Research On Portfolio Strategy Optimization Based On Data Denoising And Stock Price Prediction

The uncertainty and volatility of product value in financial markets not only bring excess returns but also risk losses.Due to the nature of seeking profit and avoiding harm,investors are not only afraid of potential risks,but also eager to obtain high returns.However,the close relationship between returns and risks determines that investors are inevitably exposed to risks when pursuing investment returns in the financial market.In order to seek a balance between risk and return,portfolio theory has emerged.In 1952,Markowitz put forward the mean variance model,which pioneered the modern portfolio theory.On this basis,scholars at home and abroad have made many efforts to constantly improve the portfolio theory,and have also made a series of research results.However,most studies overlook quantitative research on the impact of noise,which greatly limits the improvement of portfolio performance.In the process of combinatorial optimization,most of the existing studies tend to take the historical rate of return as the model input,and the investment strategy obtained may have a certain lag.How to quantify the impact of noise and establish a scientific and efficient online portfolio model based on it,providing reference for different types of investors,has become a hot topic in the field of financial research.In order to improve the current research status,this article follows the research approach of "quantifying the impact of noise→denoising the original sequence→predicting stock price portfolios→constructing a mean risk investment portfolio model",and applies literature research,empirical research,and comparative analysis methods to explore possible ways to improve the performance of financial market participants’investment portfolios from each stage of the investment portfolio,Provide reference for different types of stock market participants to allocate funds and make rational investments based on their actual situation.This article consists of six chapters:Chapter 1 is the introduction section,which first introduces and analyzes the background and significance of the topic,and elaborates on the research content,structural arrangement,and research methods of this article based on a review and sorting of research results in relevant fields.Finally,the main innovation points of this article are summarized and summarized;Chapter 2 is the theoretical overview section,which provides a detailed introduction to the models or theories involved in each subsequent chapter,providing theoretical support for the establishment of investment portfolio strategies;Chapter 3 quantifies the impact of noise on different investment portfolio models from both theoretical and empirical perspectives,and compares the degree of impact between the minimum risk model and the mean risk model;The fourth chapter establishes a stock price combination forecasting system based on model selection,original sequence noise reduction,machine learning technology and multi-objective optimization algorithm,and tests the effectiveness of the established forecasting system from three aspects of forecasting error,direction fitting degree and Goodness of fit;Based on the predicted stock prices in Chapter 4,Chapter 5 calculates the future returns of each investment target and uses this as input to the mean CVAR model to obtain asset allocation strategies for investors with different market states and risk preferences,providing reference for financial market participants;Chapter 6 summarizes and summarizes the research content and conclusions of this article,and provides prospects for future research work.This article focuses on constructing an online portfolio model based on data denoising and stock price prediction.The specific implementation path and conclusions obtained can be summarized in the following aspects:Firstly,considering the lack of theoretical basis for the application of denoising in portfolio research,this paper first decomposes noisy prices into non noisy components and noise,and theoretically derives the optimal combination weights of the minimum variance,mean variance,minimum CVaR,and mean CVaR models in both noisy and non noisy environments.By comparing their differences,it was established that noise would cause the optimal combination to deviate from its true position,elucidating the necessity of denoising.Secondly,based on theoretical analysis,this article further discusses the impact of noise on investment portfolios through empirical research,and compares and analyzes the differences in the impact of noise on different portfolio models.The empirical results indicate that the minimum risk model is less susceptible to noise interference compared to the mean risk model,the use of variance to measure risk compared to CVaR,and parameter estimation methods compared to semi parametric and non parametric methods,resulting in more robust portfolio performance.On the other hand,the mean risk model,the use of CVaR to measure risk,and semi parametric and non parametric estimation methods can yield more portfolio returns,resulting in stronger risk return equilibrium ability.In practice,it is more important to carry out denoising research on portfolio models that are highly affected by noise,such as the mean CVaR model.It can not only eliminate noise interference,improve the robustness of the model,but also harvest more portfolio returns.Secondly,in response to the issue of using historical returns as model input and less considering lag in previous portfolio models,this article constructs a stock price prediction system that integrates data denoising,prediction,and optimization,laying the foundation for the establishment of subsequent portfolio models.The design process of this prediction system can be divided into the following steps:Firstly,this article uses the most widely used data denoising techniques(EMD model,EEMD model,CEEMD model,VMD model,wavelet denoising model)to denoise the collected raw dataset.Using the MAPE values obtained from the same prediction model as the model selection criteria,the most suitable denoising methods for financial data are selected;Secondly,ARIMA,LSTM,ELMNN,and SVM models were used to predict the denoised stock prices.ARIMA models were used to capture the linear trend of stock prices,while LSTM,ELMNN,and SVM models fitted the nonlinear trend of stock price fluctuations from the perspectives of neural networks and deep learning;Once again,the prediction results obtained from these models are combined,and the multi-objective bacterial population chemotaxis algorithm optimizes the combination weights to achieve the best prediction effect of the system.The final empirical comparison results show that the VMD model always shows the best denoising effect in both high-frequency data sets and low-frequency data sets,and compared with other comparison models,the prediction system established in this paper,VMD-BCC-ALSE,is considered to be the optimal prediction model because of its higher prediction accuracy,higher direction fitting degree and Goodness of fit.Thirdly,in response to the many shortcomings of the mean variance model,this article selects a more scientific conditional value at risk as the risk measurement standard and constructs a mean CVaR investment portfolio model to guide investors in rational asset allocation.In addition,this paper also simulates the investment behavior of risk aversion investors,risk neutral investors and risk preference investors by changing the confidence level of conditional value at risk model.Based on the overall return level of the stock market,this paper selects different window periods to compare and analyze the portfolio strategies in bear market period,bull market period and stable volatility period,Analyze the differences and similarities in the optimal investment strategies of different types of investors in different market states,providing comprehensive reference for investors’ decision-making.Based on the experimental results,the following conclusions can be drawn:①Even portfolio models established based on historical data can achieve good portfolio performance,but portfolio models determined based on stock price prediction have better portfolio performance.This conclusion holds regardless of the state of the stock market;②Among the samples selected in this chapter,CVaR has always maintained the most ideal effect in all risk measurement standards,and the mean CVaR portfolio model built on this basis is also optimal in asset allocation.This result once again confirms the rationality of the mean CVaR combinatorial optimization framework proposed in this paper based on data denoising and stock price prediction.③The state of the stock market,investors’ risk tolerance,and expected return rate all have an impact on the optimal investment weight.Among them,the state of the stock market has the greatest impact on investors’ asset allocation strategies.Overall,this study has certain enlightening value,mainly reflected in:①In terms of data denoising,this paper theoretically infers that the optimal portfolio weight obtained in noisy environments is biased,which provides a reasonable explanation for previous studies that directly use denoising technology to process the original series,enriches the relevant research on portfolio to a certain extent,helps to eliminate the bottleneck of portfolio theory,and has certain theoretical significance for promoting the development of modern portfolio theory.②In terms of investment portfolio model input,based on the established stock price comprehensive prediction system,a deep exploration of the micro market situation of the stock market from a future perspective has made up for the lag in screening and combining results between different assets using historical trading data,which can help stock market participants achieve a balance between returns and risks to a greater extent and provide practical decision-making references.③In terms of model application,facing the still growing Chinese stock market,this article uses the mean risk model to help market participants achieve a balance between returns and risks.In the risk measurement section,a conditional in value risk model that better fits investor psychology is adopted,and the combination strategies of different risk preference types of investors are explored by setting different confidence levels.The final empirical results can provide reference for investors with different risk tolerance abilities,enabling them to allocate funds and make rational investments based on their own risk preferences.④In terms of overall structure,the setting of each chapter is progressive layer by layer.Chapter 3 uses the relevant theoretical knowledge introduced in Chapter 2 to derive the differences in the performance of investment portfolio models before and after denoising,demonstrating the urgency and necessity of denoising.The first step in establishing a stock price prediction system in Chapter 4 is to select the most suitable denoising technology for financial sequences from many popular denoising methods,and use each sub model to predict the denoised sequence,According to the prediction system in Chapter 4,predict the future stock price of the selected investment target,and input the processed results into the mean risk model in Chapter 5 to determine the optimal investment portfolio strategy.During the research process of this article,there are still some unresolved issues:firstly,this article only uses the Shanghai and Shenzhen 300 index dataset to compare and analyze the predictive performance of different prediction models.Increasing the types and quantity of datasets appropriately can better illustrate the superiority of the proposed combined prediction system,and the conclusions obtained will also be more convincing.Secondly,this article empirically demonstrates that investors’ risk tolerance and the market state of stocks can have an impact on investment portfolio strategies.However,the specific impact pathways have not been explained,and the mechanisms still need further research.Thirdly,for the sake of model simplification,this article only establishes and solves the mean CVaR model from a theoretical perspective,neglecting some practical factors involved in the financial market practice process,which is worth further research in the future.