支持向量机在股指现货和衍生品关系建模中的应用

Application of SVM in Modeling the Relationship Between Stock Index Spots and Derivatives

首先按照先前学者的思路,利用传统的向量自回归-误差修正(VECM)模型进行分析,结果发现期货对现货有明显的引领效应.但若对特定的异常时段进行分析,期货引领现货的效应有所减弱但仍比较明显.考虑到VECM模型自身存在的一些问题,又尝试了遗传算法、最优热路径方法等非参数统计方法.其中,遗传算法收效甚微,但最优热路径算法得到了期货长期领先现货平均2.45分钟、而在2015年股灾期间,期现货之间的领先滞后关系出现了一定程度上的反转的结论.最终本文尝试使用支持向量机(SVM)方法对这一问题进行研究,将数据的尺度从大到小进行分析,目标从寻找长期关系转到短期关系,但最终效果不甚理想.因此认为用SVM很难训练出一个让人满意的分类器,仅用期货、现货等数据预测市场走势无论是短期还是长期来看都是十分困难的.

This paper firstly makes analysis using traditional VECM model following the ideas of previous researchers, which leads to a conclusion that futures have an obvious leading effect relative to spots. However, if restricted to some periods of market shock, this leading effect of futures over spots will become less obvious but still observable. Considering that VECM model has some inherent flaws, this paper turns to non-parametric statistical methods including Genetic Algorithm and Optimal Thermal Path method. Genetic Algorithm has little effect, but Optimal Thermal Path algorithm leads to a conclusion that futures normally leads spots by 2.45 minutes, but during the stock market crash in 2015, the leadlag relationship between futures and spots reversed to some extent. In the end, we tried using SVM on this problem. We used data of various frequency, focusing on both long-term relationships and short-term relationships, but could not make any satisfactory conclusion.Therefore, we now believe that it is very difficult to train a satisfactory classifier with SVM,and to predict market trends using only futures and spots data, both in the short-term and long-term.