The Analysis Of Ultra-high-frequency Tick Data In The Stock Market

Entering the era of big data, there has no longer been a problem in getting and storing data.As in stock market, we can get the finest Tick data (i.e. UHF data) which includes almost all ofthe information in the market. The problem we need to solve is how to deal with these data toobtain the information as much as possible.Notes that, the tick data is non-equal time spacing, and the spacing is random, there are twomethods to handle such data: First, filtering the original data to generate new equally spaced data(such as minutes of data), and then use the classic model; The second method is to create a newmodel of ultra-high frequency data for a random time interval. The first approach reduced thedata frequency, corresponding to pay the price of information loss. In order to take full advantageof the information of UHF data, we should create a model on random intervals. However, wenote that the tick data is not completely random, presenting the clustering of the time interval andprice. To solve this problem, we choice the autoregressive conditional duration model (ACDmodel) to describe the structure of the autocorrelation of time interval, choice ECOGARCHmodel to describe the changes of the log price volatility, and then we combine the two modelstogether as ACD-ECOGARCH(1,1) model to describe tick data. In order to estimate the modelparameters, In the hypothesis that we can observe all price jumps, we introduce thequsi-maximum likelihood estimation method (QMLE).Finally, we use the tick data of Moutai (stock code:600519.sh) in shanghai exchange. Weuse the simulate annealing giving the estimation value of the parameters, and give a briefdescription of the leverage effect and the volatility.