Research On Stock Trading Strategy Based On Improved Transformer Network

The prosperity index of the stock market can not only reflect the prosperity of the national economy,but also stock investment is one of the main means for enterprises and institutions to seek financing.A scientific and automated stock trading strategy can not only help individuals and businesses make rational investment decisions,but also avoid market fluctuations caused by group emotional buying and selling.Therefore,the research on stock investment prediction has always been a hot topic in the financial field,with crucial theoretical significance and practical value.However,the stock market is a complex and ever-changing dynamic system.Due to the complexity of the internal structure of the stock price system and the diversity of external factors,the changes in stock prices exhibit strong nonlinearity.Accurate prediction of stock prices poses great challenges.Although research on stock price prediction has been ongoing,there are few successful examples of using stock price prediction models to develop effective trading systems.Inspired by the commonly used "moving average theory" by stock market operators to search for trading opportunities,this article proposes a definition of important trading points for stocks based on the moving average judgment strategy.This definition innovatively expands the single point prompt for stock buying and selling to the warning range prompt,alleviating the profit reduction problem caused by high-frequency trading.In addition,the paper applies the improved Transformer model to predict the important trading points of stocks for the first time.In the improved Transformer model,this paper adds a one-dimensional convolution layer(Conv1d)and a short-term memory layer(LSTM)to learn the local volatility characteristics and long-term time dependence of stock price data,so as to explore the important trading points(ETPs)that are more likely to predict the sharp rise of stock prices and capture high profit opportunities.The improved Transformer model categorizes the daily closing price series of stocks into two categories,assigning "1" to the single points that belong to the important trading point range,and "0" to the other single points.In order to evaluate the real profitability of the stock trading system designed in this paper,the cumulative rate of return and Sharpe ratio are specified as comprehensive trading evaluation indicators.And design multiple sets of comparative experimental strategies,such as the classic random buying and selling strategy(Random),buy and hold(B&H)strategy,and double moving average trading strategy(DEMA)in stock trading,to conduct backtesting and evaluation on historical data.The experimental results show that the application of the stock important trading point(ETP)strategy proposed in this article to the improved Transformer model significantly improves profitability,with its actual annual cumulative profitability being 39.02% higher than that of the random buying and selling market.