An Analysis Of High-frequency Futures Data In Framework Of DBN With Technique Analysis And Logit/probit Model

The observation of DBN in this paper is the results of smoothing price-volumeintra-day data by kernel regression and technique analysis, and the DBN systemconducts learning and inference through these observations. The DBN we considerin this paper is based on HHMM, and HHMM is transformed into DBN throughconditional probability principle and graph theory.In this paper, we add the Logit model and Probit model to DBN system in order tomodeling initial probability of market. With the help of two models, we can judgethe first state of market in every trading day.The learning of DBN is finished by a6-day history data window, then the DBNsystem conducts inference on the7-th day, finally we buy or sell Copper positionunder the direction of DBN inference results.In this paper, we consider two market states, bull state and bear state. The first oneis the situation where the price rising accompanies with the increase of tradingvolume. The last one is the situation in which price falls along with the decrease oftrading volume.The futures contract in this paper is copper. The data we considered is from SHFE,and sampling period is half second. It is found that the category based on DBNsystem really changes the distribution of yield from B&H category. Moreover, theyield in bull state is excelled the yield in bear state. In a word, DBN acts as apowerful tool in analysis of financial market, and it provides hope when handlingmore complex problem.