基于神经网络的我国沪深股市星期效应新探

Research on Day- of- the- Week Effect in Shanghai and Shenzhen Securities Markets Based on Neural Networks

考虑除日收益率外成交量、成交额和价格极差等变量对星期效应的影响,并对经过归一化与SMOTE(过抽样技术)处理后的沪深指数样本数据利用神经网络分类的方法进行检验,结果显示我国沪市存在显著的周一、周二和周四效应,我国深市存在显著的周一效应。通过神经网络敏感性分析发现星期效应对成交量、价格极差和成交额较敏感,而对收益率的敏感性较小。最后运用信息流假说对星期效应存在的原因做出了解释。

The paper uses daily mean return and also trade volume, trade sum and H -L（ high price minus low price ） to investigate the impact of different variables to day - of - the - week effect. We use neural networks to classify the data of Shenzhen and Shanghai securities markets which are disposed by normalization and SMOTE. The results show that Monday, Tuesday and Thursday effect are significant in Shanghai and Monday effect is significant in Shenzhen. The results of neural networks sensitive analysis also reveal that day - of - the - week effect is sensitive to trade volume, trade sum and H - L, but not so sensitive to daily mean return. At last, we explain the existence of day - of - the week effect using information flow hypothesis.