Quantitative Trading Strategy Improvement Based On LSTM Deep Neural Network

The expansion and institutional improvement of global capital markets in the long-cycle economy provide the underlying driving force.Based on this,with the support of the breakthrough of computer technology and the improvement of relevant theories,the attention of quantitative investment as a new investment model is gradually increasing,and the scale of quantitative related investment is also expanding rapidly.According to relevant statistics,from 2018 to the end of 2021,the scale of domestic quantitative private equity funds increased from 7% to more than 25% of the scale of private equity securities.In addition,the quantitative product categories were iterated rapidly,and the level of profitability was continuously improved.However,as the impact of the epidemic,the global liquidity crunch and geopolitical factors have a huge impact on the macro economy,the scale and income level of quantitative investment have also been affected.Therefore,the research on the potential deficiencies of quantitative investment and iterative innovation has important influence and significance for the development of quantitative investment in the future.Relative gains Machine learning techniques have been extensively used for data mining and feature extraction and prediction.On this basis,people generally use long and short term memory network(LSTM)to extract features from time series data.This paper first introduces the basic structure and working principle of LSTM,and then discusses how to use LSTM for sequence data modeling and feature extraction.This paper explores the feature extraction method of the original data using LSTM algorithm,and combines the random forest model for prediction,and finally realizes the goal of stock position construction and trading strategy construction.This study provides certain reference value for the quantitative strategy development based on LSTM feature extraction algorithm in the future.This paper attempts to focus on A total of 75 fundamental factors,technical factors and investor sentiment factors of China A shares from 2016 to 2022.After outlier screening,data standardization and neutral processing,the validity of the factors is tested,and the identified effective factors are substituted into the LSTM model for feature extraction.The extracted features were substituted into the random forest model for rolling training,and the investment strategy was established according to the predicted results.The results show that the strategy income of quantitative trading model based on LSTM neural network is obviously better than that of CSI 300 benchmark and medium index enhanced quantitative fund in the market.Compared with the original trading strategy,the improved LSTM-RF strategy has been improved in terms of income ability and risk control.Therefore,the improvement scheme of quantitative trading strategy based on neural network can improve the effectiveness of trading strategy,which has good theoretical value and practical significance.