Research On Dynamic VWAP Algorithm Trading Strategy Based On Machine Learning

In recent years,algorithmic trading has developed rapidly and been widely used in the financial field,and the research on algorithmic trading has become increasingly active.As the most widely used Volume Weighted Average Price(VWAP)algorithmic trading strategy has become one of the research hotspots.Based on the trading strategy of VWAP algorithm,this thesis divides the large order allocation problem into two core problems:The solutions to the two problems are improved and studied respectively,and a systematic solution to the large order distribution problem is proposed,which can assist institutional investment managers to formulate a more reasonable large order distribution scheme and reduce the investment cost.Firstly,based on the stock trading data,this thesis studies the correction of stock trading price under the limit of rise and fall.Based on the corrected stock trading price,it studies the cyclical characteristics of stock price and the fluctuation characteristics of stock price,providing the basis for investors to make a judgment on the market.Secondly,based on the periodicity and fluctuation characteristics of stock price,the thesis studies the forecasting problem of stock trading volume and stock price to help investors evaluate the scale and trend of stock market reasonably.Finally,based on the predicted stock trading volume and stock trading price,the optimization problem of large order allocation is studied to optimize the allocation of investors’ resources,reduce transaction costs and improve returns.The main research work of this paper is as follows:1.Trading price correction problems based on stock trading data.First,this thesis presents an unconstrained trading price correction method based on stock market trading price sequence(Sequential EM Algorithm),which can correct the trading price constrained by the limit of rise and fall in historical trading data.Then,empirical mode decomposition(EMD)algorithm is used to remove the interference in the modified trading price sequence and obtain the local periodic sub-signals(IMF).Finally,a sequence continuation algorithm is designed to fit the decomposed eigenmode functions into sinusoidal or linear functions and extract the periodic characteristics of stock prices after correction.2.Stock trading volume prediction problem.Based on the cyclical characteristics and volatility characteristics of stock prices,a stock trading volume prediction model is designed,which is called DynPre.On the basis of the original neural network model,the algorithm adds the sample building algorithm and sample extraction algorithm based on the variable length sliding time window,which can not only realize the dynamic prediction of stock trading volume,but also has the ability of minute-level iteration.This thesis uses Jiangshan Stock(stock code:600389)from 2021-10-08 to 2022-01-14,70 day of the minute level section of transaction data.The prediction effect of DynPre and traditional trading volume prediction model is compared from two aspects of prediction accuracy and operation efficiency.The experimental results show that DynPre is 47.5%more accurate than the traditional model,and the slowest running time can be finished in 283 milliseconds,which can meet the time requirement of large order splitting.3.Optimization of large order allocation.Firstly,DynPre is used to predict the stock price.Secondly,VWAP algorithmic trading strategy with stock price information is designed based on the stock price prediction data,which is called PVWAP algorithmic trading strategy.Finally,this thesis uses the minute-level slice transaction data of Jiangshan Stock from 2022-01-14 to 2022-01-18 to conduct simulation verification.Comparing the split effect of traditional VWAP algorithmic trading strategy and PVWAP strategy from three dimensions of tracking market VWAP effect,execution success rate after allocating orders using algorithmic trading strategy and whether to beat market VWAP.In the simulation experiment,the execution VWAP corresponding to the PVWAP algorithm trading strategy beats the market VWAP with an average of 0.3 yuan per transaction,and the execution VWAP corresponding to the VWAP algorithm strategy beats the market VWAP with an average of 0.64 yuan per transaction,which proves the feasibility and effectiveness of the PVWAP algorithm trading strategy.