Research On Quantitative Trading Strategies Based On Neural Network Models

With the rapid development of digital technology,the idea of combining big data distributed systems with intelligent algorithms has been applied in more and more business scenarios in the past 10 years.Quantitative trading is a typical application scenario for automated trading using big data and intelligent algorithms in financial investment.More and more financial institutions and even individual investors widely use quantitative trading for investment operations.Deep learning has made a remarkable breakthrough in recent years,surpassing the traditional machine learning approach.Its application has been extensively tried and proven successful.The industry has also tried to solve the problem of financial market transactions by applying deep learning.This thesis aims to explore models based on neural networks and deep learning to predict the future trend of treasury bond bond futures prices in China’s securities market,and use these models to build intelligent quantitative trading strategies that can obtain excess returns in order to achieve better investment results.Firstly,This thesis initially puts forward a neural network-based model and deep learning to predict the future trend of 10-year treasury bond bond futures yield.By combining historical price data and technical indicators,we employed a dichotomous and linear regression model to forecast the future price of ten-year treasury bond futures.This thesis will investigate the features and applications of these models,constructing a systematic quantitative trading strategy model and simulating trading with real-world data.The actual trading outcomes demonstrate that the utilization of these models is beneficial.This combination trading strategy can achieve excess returns.Secondly,This article adopts a popular method in the industry to construct quantitative trading strategies through deep reinforcement learning.In order to optimize the enhanced learning environment for the trading time point of 10-year treasury bond futures with a single subject,we have made some improvements on the basis of the existing methods,which can effectively improve the trading efficiency and accuracy.By using the Deep Q-learning algorithm and deep reinforcement learning methods,we can maximize returns and construct an effective trading strategy using recurrent neural networks and deep reinforcement learning methods.After practical verification,the return and risk management performance of this trading strategy is significantly superior to the strategy of randomly buying and holding based on subjective experience,and has significant advantages.Finally,for risk management,this thesis designs and implements a trading strategy verification and evaluation system.This system can simulate real transactions based on historical data,view and generate simulated trading results in real time,and analyze the risk and profitability of this strategy.Implement automated one-click deployment after verification and evaluation.