Forecasting Yield Of CSI 300 Index Based On LSTM Models

In recent years,due to the popularity of high-performance computers and the rapid development of artificial intelligence technology,artificial intelligence has challenged the ecology of various fields.The financial field is no exception.Although China’s securities market has developed rapidly and achieved remarkable achievements,the problems of insider trading and incomplete delisting system reflect that there are still many problems of mechanism and supervision in China’s securities market.In addition,the financial market is a complex dynamic system full of noise.How to accurately predict the trend of the broader market index and analyze the impact of macro factors on the securities market at this stage is a hot issue that scholars pay attention to.For investors,predicting the trend of stock price and volatility and achieving the result of high yield and low risk is their eternal pursuit.All kinds of demands naturally determine that the financial field is a fairly open field and its practice method is updated more quickly.Because of this feature plus the maturity of artificial intelligence technology,the rapid integration of artificial intelligence and finance field is an inevitable result.For example,quantitative investment is rising in recent years.After reading the strategic research reports of major institutes,it is not difficult to find that machine learning models and deep learning models are very popular.In deep learning,it is the gated recurrent neural network,such as long-short-term memory models,that is good at dealing with time series problems and long-term dependency problems.The theoretical part of this paper introduces the basic knowledge of the ARMA-GARCH family models and neural network models.In the neural network part,the back propagation algorithm and Adam optimization algorithm are introduced and deduced in detail.In the empirical part,Eviews,R language,Keras library and other tools are used to establish the ARMA-GARCH family models and LSTM models.The object of preliminary modeling is the logarithmic return series of CSI 300 Index from January 1,2008 to September 30,2019.In this paper,the model is constantly adjusted and a variety of model architectures or superparameters are tried in the process of modeling.You will see a more detailed model selection and discussion process in this paper.Finally,the figures and tables are used to show the quality of the prediction results between different models.The empirical results at the end of this paper show that after using a more reasonable sample set,the prediction accuracy of the LSTM models are higher.After learning more information by using more input features,the predictive ability of the LSTM model has been further improved.These results reflect that the LSTM models have excellent forecasting potential and flexibility.They have a broad prospect in the field of financial prediction.