracy of HA-SVM model training, the prediction accuracy and the selected stock portfolio The annual income is significantly higher than the PCA-SVM model.