Stock Price Prediction And Trading Strategy Based On Asymmetric CEEMDAN-BiLSTM-LightGBM Model

The healthy development of the stock market is one of the important basic conditions for the stable progress of the country’s economy.With the continuous improvement of China’s comprehensive national strength and international status and the continuous development of the living standards of our people,a large amount of domestic and foreign capital has poured into the Chinese stock market,financial market participants have increased,and financial products have become increasingly diverse.This has led to a rapid accumulation of financial information in the Chinese stock market in recent years.How to fully analyze and use information has become an important issue before us.In recent years,the rapid development of information technology has made it possible for us to process complicated and huge amounts of information.The algorithm-based stock price prediction and analysis have become more accurate,and its trading strategies have become more flexible and rich.Using advanced computer technology to analyze and forecast China’s stock prices and construct a more practical algorithmic trading strategy can not only provide investors with a more flexible and accurate investment method,but also provide market participants with another perspective to analyze stock price movements.In this thesis,CEEMDAN,BiLSTM and LightGBM algorithms are combined together,at the same time,taking the loss aversion into consideration,construct an asymmetric CEEMDAN-BiLSTM-LightGBM predictive analysis model with higher accuracy and stronger interpretation ability.Using the extreme value points of the predicted value formed by the asymmetric CEEMDAN-BiLSTM-LightGBM model in the process of price prediction with a certain error from the real value to form the buying and selling threshold of the trading strategyThis thesis uses three representative stocks from different industries,PINGAN,i FLYTEK,and Air China,to conduct empirical tests on stock forecasting and trading strategy construction.The prediction model shows great accuracy and applicability,and the strategy effect is significant.In the process of predictive model construction,this thesis finds that the CEEMDAN method has a better effect on stock price decomposition than the EMD and EEMD methods,the addition of adaptive white noise can effectively reduce the modal aliasing that may occur during the stock price decomposition process,which phenomenon shows that China’s stock market noise has a non-Gaussian characteristic;Compared with the single step size,the step size set based on the decomposition information of the high and low frequency shows a higher prediction accuracy in the stock price prediction effect based on the CEEMDAN-BiLSTM-LightGBM model,and the ability of some models to predict the direction has also been improved,which proves that each IMF component formed by CEEMDAN decomposition of its original stock price has certain practicality in the model construction process;The introduction of loss aversion makes the asymmetric model perform better than the symmetric model with the same conditions,which proves that the stock price exhibits the characteristic of loss aversion;In the process of the trading strategy construction,the two trading strategies based on the CEEMDAN-BiLSTM-LightGBM algorithm are far higher than the RSI trading strategy in terms of yield and Sharpe Ratio;Comparing the two trading strategies based on the CEEMDAN-BiLSTM-LightGBM algorithm,under the same conditions in other aspects,compared to its symmetric model strategy,the asymmetric trading strategy has a higher Sharpe Ratio and lower risk volatility,which can better reduce risk on the basis of effectively improving the strategic return..