Ultra Short Term Price Trend Prediction Based On Multistage Order Flow Imbalance

Financial market microstructure theory is under the microscopic structure of the established,to explore what factors had an impact on prices.Traditional studies are mostly by constructing the relevant quantity and price indicators to mining the information in the market data.And the order book as an important tool of delivering market information,is more and more attention by scholars.The order flow imbalance and other indicators constructed by the order book provide a good tool for quantifying the information contained in the order book,which is conducive to better explaining the microstructure theory.Because in the market,institutional and individual investors in the form of a limit order to trade,so there are more trading information are hidden in the order book.Compared to the order book five file stock market and futures market in our country,exchange the published data is only a quotation.And more depth of order book packing data as the Level-2 products for sale,which leads to information asymmetry,so multilevel order book has more information awaiting discovery.It is the core of this paper to analyze and model from the point of view of multilevel order book.Based on the historical Level-2 data of the listed varieties in Dalian Commod-ity Exchange,this paper selects the main contracts of iron ore,coking coal and coke which are actively traded in the black series as the research object.This article first to the original order flow imbalance,e.tc.A simple review of theory,constructs the multi-level order book imbalance index,and carries on the correla-tion analysis,and according to the Cont,Kukanov[1]the methods validation R2 change,then based on multistage order book imbalance index and the relevant relationship between price changes,build characteristic index to predict the price trend change.Next,the indexes based on the research findings were input into the long-term and short-term memory cyclic neural network,and a classifier model was constructed to link the price trend with the index changes.After training,a classifier model that could be used for trading was obtained.Finally,according to the constructed classifier model,a trading strategy is developed,and a better back-test result is obtained.This proves that the multi-level order flow imbalance index can quantify market information well.In the empirical process,it is found that the order book data under high frequency structure has obvious intraday effect.In addition,a new method of labeling classifier is introduced in this paper,and it is found that this method can effectively solve the problem of data imbalance in high-frequency data.