Research And Implementation Of Enhanced Index Tracking Algorithm Based On Clustering And LSTM

In recent years,with the rapid development of our country's economy,the people's living standards have been improved,and their disposable income has also greatly increased.Therefore,there has been an increasing demand for investment.However,the current investment methods suitable for ordinary people are very limited.Although the stock market has higher returns than other investment approaches,it also has higher risks.With the development of related technology in the field of computer,it has become a general trend to apply the technology such as computer and artificial intelligence to the investment field to meet the investment needs of ordinary people.Enhancing index tracking is one of the hot issues,the main purpose of which is to add active management to the passive investment strategy to adjust the portfolio,so as to obtain a higher excess return than the index while the tracking error is smaller.Experts on computer-related fields and financial fields have conducted certain explorations on this problem.The existing research can be roughly divided into three categories: model-based,heuristic algorithm-based,and learning algorithm-based.Through the relevant literature and comparison of experiments,it is found that these three methods have their own advantages and disadvantages.To solve the above problems,this paper proposes an enhanced index investment algorithm based on clustering and LSTM network.Firstly,the enhanced index tracking problem is described in mathematical language according to its economic definition,and a constrained multi-objective optimization model is established;then the solving process of the problem is divided into two steps.The stock selection algorithm is based on principal component analysis and clustering.And the investment weight calculation algorithm based on timewindow and LSTM network is proposed.And the algorithm proposed in this paper is compared with the existing representative algorithms,the conclusion is drawn.The research content of this paper mainly includes the following aspects:(1)Give the definition of enhance index tracking issues and the determination of evaluation indicators.After reference a large number of related literature,the enhanced index tracking problem is defined as a constrained multi-objective optimization problem.The two goals are to maximize excess return and minimize tracking error.The constraints are some trading rules of the stock market.According to the related literature,the Sharpe ratio and other indicators are determined as the final evaluation indicators of the model.(2)Principal component analysis and clustering methods are used to select stocks.Firstly,principal component analysis is used to reduce the dimension of the data.Then the advantages and disadvantages of various clustering methods are analyzed,and the appropriate clustering methods are selected based on the data characteristics.After the dimension reduction,the data are clustered using K-means algorithm,and the stock selection is carried out according to the clustering results.(3)Time-window and LSTM network are used to calculate the investment weight.Firstly,the time-window is used to split the data to obtain the training set.Cosidering the advantages of the LSTM network which is good at dealing with time series data,the tracking error and excess return are introduced into the loss function to train the LSTM network.Finally the trained LSTM network is used to calculate the stock investment weight in the portfolio.In this paper,the enhanced index tracking investment algorithm based on clustering and LSTM network is used to verify the historical data in different markets at home and abroad.For different investment weight calculation methods,the stock selection algorithm based on principal component analysis and K-means clustering is compared with other algorithms and the algorithm improves the Sharpe ratio by an average of 35.7%.For different data sets,the algorithm proposed in this paper increases the Sharpe ratio by 80.7% on average compared with other algorithms,which verifies the effectiveness of the algorithm proposed in this paper.