High Frequency Quantitative Trading Strategy Base On Deep Learning And LM Jump Detection Method

Quantitative investment,as a new method in nearly thirty years,gradually become a mainstream investment approach in today's world.At the same time,with the continuous development of Internet technology and computer science,the growth of the amount of information in financial markets has been far beyond the scope of the human brain can handle.Compared to the traditional investment methods,advantages of quantitative investment are gradually revealed.The innovation in this research field needs intersection and integration of different disciplines,so more and more researchers have been trying to construct the quantitative investment strategies using computer intelligence and other related technology combined with financial theory.At present,along with the asset price jump detection technology matures,the typical phenomenon in market microstructure have been excavated and confirm,constructing investment strategy base on the typical phenomenon has a brilliant future.Under such a background,we construct a high frequency trading strategy using the price jump detection method combined with deep learning technique.According to the back test result our strategy no matter from the returns or risk control performance are excellent thus it is very important to the theoretical researchers and practical operators.Firstly,we summarize the existing jump detection theory and method systematically,specially focuse on Lee and Mykland parameterless jump detection method and give the reason why we choose this method.On this basis,some problems of the above detection methods are given and we propose a new LM jump detection method fused with deep learning technique.Secondly,using the LM jump detection method,we monitor the 1 minute price jump of the csi 300 index futures through LM parameterless jump detection,dig out the typical event in the market: target price reflect the reality event in the form of cluster jump,then through statistical method,verify the price jumps as the profitable trading signals.After analysing the results of the initial trading strategy,we find the differences between the jumps in clusters which lead our initial trading strategy to the bad performance.Thirdly,for the price jump effectiveness classification problem,after studying the variables that affect the price jump and characteristics of time sequence data,we decide to use long short term memory network,a technique in deep learning method,as a module to improve our strategy.Finally,Through the reasonable setting of network structure and adjusting the training optimization measures,we set the level2 data(five price quotations and volume)provided by CICC as our input variables.Then,we extract the high dimensional features from the price jumps and define the effectiveness of the price jumps according to the features in order to predict price trend in the future.The results show that the optimized trading strategy has higher yield,better stability as well as stronger risk control ability.Furthermore,compared our strategy with the commonly used benchmark trading strategy in the market,we may draw the conclusion: our new quantitative trading strategy can gain the alpha profit stably and beat the market benchmark,at the same time,our strategy is also better than the benchmark performance in terms of risk control.Our innovation points are as follows:(1)using LM parameterless jump detection method to identify profitable trading signal and combining with the long short term memory network,we innovatively propose a high frequency quantitative trading strategy base on deep learning and LM jump detection method.(2)excavate the characteristics of csi 300 index futures price jump with the statistical method and find out the typical event in future market: the underlying assets absorb the information from reality through price jump clustering,which is internal profit logic of the trading signals.(3)combining with deep learning techniques,we intelligently excavate the high dimensional characteristics of stock index future in five price quotations and volume.Then we define the effectiveness of price jumps innovatively and classify the price jump of underlying assets according to our definition.