Research On Arbitrage Of Stock Index Futures Based On Order Flow Imbalance

High-frequency trading is a hot topic in the financial industry in recent years.It is created by mutual funds,hedge funds and investment banks.It usually uses auto-matic trading strategies.High-frequency trading strategies have the characteristics of short holding time and high cancellation rate.Strategy implementation is highly auto-mated,so it is not affected by traders' feelings;the immediacy of information is very demanding in the implementation process.Successful high-frequency trading strate-gies can bring higher returns,so more and more institutional investors begin to study it.However,in academia,there is still a lack of relevant research.There have been many researches on the impact of long-term market behavior on prices,but there are few related researches on the impact of short-term market behav-ior on the value of financial instruments.In the electronic market,the buying and selling intention of each market participant can be fully reflected in the order book.Cont,Kukanov and Stoikov[1]pointed out that in a short period of time,price changes are mainly driven by order flow imbalance(OFI).The study found that the linear rela-tionship between price changes and the order flow imbalance in the same period are very strong.This article first examines the correlation between the imbalance of order flow under different time scales and price changes over the same period,and shows the effect of linear fitting.Secondly,we extend the study of the impact of order flow imbalance on price to the case of investment portfolios,and test the effectiveness of this indicator from a single financial instrument to multiple financial instruments.We first construct a virtual combined market order book based on the order book data of the relevant underlying asset,and expand the order flow imbalance to multiple assets according to the definition of order flow imbalance.First,the robustness of OFI's in-terpretive performance for price changes over the same period is tested in the case of investment portfolios,and it is found that the relevant properties are still maintained.In addition,we also studied the characteristics of this correlation in different periods of the day based on the slice data of the trading time period.Based on the statistical characteristics and intraday characteristics of the data,mul-tiple market characteristics including OFI are used as input,and the problem of future price changes is transformed into a label prediction problem.Using the lightGBM al-gorithm to build a model with the tth trading day data as the training data,and apply i to the t+1th day,build an intraday high-frequency trading strategy based on the pre-diction tag,and test the profitability of the implementation on the CSI 500 stock index futures.In the case of a single asset,the strategy achieved statistically significant gains in the 2020 backtest.135 of the 241 trading days in the whole year achieved positive returns,with an average daily return of 28,706.58.In the strategy with retracement restrictions,168 trading days achieved positive returns,with an average daily return of 33048.32.Although the model also has relatively high predictive ability in the case of investment portfolio,the backtest performance is not satisfactory due to the high transaction cost.With the electronicization of China's financial market in 2004,automated trans-actions have become possible in China's financial market.We completed the above-mentioned related work using the level 1 data of stock index futures for the whole year of 2020.