基于隐马尔科夫模型的中国股票信息探测

Detecting Chinese stock information based on hidden Markov model

应用隐马尔科夫模型对不可观测的股票信息状态建模,并构建信息状态转移概率矩阵刻画信息状态在时间维度上的动态关联性.基于5分钟分时高频数据,利用贝叶斯推断与马尔科夫链蒙特卡洛模拟(MCMC)的方法估计了上证指数、上证50样本股2010年8月的信息状态与信息强度.通过实证验证了模型具有较好的信息识别能力,且发现了中国股票市场信息效应具有聚集性的特点.通过信息状态转移概率矩阵,推测出:在我国股票市场,一个信息经过100分钟能融入市场的概率是99%.

The unobservable state of stock information was modeled based on Hidden Markov Model, and a transition probabilities matrix of information state was built to describe the dynamic association properties in temporal dimension.Based on 5-minutes high frequency trading data and using Bayesian inference and MCMC sampling,the information state and strength of Shanghai Stock Index and sample stocks of SSE 50 in August 2010 were estimated.Empirical results prove the model has effective ability to identify information,and show that the information effects have characteristics of aggregation in Chinese stock market.On the base of the estimated transition probabilities matrix of information state,it is surmised that the probability that an item of information was absorbed by the market after 100 minutes is 99%in Chinese stock market.