Research On Portfolio Based On AP Cluster Selection And LSTM Asset Forecasting Configuration

Portfolios can effectively reduce investment risk by diversifying funds held by investors or financial institutions.The data studied by the portfolio is typical time series data.If the timing correlation between the selected portfolios is large,the inve stment risk cannot be truly dispersed,and the asset allocation is mainly performed by analyzing the historical time series data,which is not effective.Address the risks of future financial markets.Therefore,it is important to study time series clustering to obtain low-risk portfolios and to conduct portfolio strategy research under effective time series forecast data.Aiming at the problem of large time series correlation between portfolios,according to the financial time series clustering feasibility analysis,a time series clustering method based on AP clustering is proposed for portfolio selection.The advantages of applying AP clustering to portfolio clustering selection are analyzed.At the same time,the problem of large number of clusters and uncontrollable due to excessive data volume and high time dimension in the original AP clustering algorithm is improved.A controllable AP clustering algorithm,which sets controllable parameters in the original AP clustering algorithm,thereby controlling the clustering results and providing investors with a flexible portfolio selection method.Aiming at the problem that portfolio asset allocation based on historical data cannot effectively deal with the future risk of financial market,it is puts forward a method of portfolio asset allocation based on LSTM to forecast the data of selected portfolio,calculates the correlative risk of each time series data according to the predicted data,and proposes a dynamic timing method combining with the correlative risk,which can provide investors with favorable investment time and asset allocation ratio to cope with future financial market risks.The historical data of 300 stocks in Shanghai and Shenzhen from January 2018 to March 1,2019 are selected for experimental verification and analysis.The results show that the controllable AP clustering algorithm can effectively deal with the uncontrollable number of clusters caused by the original AP clustering,and the Silhouette Coefficient and CH evaluation are superior to the classical clustering method in dealing with portfolio selection under the same number of clusters.Establishing appropriate LSTM models to predict the stocks in the portfolio selected by clustering,comparing the gap of income between forecast data and actual data in asset allocation based on relevant risk,which in that the maximum error of single stock income is 4.7293%,the error is no more than 5%,the error of portfolio return is 0.075%,no more than 1%.The dynamic timing asset allocation method based on correlation risk proposed under the forecast data can improve the growth returns by 3.55% under the data and conditions of this study,and obtain the optimal investment timing and asset allocation ratio,which is conducive to obtaining the ideal returns for investors to cope with future risks.