Time-Selection Analysis And Application Research In Quantitative Trading

Securities market transactions are characterized by high risks and high returns,attracting a large number of investors.Traditional investment is mainly based on human will and is constrained by the information processing capabilities of the human brain,as well as human emotions,knowledge,cognition and other factors.In the investment process,it is often blind or excessive.Quantitative investment trading strategies are based on data,combined with multi-disciplinary scientific knowledge such as mathematics and probability theory,with the aid of proven financial knowledge,to formulate efficient investment strategies,and make use of the characteristics of computers such as high efficiency and high discipline to invest.It is a good solution to various deficiencies caused by human nature.Traditional quantitative investment trading strategies combined with securities data for analysis,this solution tends to use computers for technical analysis.The securities market is characterized by significant turbulence.Traditional algorithms are often unable to respond well to changes in price trends,and their generalization ability is weak.Therefore,this article uses artificial intelligence to study quantitative investment transactions.First of all,in the research process,the CNN model based on graph recognition is used to train the stock market data and predict the future trend.Using the prediction results given by the training model combined with the logistic regression strategy model,backtesting was performed on some data from 2016 to 2019.The backtest results show that the quantitative trading strategy model based on graphic recognition can predict the trend of the future period of time to a certain extent,and the timing results can also bring above-average returns with a high probability.The accuracy rate of the CNN graph recognition model's over-the-top trend is about 75%.Combined with the logistic regression model to predict trading points,the accuracy rate is about 80%.At the same time,because the simple pattern recognition algorithm cannot accurately determine the buying point and selling point,and the average average difference days is high,an Adaboost algorithm model based on the pattern recognition result is proposed to find more accurate buying points and buying points,and give a specific quantification Trading straregy.In the process of processing,due to the large gap between the prices of different stocks,in order to ensure the uniformity of the data,the z-score algorithm is used to normalize the data.After optimization using the Adaboost algorithm model,the average mean difference days decreased to about 5 days.Therefore,it shows that the multi-factor Adaboost stock trading point timing algorithm based on deep learning of Kline graphics has certain practical value.